OpenBMB/ChatDev
1
1
Fork 0
mirror of https://github.com/OpenBMB/ChatDev.git synced 2024-07-20 08:10:24 +03:00
Code Issues Actions Packages Projects Releases Wiki Activity

MultiAgentEbook

Browse Source
This commit is contained in:
钱忱 2024-06-25 22:25:48 +08:00
parent e49d851e53
commit 7afa42b399
165 changed files with 4818 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
MultiAgentEbook/README.md Normal file
View File

@ -0,0 +1 @@
# Coming Soon

29
MultiAgentEbook/book_communication/data.csv Executable file
View File

@ -0,0 +1,29 @@
,image_path,title,author,summary,affiliation
0,./images/1d.png,AgentCF: Collaborative Learning with Autonomous Language Agents for Recommender Systems,"Junjie Zhang, Yupeng Hou, Ruobing Xie, Wenqi Sun, Julian McAuley, Wayne Xin Zhao, Leyu Lin, Ji-Rong Wen","Recently, there has been an emergence of employing LLM-poweredagents as believable human proxies, based on their remarkabledecision-making capability. However, existing studies mainly focuson simulating human dialogue. Human non-verbal behaviors, suchas item clicking in recommender systems, although implicitly ex-hibiting user preferences and could enhance the modeling of users,have not been deeply explored. The main reasons lie in the gapbetween language modeling and behavior modeling, as well as theincomprehension of LLMs about user-item relations.To address this issue, we propose AgentCF for simulating user-item interactions in recommender systems through agent-basedcollaborative filtering. We creatively consider not only users butalso items as agents, and develop a collaborative learning approachthat optimizes both kinds of agents together. Specifically, at eachtime step, we first prompt the user and item agents to interact au-tonomously. Then, based on the disparities between the agents’decisions and real-world interaction records, user and item agentsare prompted to reflect on and adjust the misleading simulationscollaboratively, thereby modeling their two-sided relations. The op-timized agents can also propagate their preferences to other agentsin subsequent interactions, implicitly capturing the collaborative fil-tering idea. Overall, the optimized agents exhibit diverse interactionbehaviors within our framework, including user-item, user-user,item-item, and collective interactions. The results show that theseagents can demonstrate personalized behaviors akin to those of real-world individuals, sparking the development of next-generationuser behavior simulation.","Renmin University of China, UC San Diego, Tencent"
1,./images/agentcf_collaborative_learning_with_20231013.png,AgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors,"Weize Chen, Yusheng Su, Jingwei Zuo, Cheng Yang, Chenfei Yuan, Chi-Min Chan, Heyang Yu, Yaxi Lu, Yi-Hsin Hung, Chen Qian, Yujia Qin, Xin Cong, Ruobing Xie, Zhiyuan Liu, Maosong Sun, Jie Zhou","Autonomous agents empowered by Large Language Models (LLMs) have under-gone significant improvements, enabling them to generalize across a broad spec-trum of tasks. However, in real-world scenarios, cooperation among individuals isoften required to enhance the efficiency and effectiveness of task accomplishment.Hence, inspired by human group dynamics, we propose a multi-agent frameworkAGENTVERSE that can effectively orchestrate a collaborative group of expert agentsas a greater-than-the-sum-of-its-parts system. Our experiments demonstrate thatAGENTVERSE can proficiently deploy multi-agent groups that outperform a singleagent. Extensive experiments on text understanding, reasoning, coding, tool utiliza-tion, and embodied AI confirm the effectiveness of AGENTVERSE. Moreover, ouranalysis of agent interactions within AGENTVERSE reveals the emergence of spe-cific collaborative behaviors, contributing to heightened group efficiency. Our codehas been released at https://github.com/OpenBMB/AgentVerse/.","Tsinghua University, Beijing University of Posts and Telecommunications, Tencent Inc."
2,./images/agentverse_facilitating_multi-agent_collaboration_20230821.png,Apollo's Oracle: Retrieval-Augmented Reasoning in Multi-Agent Debates,"Haotian Wang, Xiyuan Du, Weijiang Yu, Qianglong Chen, Kun Zhu, Zheng Chu, Lian Yan, Yi Guan","Multi-agent debate systems are designed to derive accurate and consistent conclusions through adversarial interactions among agents. However, these systems often encounter challenges due to cognitive constraints, manifesting as (1) agents' obstinate adherence to incorrect viewpoints and (2) their propensity to abandon correct viewpoints. These issues are primarily responsible for the ineffectiveness of such debates. Addressing the challenge of cognitive constraints, we introduce a novel framework, the Multi-Agent Debate with Retrieval Augmented (MADRA). MADRA incorporates retrieval of prior knowledge into the debate process, effectively breaking cognitive constraints and enhancing the agents' reasoning capabilities. Furthermore, we have developed a self-selection module within this framework, enabling agents to autonomously select pertinent evidence, thereby minimizing the impact of irrelevant or noisy data. We have comprehensively tested and analyzed MADRA across six diverse datasets. The experimental results demonstrate that our approach significantly enhances performance across various tasks, proving the effectiveness of our proposed method.","Harbin Institute of Technology, Sun Yat-sen University, Zhejiang University"
3,./images/apollo's_oracle_retrieval-augmented_reasoning_20231208.png,ATM: Adversarial Tuning Multi-agent System Makes a Robust Retrieval-Augmented Generator,"Junda Zhu, Lingyong Yan, Haibo Shi, Dawei Yin, Lei Sha","Large language models (LLMs) are proven tobenefit a lot from retrieval-augmented genera-tion (RAG) in alleviating hallucinations con-fronted with knowledge-intensive questions.RAG adopts information retrieval techniquesto inject external knowledge from semantic-relevant documents as input contexts. How-ever, due to today’s Internet being flooded withnumerous noisy and fabricating content, it isinevitable that RAG systems are vulnerableto these noises and prone to respond incor-rectly. To this end, we propose to optimizethe retrieval-augmented GENERATOR with aAdversarial Tuning Multi-agent system (ATM).The ATM steers the GENERATOR to have a ro-bust perspective of useful documents for ques-tion answering with the help of an auxiliaryATTACKER agent. The GENERATOR and theATTACKER are tuned adversarially for severaliterations. After rounds of multi-agent itera-tive tuning, the GENERATOR can eventuallybetter discriminate useful documents amongstfabrications. The experimental results verifythe effectiveness of ATM and we also observethat the GENERATOR can achieve better perfor-mance compared to state-of-the-art baselines.","Beihang University, Baidu Inc."
4,./images/atm_adversarial_tuning_multi-agent_20240528.png,Auto Arena of LLMs: Automating LLM Evaluations with Agent Peer-battles and Committee Discussions,"Ruochen Zhao, Wenxuan Zhang, Yew Ken Chia, Deli Zhao, Lidong Bing","As LLMs evolve on a daily basis, there is an urgent need for a trustworthy evaluationmethod that can provide robust evaluation results in a timely fashion. Currently,as static benchmarks are prone to contamination concerns, users tend to trusthuman voting platforms, such as Chatbot Arena. However, human annotationsrequire extensive manual efforts. To provide an automatic, robust, and trustworthyevaluation framework, we innovatively propose the Auto-Arena of LLMs, whichautomates the entire evaluation process with LLM agents. Firstly, an examinerLLM devises queries. Then, a pair of candidate LLMs engage in a multi-round peer-battle around the query, during which the LLM’s true performance gaps becomevisible. Finally, a committee of LLM judges collectively discuss and determine thewinner, which alleviates bias and promotes fairness. In our extensive experimenton the 17 newest LLMs, Auto-Arena shows the highest correlation with humanpreferences, providing a promising alternative to human evaluation platforms.","Nanyang Technological University, Alibaba Group, Singapore University of Technology and Design"
5,./images/auto_arena_of_llms_20240530.png,Autonomous Agents for Collaborative Task under Information Asymmetry,"Wei Liu, Chenxi Wang, Yifei Wang, Zihao Xie, Rennai Qiu, Yufan Dang, Zhuoyun Du, Weize Chen, Cheng Yang, Chen Qian","Large Language Model Multi-Agent Systems (LLM-MAS) have achieved greatprogress in solving complex tasks. It performs communication among agents withinthe system to collaboratively solve tasks, under the premise of shared information.However, when agents’ communication is leveraged to enhance human cooperation,a new challenge arises due to information asymmetry, since each agent can onlyaccess the information of its human user. Previous MAS struggle to complete tasksunder this condition. To address this, we propose a new MAS paradigm termediAgents, which denotes Informative Multi-Agent Systems. In iAgents, the humansocial network is mirrored in the agent network, where agents proactively exchangehuman information necessary for task resolution, thereby overcoming informationasymmetry. iAgents employs a novel agent reasoning mechanism, InfoNav, tonavigate agents’ communication towards effective information exchange. Togetherwith InfoNav, iAgents organizes human information in a mixed memory to provideagents with accurate and comprehensive information for exchange. Additionally,we introduce InformativeBench, the first benchmark tailored for evaluating LLMagents’ task-solving ability under information asymmetry. Experimental resultsshow that iAgents can collaborate within a social network of 140 individualsand 588 relationships, autonomously communicate over 30 turns, and retrieveinformation from nearly 70,000 messages to complete tasks within 3 minutes.","Tsinghua University, Beijing University of Posts and Telecommunications"
6,./images/autonomous_agents_for_collaborative_20240621.png,Avalon's Game of Thoughts: Battle Against Deception through Recursive Contemplation,"Shenzhi Wang, Chang Liu, Zilong Zheng, Siyuan Qi, Shuo Chen, Qisen Yang, Andrew Zhao, Chaofei Wang, Shiji Song, Gao Huang","Recent breakthroughs in large language models (LLMs) have brought remark-able success in the field of LLM-as-Agent. Nevertheless, a prevalent assumptionis that the information processed by LLMs is consistently honest, neglecting thepervasive deceptive or misleading information in human society and AI-generatedcontent.This oversight makes LLMs susceptible to malicious manipulations,potentially resulting in detrimental outcomes. This study utilizes the intricateAvalon game as a testbed to explore LLMs’ potential in deceptive environments.Avalon, full of misinformation and requiring sophisticated logic, manifests as a“Game-of-Thoughts”. Inspired by the efficacy of humans’ recursive thinking andperspective-taking in the Avalon game, we introduce a novel framework, Recur-sive Contemplation (ReCon), to enhance LLMs’ ability to identify and counteractdeceptive information. ReCon combines formulation and refinement contempla-tion processes; formulation contemplation produces initial thoughts and speech,while refinement contemplation further polishes them. Additionally, we incor-porate first-order and second-order perspective transitions into these processesrespectively. Specifically, the first-order allows an LLM agent to infer others’mental states, and the second-order involves understanding how others perceivethe agent’s mental state.......","Tsinghua University, BIGAI, Technical University of Munich"
7,./images/avalon's_game_of_thoughts_20231002.png,Beyond Natural Language: LLMs Leveraging Alternative Formats for Enhanced Reasoning and Communication,"Weize Chen, Chenfei Yuan, Jiarui Yuan, Yusheng Su, Chen Qian, Cheng Yang, Ruobing Xie, Zhiyuan Liu, Maosong Sun","Natural language (NL) has long been the predominant format for human cognition and communication, and by extension, has been similarly pivotal in the development and application of Large Language Models (LLMs). Yet, besides NL, LLMs have seen various non-NL formats during pre-training, such as code and logical expression. NL's status as the optimal format for LLMs, particularly in single-LLM reasoning and multi-agent communication, has not been thoroughly examined. In this work, we challenge the default use of NL by exploring the utility of non-NL formats in these contexts. We show that allowing LLMs to autonomously select the most suitable format before reasoning or communicating leads to a 3.3 to 5.7\% improvement in reasoning efficiency for different LLMs, and up to a 72.7\% reduction in token usage in multi-agent communication, all while maintaining communicative effectiveness. Our comprehensive analysis further reveals that LLMs can devise a format from limited task instructions and that the devised format is effectively transferable across different LLMs. Intriguingly, the structured communication format decided by LLMs exhibits notable parallels with established agent communication languages, suggesting a natural evolution towards efficient, structured communication in agent communication.","Tsinghua University, Tencent, Beijing University of Posts and Telecommunications"
8,./images/beyond_natural_language_llms_20240228.png,Building Cooperative Embodied Agents Modularly with Large Language Models,"Hongxin Zhang, Weihua Du, Jiaming Shan, Qinhong Zhou, Yilun Du, Joshua B. Tenenbaum, Tianmin Shu, Chuang Gan","In this work, we address challenging multi-agent cooperation problems with de-centralized control, raw sensory observations, costly communication, and multi-objective tasks instantiated in various embodied environments. While previous re-search either presupposes a cost-free communication channel or relies on a central-ized controller with shared observations, we harness the commonsense knowledge,reasoning ability, language comprehension, and text generation prowess of LLMsand seamlessly incorporate them into a cognitive-inspired modular framework thatintegrates with perception, memory, and execution. Thus building a CooperativeEmbodied Language Agent CoELA, who can plan, communicate, and cooperatewith others to accomplish long-horizon tasks efficiently. Our experiments on C-WAH and TDW-MAT demonstrate that CoELA driven by GPT-4 can surpass strongplanning-based methods and exhibit emergent effective communication. Thoughcurrent Open LMs like LLAMA-2 still underperform, we fine-tune a CoLLAMAwith data collected with our agents and show how they can achieve promisingperformance. We also conducted a user study for human-agent interaction anddiscovered that CoELA communicating in natural language can earn more trust andcooperate more effectively with humans. Our research underscores the potential ofLLMs for future research in multi-agent cooperation. Videos can be found on theproject website https://vis-www.cs.umass.edu/Co-LLM-Agents/.","University of Massachusetts Amherst, Tsinghua University, Shanghai Jiao Tong University, MIT, MIT-IBM Watson AI Lab"
9,./images/building_cooperative_embodied_agents_20230705.png,"CAMEL: Communicative Agents for ""Mind"" Exploration of Large Language Model Society","Guohao Li, Hasan Abed Al Kader Hammoud, Hani Itani, Dmitrii Khizbullin, Bernard Ghanem","The rapid advancement of chat-based language models has led to remarkableprogress in complex task-solving. However, their success heavily relies on humaninput to guide the conversation, which can be challenging and time-consuming.This paper explores the potential of building scalable techniques to facilitate au-tonomous cooperation among communicative agents, and provides insight intotheir “cognitive” processes. To address the challenges of achieving autonomouscooperation, we propose a novel communicative agent framework named role-playing . Our approach involves using inception prompting to guide chat agentstoward task completion while maintaining consistency with human intentions. Weshowcase how role-playing can be used to generate conversational data for studyingthe behaviors and capabilities of a society of agents, providing a valuable resourcefor investigating conversational language models. In particular, we conduct com-prehensive studies on instruction-following cooperation in multi-agent settings.Our contributions include introducing a novel communicative agent framework,offering a scalable approach for studying the cooperative behaviors and capabili-ties of multi-agent systems, and open-sourcing our library to support research oncommunicative agents and beyond: https://github.com/camel-ai/camel.",King Abdullah University of Science and Technology
10,./images/camel_communicative_agents_for_20230331.png,ChatDev: Communicative Agents for Software Development,"Chen Qian, Wei Liu, Hongzhang Liu, Nuo Chen, Yufan Dang, Jiahao Li, Cheng Yang, Weize Chen, Yusheng Su, Xin Cong, Juyuan Xu, Dahai Li, Zhiyuan Liu, Maosong Sun","Software development is a complex task thatnecessitates cooperation among multiple mem-bers with diverse skills. Numerous studies useddeep learning to improve specific phases in awaterfall model, such as design, coding, andtesting.However, the deep learning modelin each phase requires unique designs, lead-ing to technical inconsistencies across variousphases, which results in a fragmented and in-effective development process. In this paper,we introduce ChatDev, a chat-powered soft-ware development framework in which special-ized agents driven by large language models(LLMs) are guided in what to communicate(via chat chain) and how to communicate (viacommunicative dehallucination). These agentsactively contribute to the design, coding, andtesting phases through unified language-basedcommunication, with solutions derived fromtheir multi-turn dialogues. We found their uti-lization of natural language is advantageousfor system design, and communicating in pro-gramming language proves helpful in debug-ging. This paradigm demonstrates how linguis-tic communication facilitates multi-agent col-laboration, establishing language as a unify-ing bridge for autonomous task-solving amongLLM agents. The code and data are availableat https://github.com/OpenBMB/ChatDev.","Tsinghua University, The University of Sydney, BUPT, Modelbest Inc."
11,./images/chatdev_communicative_agents_for_20230716.png,Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate,"Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Zhaopeng Tu, Shuming Shi","Modern large language models (LLMs) likeChatGPT have shown remarkable performanceon general language tasks but still struggle oncomplex reasoning tasks, which drives the re-search on cognitive behaviors of LLMs to ex-plore human-like problem-solving strategies.Along this direction, one representative strat-egy is self-reflection, which asks an LLM torefine the solution with the feedback gener-ated by itself iteratively. However, our studyshows that such reflection-style methods suf-fer from the Degeneration-of-Thought (DoT)problem: once the LLM has established confi-dence in its solutions, it is unable to generatenovel thoughts later through reflection even ifits initial stance is incorrect. To address theDoT problem, we propose a Multi-Agent De-bate (MAD) framework, in which multipleagents express their arguments in the state of“tit for tat” and a judge manages the debateprocess to obtain a final solution. Clearly, ourMAD framework encourages divergent think-ing in LLMs which would be helpful for tasksthat require deep levels of contemplation. Ex-periment results on two challenging datasets,commonsense machine translation and counter-intuitive arithmetic reasoning, demonstrate theeffectiveness of our MAD framework. Exten-sive analyses suggest that the adaptive break ofdebate and the modest level of “tit for tat” stateare required for MAD to obtain good perfor-mance. Moreover, we find that LLMs might notbe a fair judge if different LLMs are used foragents. Code is available at https://github.com/Skytliang/Multi-Agents-Debate.","Tsinghua University, Shanghai Jiao Tong University, Tencent AI Lab"
12,./images/encouraging_divergent_thinking_in_20230530.png,Examining Inter-Consistency of Large Language Models Collaboration: An In-depth Analysis via Debate,"Kai Xiong, Xiao Ding, Yixin Cao, Ting Liu, Bing Qin","Large Language Models (LLMs) have shownimpressive capabilities in various applications,but they still face various inconsistency issues.Existing works primarily focus on the incon-sistency issues within a single LLM, while wecomplementarily explore the inter-consistencyamong multiple LLMs for collaboration. Toexamine whether LLMs can collaborate effec-tively to achieve a consensus for a shared goal,we focus on commonsense reasoning, and in-troduce a formal debate framework (FORD)to conduct a three-stage debate among LLMswith real-world scenarios alignment: fair de-bate, mismatched debate, and roundtable de-bate. Through extensive experiments on var-ious datasets, LLMs can effectively collabo-rate to reach a consensus despite noticeableinter-inconsistencies, but imbalances in theirabilities can lead to domination by superiorLLMs. Leveraging a more advanced LLM likeGPT-4 as an authoritative judge can boost col-laboration performance. Our work contributesto understanding the inter-consistency amongLLMs and lays the foundation for develop-ing future collaboration methods. Codes anddata are available at https://github.com/Waste-Wood/FORD.","Harbin Institute of Technology, Singapore Management University"
13,./images/examining_inter-consistency_of_large_20230519.png,Exploring Large Language Models for Communication Games: An Empirical Study on Werewolf,"Yuzhuang Xu, Shuo Wang, Peng Li, Fuwen Luo, Xiaolong Wang, Weidong Liu, Yang Liu","Communication games, which we refer to asincomplete information games that heavily de-pend on natural language communication, holdsignificant research value in fields such as eco-nomics, social science, and artificial intelli-gence. In this work, we explore the problem ofhow to engage large language models (LLMs)in communication games, and in response, pro-pose a tuning-free framework. Our approachkeeps LLMs frozen, and relies on the retrievaland reflection on past communications and ex-periences for improvement. An empirical studyon the representative and widely-studied com-munication game, “Werewolf”, demonstratesthat our framework can effectively play Were-wolf game without tuning the parameters of theLLMs. More importantly, strategic behaviorsbegin to emerge in our experiments, suggest-ing that it will be a fruitful journey to engageLLMs in communication games and associateddomains.","Tsinghua University, Zhongguancun Laboratory"
14,./images/exploring_large_language_models_20230909.png,Generative Agents: Interactive Simulacra of Human Behavior,"Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein","Believable proxies of human behavior can empower interactive applications ranging from immersive environments to rehearsal spaces for interpersonal communication to prototyping tools. In this paper, we introduce generative agents--computational software agents that simulate believable human behavior. Generative agents wake up, cook breakfast, and head to work; artists paint, while authors write; they form opinions, notice each other, and initiate conversations; they remember and reflect on days past as they plan the next day. To enable generative agents, we describe an architecture that extends a large language model to store a complete record of the agent's experiences using natural language, synthesize those memories over time into higher-level reflections, and retrieve them dynamically to plan behavior. We instantiate generative agents to populate an interactive sandbox environment inspired by The Sims, where end users can interact with a small town of twenty five agents using natural language. In an evaluation, these generative agents produce believable individual and emergent social behaviors: for example, starting with only a single user-specified notion that one agent wants to throw a Valentine's Day party, the agents autonomously spread invitations to the party over the next two days, make new acquaintances, ask each other out on dates to the party, and coordinate to show up for the party together at the right time. We demonstrate through ablation that the components of our agent architecture--observation, planning, and reflection--each contribute critically to the believability of agent behavior. By fusing large language models with computational, interactive agents, this work introduces architectural and interaction patterns for enabling believable simulations of human behavior.","Stanford University, Google Research, Google DeepMind"
15,./images/generative_agents_interactive_simulacra_20230407.png,Improving Factuality and Reasoning in Language Models through Multiagent Debate,"Yilun Du, Shuang Li, Antonio Torralba, Joshua B. Tenenbaum, Igor Mordatch","Large language models (LLMs) have demonstrated remarkable capabilities inlanguage generation, understanding, and few-shot learning in recent years. Anextensive body of work has explored how their performance may be further im-proved through the tools of prompting, ranging from verification, self-consistency,or intermediate scratchpads. In this paper, we present a complementary approachto improve language responses where multiple language model instances proposeand debate their individual responses and reasoning processes over multiple roundsto arrive at a common final answer. Our findings indicate that this approachsignificantly enhances mathematical and strategic reasoning across a number oftasks. We also demonstrate that our approach improves the factual validity ofgenerated content, reducing fallacious answers and hallucinations that contem-porary models are prone to. Our approach may be directly applied to existingblack-box models and uses identical procedure and prompts for all tasks we inves-tigate. Overall, our findings suggest that such ""society of minds"" approach has thepotential to significantly advance the capabilities of LLMs and pave the way forfurther breakthroughs in language generation and understanding. Project websiteat https://composable-models.github.io/llm_debate/.","MIT CSAIL, Google Brain"
16,./images/improving_factuality_and_reasoning_20230523.png,Improving Language Model Negotiation with Self-Play and In-Context Learning from AI Feedback,"Yao Fu, Hao Peng, Tushar Khot, Mirella Lapata","We study whether multiple large language models (LLMs) can autonomouslyimprove each other in a negotiation game by playing, reflecting, and criticizing.We are interested in this question because if LLMs were able to improve eachother, it would imply the possibility of creating strong AI agents with minimalhuman intervention. We ask two LLMs to negotiate with each other, playingthe roles of a buyer and a seller, respectively. They aim to reach a deal withthe buyer targeting a lower price and the seller a higher one. A third languagemodel, playing the critic, provides feedback to a player to improve the player’snegotiation strategies. We let the two agents play multiple rounds, using previousnegotiation history and AI feedback as in-context demonstrations to improve themodel’s negotiation strategy iteratively. We use different LLMs (GPT and Claude)for different roles and use the deal price as the evaluation metric. Our experimentsreveal multiple intriguing findings: (","University of Edinburgh, Allen Institute for AI, University of Edinburgh"
17,./images/improving_language_model_negotiation_20230517.png,Improving Multi-Agent Debate with Sparse Communication Topology,"Yunxuan Li, Yibing Du, Jiageng Zhang, Le Hou, Peter Grabowski, Yeqing Li, Eugene Ie","Multi-agent debate has proven effective in im-proving large language models quality for rea-soning and factuality tasks. While various role-playing strategies in multi-agent debates havebeen explored, in terms of the communica-tion among agents, existing approaches adopta brute force algorithm – each agent can com-municate with all other agents. In this paper,we systematically investigate the effect of com-munication connectivity in multi-agent systems.Our experiments on GPT and Mistral models re-veal that multi-agent debates leveraging sparsecommunication topology can achieve compara-ble or superior performance while significantlyreducing computational costs. Furthermore, weextend the multi-agent debate framework tomultimodal reasoning and alignment labelingtasks, showcasing its broad applicability andeffectiveness. Our findings underscore the im-portance of communication connectivity on en-hancing the efficiency and effectiveness of the“society of minds” approach.","Google, Google DeepMind"
18,./images/improving_multi-agent_debate_with_20240617.png,LLM-Based Agent Society Investigation: Collaboration and Confrontation in Avalon Gameplay,"Yihuai Lan, Zhiqiang Hu, Lei Wang, Yang Wang, Deheng Ye, Peilin Zhao, Ee-Peng Lim, Hui Xiong, Hao Wang","This paper explores the open research prob-lem of understanding the social behaviors ofLLM-based agents. Using Avalon as a testbed,we employ system prompts to guide LLMagents in gameplay. While previous studieshave touched on gameplay with LLM agents,research on their social behaviors is lacking.We propose a novel framework, tailored forAvalon, features a multi-agent system facil-itating efficient communication and interac-tion. We evaluate its performance based ongame success and analyze LLM agents’ so-cial behaviors. Results affirm the framework’seffectiveness in creating adaptive agents andsuggest LLM-based agents’ potential in nav-igating dynamic social interactions. By ex-amining collaboration and confrontation be-haviors, we offer insights into this field’s re-search and applications.Our code is pub-licly available at https://github.com/3DAgentWorld/LLM-Game-Agent","The Hong Kong University of Science and Technology (Guangzhou), Singapore University of Technology and Design, Singapore Management University, Verily Life Sciences, Tencent"
19,./images/llm-based_agent_society_investigation_20231023.png,LM vs LM: Detecting Factual Errors via Cross Examination,"Roi Cohen, May Hamri, Mor Geva, Amir Globerson","A prominent weakness of modern languagemodels (LMs) is their tendency to generate fac-tually incorrect text, which hinders their us-ability. A natural question is whether such fac-tual errors can be detected automatically. In-spired by truth-seeking mechanisms in law, wepropose a factuality evaluation framework forLMs that is based on cross-examination. Ourkey idea is that an incorrect claim is likely toresult in inconsistency with other claims thatthe model generates. To discover such incon-sistencies, we facilitate a multi-turn interactionbetween the LM that generated the claim andanother LM (acting as an examiner) which in-troduces questions to discover inconsistencies.We empirically evaluate our method on factualclaims made by multiple recent LMs on fourbenchmarks, finding that it outperforms exist-ing methods and baselines, often by a largegap. Our results demonstrate the potential ofusing interacting LMs to capture factual errors.","Tel Aviv University, Google DeepMind, Google Research"
20,./images/lm_vs_lm_detecting_20230522.png,PLAYER*: Enhancing LLM-based Multi-Agent Communication and Interaction in Murder Mystery Games,"Qinglin Zhu, Runcong Zhao, Jinhua Du, Lin Gui, Yulan He","We propose PLAYER*, a novel framework that addresses the limitations of existing agent-based approaches built on Large Language Models (LLMs) in handling complex questions and understanding interpersonal relationships in dynamic environments. PLAYER* enhances path planning in Murder Mystery Games (MMGs) using an anytime sampling-based planner and a questioning-driven search framework. By equipping agents with a set of sensors, PLAYER* eliminates the need for pre-defined questions and enables agents to navigate complex social interactions. We additionally make a contribution by introducing a quantifiable evaluation method using multiple-choice questions and present WellPlay, a dataset containing 1,482 question-answer pairs. Experimental results demonstrate PLAYER*'s superiority over existing multi-agent methods, enhancing the generalisability and adaptability of agents in MMGs and paving the way for more effective multi-agent interactions.","King’s College London, Huawei London Research Centre, The Alan Turing Institute"
21,./images/player_enhancing_llm-based_multi-agent_20240426.png,RoCo: Dialectic Multi-Robot Collaboration with Large Language Models,"Zhao Mandi, Shreeya Jain, Shuran Song",": We propose a novel approach to multi-robot collaboration that har-nesses the power of pre-trained large language models (LLMs) for both high-levelcommunication and low-level path planning. Robots are equipped with LLMs todiscuss and collectively reason task strategies. They then generate sub-task plansand task space waypoint paths, which are used by a multi-arm motion planner toaccelerate trajectory planning. We also provide feedback from the environment,such as collision checking, and prompt the LLM agents to improve their plan andwaypoints in-context. For evaluation, we introduce RoCoBench, a 6-task bench-mark covering a wide range of multi-robot collaboration scenarios, accompaniedby a text-only dataset for agent representation and reasoning. We experimentallydemonstrate the effectiveness of our approach – it achieves high success ratesacross all tasks in RoCoBench and adapts to variations in task semantics. Our di-alog setup offers high interpretability and flexibility – in real world experiments,we show RoCo easily incorporates human-in-the-loop, where a user can commu-nicate and collaborate with a robot agent to complete tasks together. See projectwebsite project-roco.github.io for videos and code.",Columbia University
22,./images/roco_dialectic_multi-robot_collaboration_20230710.png,Scaling Large-Language-Model-based Multi-Agent Collaboration,"Chen Qian, Zihao Xie, Yifei Wang, Wei Liu, Yufan Dang, Zhuoyun Du, Weize Chen, Cheng Yang, Zhiyuan Liu, Maosong Sun","Pioneering advancements in large languagemodel-powered agents have underscored thedesign pattern of multi-agent collaboration,demonstrating that collective intelligence cansurpass the capabilities of each individual. In-spired by the neural scaling law, which positsthat increasing neurons leads to emergent abil-ities, this study investigates whether a simi-lar principle applies to increasing agents inmulti-agent collaboration.Technically, wepropose ::multi-agent:collaboration::networks(MACNET), which utilize directed acyclicgraphs to organize agents and streamline theirinteractive reasoning via topological ordering,with solutions derived from their dialogues.Extensive experiments show that MACNETconsistently outperforms baseline models, en-abling effective agent collaboration across var-ious network topologies and supporting coop-eration among more than a thousand agents.Notably, we observed a small-world collabo-ration phenomenon, where topologies resem-bling small-world properties achieved supe-rior performance. Additionally, we identifieda collaborative scaling law, indicating thatnormalized solution quality follows a logisticgrowth pattern as scaling agents, with collabo-rative emergence occurring much earlier thanpreviously observed instances of neural emer-gence. The code and data will be available athttps://github.com/OpenBMB/ChatDev.","Tsinghua University, Beijing University of Posts and Telecommunications"
23,./images/scaling_large-language-model-based_multi-agent_collaboration_20240611.png,The Impact of Language on Arithmetic Proficiency- A Multilingual Investigation with Cross-Agent Checking Computation,"Chung-Chi Chen, Hiroya Takamura, Ichiro Kobayashi, Yusuke Miyao","This paper critically examines the arithmetic capabilities of Large Language Models (LLMs), uncovering significant limitations in their performance. Our research reveals a notable decline in accuracy for complex calculations involving large numbers, with addition and subtraction tasks showing varying degrees of proficiency. Additionally, we challenge the notion that arithmetic is language-independent, finding up to a 10% difference in performance across twenty languages. The study also compares self-verification methods with cross-agent collaborations, showing that a single model often outperforms collaborative approaches in basic arithmetic tasks. These findings suggest a need to reassess the effectiveness of LLMs in tasks requiring numerical accuracy and precision.","AIST, University of Tokyo"
24,./images/the_impact_of_language_20240616.png,Theory of Mind for Multi-Agent Collaboration via Large Language Models,"Huao Li, Yu Quan Chong, Simon Stepputtis, Joseph Campbell, Dana Hughes, Michael Lewis, Katia Sycara","While Large Language Models (LLMs) havedemonstrated impressive accomplishments inboth reasoning and planning, their abilitiesin multi-agent collaborations remains largelyunexplored.This study evaluates LLM-based agents in a multi-agent cooperative textgame with Theory of Mind (ToM) inferencetasks, comparing their performance with Multi-Agent Reinforcement Learning (MARL) andplanning-based baselines. We observed evi-dence of emergent collaborative behaviors andhigh-order Theory of Mind capabilities amongLLM-based agents. Our results reveal limi-tations in LLM-based agents’ planning opti-mization due to systematic failures in managinglong-horizon contexts and hallucination aboutthe task state. We explore the use of explicitbelief state representations to mitigate these is-sues, finding that it enhances task performanceand the accuracy of ToM inferences for LLM-based agents.","University of Pittsburgh, Carnegie Mellon University"
25,./images/theory_of_mind_for_20231016.png,Toward Optimal LLM Alignments Using Two-Player Games,"Rui Zheng, Hongyi Guo, Zhihan Liu, Xiaoying Zhang, Yuanshun Yao, Xiaojun Xu, Zhaoran Wang, Zhiheng Xi, Tao Gui, Qi Zhang, Xuanjing Huang, Hang Li, Yang Liu","Alignment of large language models is a critical process designed to ensure thatthe model’s responses to user prompts accurately reflect human intentions andadhere to societal values. The standard Reinforcement Learning from HumanFeedback (RLHF) framework primarily focuses on optimizing the performance oflarge language models using pre-collected prompts. However, collecting promptsthat provide comprehensive coverage is both tedious and challenging, and oftenfails to include scenarios that LLMs need to improve on the most. In this paper,we investigate alignment through the lens of two-agent games, involving iterativeinteractions between an adversarial and a defensive agent. The adversarial agent’stask at each step is to generate prompts that expose the weakness of the defensiveagent. In return, the defensive agent seeks to improve its responses to these newlyidentified prompts it “struggled"" with, based on feedback from the reward model.We theoretically demonstrate that this iterative reinforcement learning optimizationconverges to a Nash Equilibrium for the game induced by the agents. Experi-mental results in safety scenarios demonstrate that learning in such a competitiveenvironment not only fully trains agents but also leads to policies with enhancedgeneralization capabilities for both adversarial and defensive agents. Our code isreleased at https://github.com/ruizheng20/gpo.","Fudan University, Northwestern University, ByteDance Research"
26,./images/toward_optimal_llm_alignments_20240616.png,Towards Detecting LLMs Hallucination via Markov Chain-based Multi-agent Debate Framework,"Xiaoxi Sun, Jinpeng Li, Yan Zhong, Dongyan Zhao, Rui Yan","The advent of large language models (LLMs)has facilitated the development of natural lan-guage text generation. It also poses unprece-dented challenges, with content hallucinationemerging as a significant concern. Existingsolutions often involve expensive and complexinterventions during the training process. More-over, some approaches emphasize problem dis-assembly while neglecting the crucial valida-tion process, leading to performance degrada-tion or limited applications. To overcome theselimitations, we propose a Markov Chain-basedmulti-agent debate verification framework toenhance hallucination detection accuracy inconcise claims. Our method integrates the fact-checking process, including claim detection,evidence retrieval, and multi-agent verification.In the verification stage, we deploy multipleagents through flexible Markov Chain-baseddebates to validate individual claims, ensuringmeticulous verification outcomes. Experimen-tal results across three generative tasks demon-strate that our approach achieves significantimprovements over baselines.","Peking University, Renmin University of China"
27,./images/towards_detecting_llms_hallucination_20240605.png,To be Continued...,Your Contributions are Welcome!,,
1 image_path title author summary affiliation
2 0 ./images/1d.png AgentCF: Collaborative Learning with Autonomous Language Agents for Recommender Systems Junjie Zhang, Yupeng Hou, Ruobing Xie, Wenqi Sun, Julian McAuley, Wayne Xin Zhao, Leyu Lin, Ji-Rong Wen Recently, there has been an emergence of employing LLM-poweredagents as believable human proxies, based on their remarkabledecision-making capability. However, existing studies mainly focuson simulating human dialogue. Human non-verbal behaviors, suchas item clicking in recommender systems, although implicitly ex-hibiting user preferences and could enhance the modeling of users,have not been deeply explored. The main reasons lie in the gapbetween language modeling and behavior modeling, as well as theincomprehension of LLMs about user-item relations.To address this issue, we propose AgentCF for simulating user-item interactions in recommender systems through agent-basedcollaborative filtering. We creatively consider not only users butalso items as agents, and develop a collaborative learning approachthat optimizes both kinds of agents together. Specifically, at eachtime step, we first prompt the user and item agents to interact au-tonomously. Then, based on the disparities between the agents’decisions and real-world interaction records, user and item agentsare prompted to reflect on and adjust the misleading simulationscollaboratively, thereby modeling their two-sided relations. The op-timized agents can also propagate their preferences to other agentsin subsequent interactions, implicitly capturing the collaborative fil-tering idea. Overall, the optimized agents exhibit diverse interactionbehaviors within our framework, including user-item, user-user,item-item, and collective interactions. The results show that theseagents can demonstrate personalized behaviors akin to those of real-world individuals, sparking the development of next-generationuser behavior simulation. Renmin University of China, UC San Diego, Tencent
3 1 ./images/agentcf_collaborative_learning_with_20231013.png AgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors Weize Chen, Yusheng Su, Jingwei Zuo, Cheng Yang, Chenfei Yuan, Chi-Min Chan, Heyang Yu, Yaxi Lu, Yi-Hsin Hung, Chen Qian, Yujia Qin, Xin Cong, Ruobing Xie, Zhiyuan Liu, Maosong Sun, Jie Zhou Autonomous agents empowered by Large Language Models (LLMs) have under-gone significant improvements, enabling them to generalize across a broad spec-trum of tasks. However, in real-world scenarios, cooperation among individuals isoften required to enhance the efficiency and effectiveness of task accomplishment.Hence, inspired by human group dynamics, we propose a multi-agent frameworkAGENTVERSE that can effectively orchestrate a collaborative group of expert agentsas a greater-than-the-sum-of-its-parts system. Our experiments demonstrate thatAGENTVERSE can proficiently deploy multi-agent groups that outperform a singleagent. Extensive experiments on text understanding, reasoning, coding, tool utiliza-tion, and embodied AI confirm the effectiveness of AGENTVERSE. Moreover, ouranalysis of agent interactions within AGENTVERSE reveals the emergence of spe-cific collaborative behaviors, contributing to heightened group efficiency. Our codehas been released at https://github.com/OpenBMB/AgentVerse/. Tsinghua University, Beijing University of Posts and Telecommunications, Tencent Inc.
4 2 ./images/agentverse_facilitating_multi-agent_collaboration_20230821.png Apollo's Oracle: Retrieval-Augmented Reasoning in Multi-Agent Debates Haotian Wang, Xiyuan Du, Weijiang Yu, Qianglong Chen, Kun Zhu, Zheng Chu, Lian Yan, Yi Guan Multi-agent debate systems are designed to derive accurate and consistent conclusions through adversarial interactions among agents. However, these systems often encounter challenges due to cognitive constraints, manifesting as (1) agents' obstinate adherence to incorrect viewpoints and (2) their propensity to abandon correct viewpoints. These issues are primarily responsible for the ineffectiveness of such debates. Addressing the challenge of cognitive constraints, we introduce a novel framework, the Multi-Agent Debate with Retrieval Augmented (MADRA). MADRA incorporates retrieval of prior knowledge into the debate process, effectively breaking cognitive constraints and enhancing the agents' reasoning capabilities. Furthermore, we have developed a self-selection module within this framework, enabling agents to autonomously select pertinent evidence, thereby minimizing the impact of irrelevant or noisy data. We have comprehensively tested and analyzed MADRA across six diverse datasets. The experimental results demonstrate that our approach significantly enhances performance across various tasks, proving the effectiveness of our proposed method. Harbin Institute of Technology, Sun Yat-sen University, Zhejiang University
5 3 ./images/apollo's_oracle_retrieval-augmented_reasoning_20231208.png ATM: Adversarial Tuning Multi-agent System Makes a Robust Retrieval-Augmented Generator Junda Zhu, Lingyong Yan, Haibo Shi, Dawei Yin, Lei Sha Large language models (LLMs) are proven tobenefit a lot from retrieval-augmented genera-tion (RAG) in alleviating hallucinations con-fronted with knowledge-intensive questions.RAG adopts information retrieval techniquesto inject external knowledge from semantic-relevant documents as input contexts. How-ever, due to today’s Internet being flooded withnumerous noisy and fabricating content, it isinevitable that RAG systems are vulnerableto these noises and prone to respond incor-rectly. To this end, we propose to optimizethe retrieval-augmented GENERATOR with aAdversarial Tuning Multi-agent system (ATM).The ATM steers the GENERATOR to have a ro-bust perspective of useful documents for ques-tion answering with the help of an auxiliaryATTACKER agent. The GENERATOR and theATTACKER are tuned adversarially for severaliterations. After rounds of multi-agent itera-tive tuning, the GENERATOR can eventuallybetter discriminate useful documents amongstfabrications. The experimental results verifythe effectiveness of ATM and we also observethat the GENERATOR can achieve better perfor-mance compared to state-of-the-art baselines. Beihang University, Baidu Inc.
6 4 ./images/atm_adversarial_tuning_multi-agent_20240528.png Auto Arena of LLMs: Automating LLM Evaluations with Agent Peer-battles and Committee Discussions Ruochen Zhao, Wenxuan Zhang, Yew Ken Chia, Deli Zhao, Lidong Bing As LLMs evolve on a daily basis, there is an urgent need for a trustworthy evaluationmethod that can provide robust evaluation results in a timely fashion. Currently,as static benchmarks are prone to contamination concerns, users tend to trusthuman voting platforms, such as Chatbot Arena. However, human annotationsrequire extensive manual efforts. To provide an automatic, robust, and trustworthyevaluation framework, we innovatively propose the Auto-Arena of LLMs, whichautomates the entire evaluation process with LLM agents. Firstly, an examinerLLM devises queries. Then, a pair of candidate LLMs engage in a multi-round peer-battle around the query, during which the LLM’s true performance gaps becomevisible. Finally, a committee of LLM judges collectively discuss and determine thewinner, which alleviates bias and promotes fairness. In our extensive experimenton the 17 newest LLMs, Auto-Arena shows the highest correlation with humanpreferences, providing a promising alternative to human evaluation platforms. Nanyang Technological University, Alibaba Group, Singapore University of Technology and Design
7 5 ./images/auto_arena_of_llms_20240530.png Autonomous Agents for Collaborative Task under Information Asymmetry Wei Liu, Chenxi Wang, Yifei Wang, Zihao Xie, Rennai Qiu, Yufan Dang, Zhuoyun Du, Weize Chen, Cheng Yang, Chen Qian Large Language Model Multi-Agent Systems (LLM-MAS) have achieved greatprogress in solving complex tasks. It performs communication among agents withinthe system to collaboratively solve tasks, under the premise of shared information.However, when agents’ communication is leveraged to enhance human cooperation,a new challenge arises due to information asymmetry, since each agent can onlyaccess the information of its human user. Previous MAS struggle to complete tasksunder this condition. To address this, we propose a new MAS paradigm termediAgents, which denotes Informative Multi-Agent Systems. In iAgents, the humansocial network is mirrored in the agent network, where agents proactively exchangehuman information necessary for task resolution, thereby overcoming informationasymmetry. iAgents employs a novel agent reasoning mechanism, InfoNav, tonavigate agents’ communication towards effective information exchange. Togetherwith InfoNav, iAgents organizes human information in a mixed memory to provideagents with accurate and comprehensive information for exchange. Additionally,we introduce InformativeBench, the first benchmark tailored for evaluating LLMagents’ task-solving ability under information asymmetry. Experimental resultsshow that iAgents can collaborate within a social network of 140 individualsand 588 relationships, autonomously communicate over 30 turns, and retrieveinformation from nearly 70,000 messages to complete tasks within 3 minutes. Tsinghua University, Beijing University of Posts and Telecommunications
8 6 ./images/autonomous_agents_for_collaborative_20240621.png Avalon's Game of Thoughts: Battle Against Deception through Recursive Contemplation Shenzhi Wang, Chang Liu, Zilong Zheng, Siyuan Qi, Shuo Chen, Qisen Yang, Andrew Zhao, Chaofei Wang, Shiji Song, Gao Huang Recent breakthroughs in large language models (LLMs) have brought remark-able success in the field of LLM-as-Agent. Nevertheless, a prevalent assumptionis that the information processed by LLMs is consistently honest, neglecting thepervasive deceptive or misleading information in human society and AI-generatedcontent.This oversight makes LLMs susceptible to malicious manipulations,potentially resulting in detrimental outcomes. This study utilizes the intricateAvalon game as a testbed to explore LLMs’ potential in deceptive environments.Avalon, full of misinformation and requiring sophisticated logic, manifests as a“Game-of-Thoughts”. Inspired by the efficacy of humans’ recursive thinking andperspective-taking in the Avalon game, we introduce a novel framework, Recur-sive Contemplation (ReCon), to enhance LLMs’ ability to identify and counteractdeceptive information. ReCon combines formulation and refinement contempla-tion processes; formulation contemplation produces initial thoughts and speech,while refinement contemplation further polishes them. Additionally, we incor-porate first-order and second-order perspective transitions into these processesrespectively. Specifically, the first-order allows an LLM agent to infer others’mental states, and the second-order involves understanding how others perceivethe agent’s mental state....... Tsinghua University, BIGAI, Technical University of Munich
9 7 ./images/avalon's_game_of_thoughts_20231002.png Beyond Natural Language: LLMs Leveraging Alternative Formats for Enhanced Reasoning and Communication Weize Chen, Chenfei Yuan, Jiarui Yuan, Yusheng Su, Chen Qian, Cheng Yang, Ruobing Xie, Zhiyuan Liu, Maosong Sun Natural language (NL) has long been the predominant format for human cognition and communication, and by extension, has been similarly pivotal in the development and application of Large Language Models (LLMs). Yet, besides NL, LLMs have seen various non-NL formats during pre-training, such as code and logical expression. NL's status as the optimal format for LLMs, particularly in single-LLM reasoning and multi-agent communication, has not been thoroughly examined. In this work, we challenge the default use of NL by exploring the utility of non-NL formats in these contexts. We show that allowing LLMs to autonomously select the most suitable format before reasoning or communicating leads to a 3.3 to 5.7\% improvement in reasoning efficiency for different LLMs, and up to a 72.7\% reduction in token usage in multi-agent communication, all while maintaining communicative effectiveness. Our comprehensive analysis further reveals that LLMs can devise a format from limited task instructions and that the devised format is effectively transferable across different LLMs. Intriguingly, the structured communication format decided by LLMs exhibits notable parallels with established agent communication languages, suggesting a natural evolution towards efficient, structured communication in agent communication. Tsinghua University, Tencent, Beijing University of Posts and Telecommunications
10 8 ./images/beyond_natural_language_llms_20240228.png Building Cooperative Embodied Agents Modularly with Large Language Models Hongxin Zhang, Weihua Du, Jiaming Shan, Qinhong Zhou, Yilun Du, Joshua B. Tenenbaum, Tianmin Shu, Chuang Gan In this work, we address challenging multi-agent cooperation problems with de-centralized control, raw sensory observations, costly communication, and multi-objective tasks instantiated in various embodied environments. While previous re-search either presupposes a cost-free communication channel or relies on a central-ized controller with shared observations, we harness the commonsense knowledge,reasoning ability, language comprehension, and text generation prowess of LLMsand seamlessly incorporate them into a cognitive-inspired modular framework thatintegrates with perception, memory, and execution. Thus building a CooperativeEmbodied Language Agent CoELA, who can plan, communicate, and cooperatewith others to accomplish long-horizon tasks efficiently. Our experiments on C-WAH and TDW-MAT demonstrate that CoELA driven by GPT-4 can surpass strongplanning-based methods and exhibit emergent effective communication. Thoughcurrent Open LMs like LLAMA-2 still underperform, we fine-tune a CoLLAMAwith data collected with our agents and show how they can achieve promisingperformance. We also conducted a user study for human-agent interaction anddiscovered that CoELA communicating in natural language can earn more trust andcooperate more effectively with humans. Our research underscores the potential ofLLMs for future research in multi-agent cooperation. Videos can be found on theproject website https://vis-www.cs.umass.edu/Co-LLM-Agents/. University of Massachusetts Amherst, Tsinghua University, Shanghai Jiao Tong University, MIT, MIT-IBM Watson AI Lab
11 9 ./images/building_cooperative_embodied_agents_20230705.png CAMEL: Communicative Agents for "Mind" Exploration of Large Language Model Society Guohao Li, Hasan Abed Al Kader Hammoud, Hani Itani, Dmitrii Khizbullin, Bernard Ghanem The rapid advancement of chat-based language models has led to remarkableprogress in complex task-solving. However, their success heavily relies on humaninput to guide the conversation, which can be challenging and time-consuming.This paper explores the potential of building scalable techniques to facilitate au-tonomous cooperation among communicative agents, and provides insight intotheir “cognitive” processes. To address the challenges of achieving autonomouscooperation, we propose a novel communicative agent framework named role-playing . Our approach involves using inception prompting to guide chat agentstoward task completion while maintaining consistency with human intentions. Weshowcase how role-playing can be used to generate conversational data for studyingthe behaviors and capabilities of a society of agents, providing a valuable resourcefor investigating conversational language models. In particular, we conduct com-prehensive studies on instruction-following cooperation in multi-agent settings.Our contributions include introducing a novel communicative agent framework,offering a scalable approach for studying the cooperative behaviors and capabili-ties of multi-agent systems, and open-sourcing our library to support research oncommunicative agents and beyond: https://github.com/camel-ai/camel. King Abdullah University of Science and Technology
12 10 ./images/camel_communicative_agents_for_20230331.png ChatDev: Communicative Agents for Software Development Chen Qian, Wei Liu, Hongzhang Liu, Nuo Chen, Yufan Dang, Jiahao Li, Cheng Yang, Weize Chen, Yusheng Su, Xin Cong, Juyuan Xu, Dahai Li, Zhiyuan Liu, Maosong Sun Software development is a complex task thatnecessitates cooperation among multiple mem-bers with diverse skills. Numerous studies useddeep learning to improve specific phases in awaterfall model, such as design, coding, andtesting.However, the deep learning modelin each phase requires unique designs, lead-ing to technical inconsistencies across variousphases, which results in a fragmented and in-effective development process. In this paper,we introduce ChatDev, a chat-powered soft-ware development framework in which special-ized agents driven by large language models(LLMs) are guided in what to communicate(via chat chain) and how to communicate (viacommunicative dehallucination). These agentsactively contribute to the design, coding, andtesting phases through unified language-basedcommunication, with solutions derived fromtheir multi-turn dialogues. We found their uti-lization of natural language is advantageousfor system design, and communicating in pro-gramming language proves helpful in debug-ging. This paradigm demonstrates how linguis-tic communication facilitates multi-agent col-laboration, establishing language as a unify-ing bridge for autonomous task-solving amongLLM agents. The code and data are availableat https://github.com/OpenBMB/ChatDev. Tsinghua University, The University of Sydney, BUPT, Modelbest Inc.
13 11 ./images/chatdev_communicative_agents_for_20230716.png Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Zhaopeng Tu, Shuming Shi Modern large language models (LLMs) likeChatGPT have shown remarkable performanceon general language tasks but still struggle oncomplex reasoning tasks, which drives the re-search on cognitive behaviors of LLMs to ex-plore human-like problem-solving strategies.Along this direction, one representative strat-egy is self-reflection, which asks an LLM torefine the solution with the feedback gener-ated by itself iteratively. However, our studyshows that such reflection-style methods suf-fer from the Degeneration-of-Thought (DoT)problem: once the LLM has established confi-dence in its solutions, it is unable to generatenovel thoughts later through reflection even ifits initial stance is incorrect. To address theDoT problem, we propose a Multi-Agent De-bate (MAD) framework, in which multipleagents express their arguments in the state of“tit for tat” and a judge manages the debateprocess to obtain a final solution. Clearly, ourMAD framework encourages divergent think-ing in LLMs which would be helpful for tasksthat require deep levels of contemplation. Ex-periment results on two challenging datasets,commonsense machine translation and counter-intuitive arithmetic reasoning, demonstrate theeffectiveness of our MAD framework. Exten-sive analyses suggest that the adaptive break ofdebate and the modest level of “tit for tat” stateare required for MAD to obtain good perfor-mance. Moreover, we find that LLMs might notbe a fair judge if different LLMs are used foragents. Code is available at https://github.com/Skytliang/Multi-Agents-Debate. Tsinghua University, Shanghai Jiao Tong University, Tencent AI Lab
14 12 ./images/encouraging_divergent_thinking_in_20230530.png Examining Inter-Consistency of Large Language Models Collaboration: An In-depth Analysis via Debate Kai Xiong, Xiao Ding, Yixin Cao, Ting Liu, Bing Qin Large Language Models (LLMs) have shownimpressive capabilities in various applications,but they still face various inconsistency issues.Existing works primarily focus on the incon-sistency issues within a single LLM, while wecomplementarily explore the inter-consistencyamong multiple LLMs for collaboration. Toexamine whether LLMs can collaborate effec-tively to achieve a consensus for a shared goal,we focus on commonsense reasoning, and in-troduce a formal debate framework (FORD)to conduct a three-stage debate among LLMswith real-world scenarios alignment: fair de-bate, mismatched debate, and roundtable de-bate. Through extensive experiments on var-ious datasets, LLMs can effectively collabo-rate to reach a consensus despite noticeableinter-inconsistencies, but imbalances in theirabilities can lead to domination by superiorLLMs. Leveraging a more advanced LLM likeGPT-4 as an authoritative judge can boost col-laboration performance. Our work contributesto understanding the inter-consistency amongLLMs and lays the foundation for develop-ing future collaboration methods. Codes anddata are available at https://github.com/Waste-Wood/FORD. Harbin Institute of Technology, Singapore Management University
15 13 ./images/examining_inter-consistency_of_large_20230519.png Exploring Large Language Models for Communication Games: An Empirical Study on Werewolf Yuzhuang Xu, Shuo Wang, Peng Li, Fuwen Luo, Xiaolong Wang, Weidong Liu, Yang Liu Communication games, which we refer to asincomplete information games that heavily de-pend on natural language communication, holdsignificant research value in fields such as eco-nomics, social science, and artificial intelli-gence. In this work, we explore the problem ofhow to engage large language models (LLMs)in communication games, and in response, pro-pose a tuning-free framework. Our approachkeeps LLMs frozen, and relies on the retrievaland reflection on past communications and ex-periences for improvement. An empirical studyon the representative and widely-studied com-munication game, “Werewolf”, demonstratesthat our framework can effectively play Were-wolf game without tuning the parameters of theLLMs. More importantly, strategic behaviorsbegin to emerge in our experiments, suggest-ing that it will be a fruitful journey to engageLLMs in communication games and associateddomains. Tsinghua University, Zhongguancun Laboratory
16 14 ./images/exploring_large_language_models_20230909.png Generative Agents: Interactive Simulacra of Human Behavior Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein Believable proxies of human behavior can empower interactive applications ranging from immersive environments to rehearsal spaces for interpersonal communication to prototyping tools. In this paper, we introduce generative agents--computational software agents that simulate believable human behavior. Generative agents wake up, cook breakfast, and head to work; artists paint, while authors write; they form opinions, notice each other, and initiate conversations; they remember and reflect on days past as they plan the next day. To enable generative agents, we describe an architecture that extends a large language model to store a complete record of the agent's experiences using natural language, synthesize those memories over time into higher-level reflections, and retrieve them dynamically to plan behavior. We instantiate generative agents to populate an interactive sandbox environment inspired by The Sims, where end users can interact with a small town of twenty five agents using natural language. In an evaluation, these generative agents produce believable individual and emergent social behaviors: for example, starting with only a single user-specified notion that one agent wants to throw a Valentine's Day party, the agents autonomously spread invitations to the party over the next two days, make new acquaintances, ask each other out on dates to the party, and coordinate to show up for the party together at the right time. We demonstrate through ablation that the components of our agent architecture--observation, planning, and reflection--each contribute critically to the believability of agent behavior. By fusing large language models with computational, interactive agents, this work introduces architectural and interaction patterns for enabling believable simulations of human behavior. Stanford University, Google Research, Google DeepMind
17 15 ./images/generative_agents_interactive_simulacra_20230407.png Improving Factuality and Reasoning in Language Models through Multiagent Debate Yilun Du, Shuang Li, Antonio Torralba, Joshua B. Tenenbaum, Igor Mordatch Large language models (LLMs) have demonstrated remarkable capabilities inlanguage generation, understanding, and few-shot learning in recent years. Anextensive body of work has explored how their performance may be further im-proved through the tools of prompting, ranging from verification, self-consistency,or intermediate scratchpads. In this paper, we present a complementary approachto improve language responses where multiple language model instances proposeand debate their individual responses and reasoning processes over multiple roundsto arrive at a common final answer. Our findings indicate that this approachsignificantly enhances mathematical and strategic reasoning across a number oftasks. We also demonstrate that our approach improves the factual validity ofgenerated content, reducing fallacious answers and hallucinations that contem-porary models are prone to. Our approach may be directly applied to existingblack-box models and uses identical procedure and prompts for all tasks we inves-tigate. Overall, our findings suggest that such "society of minds" approach has thepotential to significantly advance the capabilities of LLMs and pave the way forfurther breakthroughs in language generation and understanding. Project websiteat https://composable-models.github.io/llm_debate/. MIT CSAIL, Google Brain
18 16 ./images/improving_factuality_and_reasoning_20230523.png Improving Language Model Negotiation with Self-Play and In-Context Learning from AI Feedback Yao Fu, Hao Peng, Tushar Khot, Mirella Lapata We study whether multiple large language models (LLMs) can autonomouslyimprove each other in a negotiation game by playing, reflecting, and criticizing.We are interested in this question because if LLMs were able to improve eachother, it would imply the possibility of creating strong AI agents with minimalhuman intervention. We ask two LLMs to negotiate with each other, playingthe roles of a buyer and a seller, respectively. They aim to reach a deal withthe buyer targeting a lower price and the seller a higher one. A third languagemodel, playing the critic, provides feedback to a player to improve the player’snegotiation strategies. We let the two agents play multiple rounds, using previousnegotiation history and AI feedback as in-context demonstrations to improve themodel’s negotiation strategy iteratively. We use different LLMs (GPT and Claude)for different roles and use the deal price as the evaluation metric. Our experimentsreveal multiple intriguing findings: ( University of Edinburgh, Allen Institute for AI, University of Edinburgh
19 17 ./images/improving_language_model_negotiation_20230517.png Improving Multi-Agent Debate with Sparse Communication Topology Yunxuan Li, Yibing Du, Jiageng Zhang, Le Hou, Peter Grabowski, Yeqing Li, Eugene Ie Multi-agent debate has proven effective in im-proving large language models quality for rea-soning and factuality tasks. While various role-playing strategies in multi-agent debates havebeen explored, in terms of the communica-tion among agents, existing approaches adopta brute force algorithm – each agent can com-municate with all other agents. In this paper,we systematically investigate the effect of com-munication connectivity in multi-agent systems.Our experiments on GPT and Mistral models re-veal that multi-agent debates leveraging sparsecommunication topology can achieve compara-ble or superior performance while significantlyreducing computational costs. Furthermore, weextend the multi-agent debate framework tomultimodal reasoning and alignment labelingtasks, showcasing its broad applicability andeffectiveness. Our findings underscore the im-portance of communication connectivity on en-hancing the efficiency and effectiveness of the“society of minds” approach. Google, Google DeepMind
20 18 ./images/improving_multi-agent_debate_with_20240617.png LLM-Based Agent Society Investigation: Collaboration and Confrontation in Avalon Gameplay Yihuai Lan, Zhiqiang Hu, Lei Wang, Yang Wang, Deheng Ye, Peilin Zhao, Ee-Peng Lim, Hui Xiong, Hao Wang This paper explores the open research prob-lem of understanding the social behaviors ofLLM-based agents. Using Avalon as a testbed,we employ system prompts to guide LLMagents in gameplay. While previous studieshave touched on gameplay with LLM agents,research on their social behaviors is lacking.We propose a novel framework, tailored forAvalon, features a multi-agent system facil-itating efficient communication and interac-tion. We evaluate its performance based ongame success and analyze LLM agents’ so-cial behaviors. Results affirm the framework’seffectiveness in creating adaptive agents andsuggest LLM-based agents’ potential in nav-igating dynamic social interactions. By ex-amining collaboration and confrontation be-haviors, we offer insights into this field’s re-search and applications.Our code is pub-licly available at https://github.com/3DAgentWorld/LLM-Game-Agent The Hong Kong University of Science and Technology (Guangzhou), Singapore University of Technology and Design, Singapore Management University, Verily Life Sciences, Tencent
21 19 ./images/llm-based_agent_society_investigation_20231023.png LM vs LM: Detecting Factual Errors via Cross Examination Roi Cohen, May Hamri, Mor Geva, Amir Globerson A prominent weakness of modern languagemodels (LMs) is their tendency to generate fac-tually incorrect text, which hinders their us-ability. A natural question is whether such fac-tual errors can be detected automatically. In-spired by truth-seeking mechanisms in law, wepropose a factuality evaluation framework forLMs that is based on cross-examination. Ourkey idea is that an incorrect claim is likely toresult in inconsistency with other claims thatthe model generates. To discover such incon-sistencies, we facilitate a multi-turn interactionbetween the LM that generated the claim andanother LM (acting as an examiner) which in-troduces questions to discover inconsistencies.We empirically evaluate our method on factualclaims made by multiple recent LMs on fourbenchmarks, finding that it outperforms exist-ing methods and baselines, often by a largegap. Our results demonstrate the potential ofusing interacting LMs to capture factual errors. Tel Aviv University, Google DeepMind, Google Research
22 20 ./images/lm_vs_lm_detecting_20230522.png PLAYER*: Enhancing LLM-based Multi-Agent Communication and Interaction in Murder Mystery Games Qinglin Zhu, Runcong Zhao, Jinhua Du, Lin Gui, Yulan He We propose PLAYER*, a novel framework that addresses the limitations of existing agent-based approaches built on Large Language Models (LLMs) in handling complex questions and understanding interpersonal relationships in dynamic environments. PLAYER* enhances path planning in Murder Mystery Games (MMGs) using an anytime sampling-based planner and a questioning-driven search framework. By equipping agents with a set of sensors, PLAYER* eliminates the need for pre-defined questions and enables agents to navigate complex social interactions. We additionally make a contribution by introducing a quantifiable evaluation method using multiple-choice questions and present WellPlay, a dataset containing 1,482 question-answer pairs. Experimental results demonstrate PLAYER*'s superiority over existing multi-agent methods, enhancing the generalisability and adaptability of agents in MMGs and paving the way for more effective multi-agent interactions. King’s College London, Huawei London Research Centre, The Alan Turing Institute
23 21 ./images/player_enhancing_llm-based_multi-agent_20240426.png RoCo: Dialectic Multi-Robot Collaboration with Large Language Models Zhao Mandi, Shreeya Jain, Shuran Song : We propose a novel approach to multi-robot collaboration that har-nesses the power of pre-trained large language models (LLMs) for both high-levelcommunication and low-level path planning. Robots are equipped with LLMs todiscuss and collectively reason task strategies. They then generate sub-task plansand task space waypoint paths, which are used by a multi-arm motion planner toaccelerate trajectory planning. We also provide feedback from the environment,such as collision checking, and prompt the LLM agents to improve their plan andwaypoints in-context. For evaluation, we introduce RoCoBench, a 6-task bench-mark covering a wide range of multi-robot collaboration scenarios, accompaniedby a text-only dataset for agent representation and reasoning. We experimentallydemonstrate the effectiveness of our approach – it achieves high success ratesacross all tasks in RoCoBench and adapts to variations in task semantics. Our di-alog setup offers high interpretability and flexibility – in real world experiments,we show RoCo easily incorporates human-in-the-loop, where a user can commu-nicate and collaborate with a robot agent to complete tasks together. See projectwebsite project-roco.github.io for videos and code. Columbia University
24 22 ./images/roco_dialectic_multi-robot_collaboration_20230710.png Scaling Large-Language-Model-based Multi-Agent Collaboration Chen Qian, Zihao Xie, Yifei Wang, Wei Liu, Yufan Dang, Zhuoyun Du, Weize Chen, Cheng Yang, Zhiyuan Liu, Maosong Sun Pioneering advancements in large languagemodel-powered agents have underscored thedesign pattern of multi-agent collaboration,demonstrating that collective intelligence cansurpass the capabilities of each individual. In-spired by the neural scaling law, which positsthat increasing neurons leads to emergent abil-ities, this study investigates whether a simi-lar principle applies to increasing agents inmulti-agent collaboration.Technically, wepropose ::multi-agent:collaboration::networks(MACNET), which utilize directed acyclicgraphs to organize agents and streamline theirinteractive reasoning via topological ordering,with solutions derived from their dialogues.Extensive experiments show that MACNETconsistently outperforms baseline models, en-abling effective agent collaboration across var-ious network topologies and supporting coop-eration among more than a thousand agents.Notably, we observed a small-world collabo-ration phenomenon, where topologies resem-bling small-world properties achieved supe-rior performance. Additionally, we identifieda collaborative scaling law, indicating thatnormalized solution quality follows a logisticgrowth pattern as scaling agents, with collabo-rative emergence occurring much earlier thanpreviously observed instances of neural emer-gence. The code and data will be available athttps://github.com/OpenBMB/ChatDev. Tsinghua University, Beijing University of Posts and Telecommunications
25 23 ./images/scaling_large-language-model-based_multi-agent_collaboration_20240611.png The Impact of Language on Arithmetic Proficiency- A Multilingual Investigation with Cross-Agent Checking Computation Chung-Chi Chen, Hiroya Takamura, Ichiro Kobayashi, Yusuke Miyao This paper critically examines the arithmetic capabilities of Large Language Models (LLMs), uncovering significant limitations in their performance. Our research reveals a notable decline in accuracy for complex calculations involving large numbers, with addition and subtraction tasks showing varying degrees of proficiency. Additionally, we challenge the notion that arithmetic is language-independent, finding up to a 10% difference in performance across twenty languages. The study also compares self-verification methods with cross-agent collaborations, showing that a single model often outperforms collaborative approaches in basic arithmetic tasks. These findings suggest a need to reassess the effectiveness of LLMs in tasks requiring numerical accuracy and precision. AIST, University of Tokyo
26 24 ./images/the_impact_of_language_20240616.png Theory of Mind for Multi-Agent Collaboration via Large Language Models Huao Li, Yu Quan Chong, Simon Stepputtis, Joseph Campbell, Dana Hughes, Michael Lewis, Katia Sycara While Large Language Models (LLMs) havedemonstrated impressive accomplishments inboth reasoning and planning, their abilitiesin multi-agent collaborations remains largelyunexplored.This study evaluates LLM-based agents in a multi-agent cooperative textgame with Theory of Mind (ToM) inferencetasks, comparing their performance with Multi-Agent Reinforcement Learning (MARL) andplanning-based baselines. We observed evi-dence of emergent collaborative behaviors andhigh-order Theory of Mind capabilities amongLLM-based agents. Our results reveal limi-tations in LLM-based agents’ planning opti-mization due to systematic failures in managinglong-horizon contexts and hallucination aboutthe task state. We explore the use of explicitbelief state representations to mitigate these is-sues, finding that it enhances task performanceand the accuracy of ToM inferences for LLM-based agents. University of Pittsburgh, Carnegie Mellon University
27 25 ./images/theory_of_mind_for_20231016.png Toward Optimal LLM Alignments Using Two-Player Games Rui Zheng, Hongyi Guo, Zhihan Liu, Xiaoying Zhang, Yuanshun Yao, Xiaojun Xu, Zhaoran Wang, Zhiheng Xi, Tao Gui, Qi Zhang, Xuanjing Huang, Hang Li, Yang Liu Alignment of large language models is a critical process designed to ensure thatthe model’s responses to user prompts accurately reflect human intentions andadhere to societal values. The standard Reinforcement Learning from HumanFeedback (RLHF) framework primarily focuses on optimizing the performance oflarge language models using pre-collected prompts. However, collecting promptsthat provide comprehensive coverage is both tedious and challenging, and oftenfails to include scenarios that LLMs need to improve on the most. In this paper,we investigate alignment through the lens of two-agent games, involving iterativeinteractions between an adversarial and a defensive agent. The adversarial agent’stask at each step is to generate prompts that expose the weakness of the defensiveagent. In return, the defensive agent seeks to improve its responses to these newlyidentified prompts it “struggled" with, based on feedback from the reward model.We theoretically demonstrate that this iterative reinforcement learning optimizationconverges to a Nash Equilibrium for the game induced by the agents. Experi-mental results in safety scenarios demonstrate that learning in such a competitiveenvironment not only fully trains agents but also leads to policies with enhancedgeneralization capabilities for both adversarial and defensive agents. Our code isreleased at https://github.com/ruizheng20/gpo. Fudan University, Northwestern University, ByteDance Research
28 26 ./images/toward_optimal_llm_alignments_20240616.png Towards Detecting LLMs Hallucination via Markov Chain-based Multi-agent Debate Framework Xiaoxi Sun, Jinpeng Li, Yan Zhong, Dongyan Zhao, Rui Yan The advent of large language models (LLMs)has facilitated the development of natural lan-guage text generation. It also poses unprece-dented challenges, with content hallucinationemerging as a significant concern. Existingsolutions often involve expensive and complexinterventions during the training process. More-over, some approaches emphasize problem dis-assembly while neglecting the crucial valida-tion process, leading to performance degrada-tion or limited applications. To overcome theselimitations, we propose a Markov Chain-basedmulti-agent debate verification framework toenhance hallucination detection accuracy inconcise claims. Our method integrates the fact-checking process, including claim detection,evidence retrieval, and multi-agent verification.In the verification stage, we deploy multipleagents through flexible Markov Chain-baseddebates to validate individual claims, ensuringmeticulous verification outcomes. Experimen-tal results across three generative tasks demon-strate that our approach achieves significantimprovements over baselines. Peking University, Renmin University of China
29 27 ./images/towards_detecting_llms_hallucination_20240605.png To be Continued... Your Contributions are Welcome!

94
MultiAgentEbook/book_communication/script.js Executable file
View File

@ -0,0 +1,94 @@
document.addEventListener("DOMContentLoaded", function() {
const csvFilePath = './book_communication/data.csv';
function loadCSV(filePath) {
return fetch(filePath)
.then(response => response.text())
.then(text => Papa.parse(text, { header: true }).data);
}
function createFlipBook(pages) {
const container = document.getElementById('flip_book_container');
const numPages = pages.length;
let flipBookHTML = '';
let style = document.createElement('style');
let css = '';
flipBookHTML += `<input type="checkbox" id="cover_checkbox">\n`;
for (let i = 0; i < numPages - 1; i++) {
flipBookHTML += `<input type="checkbox" id="page${i + 1}_checkbox">\n`;
}
flipBookHTML += `<div id="flip_book">\n`;
flipBookHTML += `<div class="front_cover">
<label for="cover_checkbox" id="cover">
<img src="./images/1.png" alt="Book Cover" class="cover_image">
</label>
</div>`
for (let i = 0; i < numPages - 1; i++) {
console.log(i)
const page = pages[i];
const pageIndex = i + 1;
flipBookHTML += `
<div class="page" id="page${pageIndex}">
<div class="front_page">
<label for="page${pageIndex}_checkbox"></label>
<img class="back_content" src="${page.image_path}" alt="Back content">
</div>
<div class="back_page">
<label for="page${pageIndex}_checkbox"></label>
<img class="edge_shading" src="./images/back_page_edge_shading.png" alt="Back page edge shading">
<div class="text_content">
<h1>${page.title}</h1>
<p class="author">${page.author}</p>
<p class="author">${page.affiliation}</p>
<div class="text_content_summary"><p class="summary">${page.summary}</p></div>
</div>
</div>
</div>\n`;
css += `
#page${pageIndex} {
z-index: ${numPages - i};
}
#page${pageIndex}_checkbox:checked~#flip_book #page${pageIndex} {
transform: rotateY(-180deg);
z-index: ${i + 1};
}\n`;
}
flipBookHTML += `<div class="back_cover">
<img src="./images/1a.png" alt="Back Cover" class="cover_image">
</div>`;
container.innerHTML = flipBookHTML;
style.innerHTML = css;
document.head.appendChild(style);
const md = window.markdownit();
const summaryElements = document.querySelectorAll('.summary');
summaryElements.forEach(el => {
el.innerHTML = md.render(el.textContent);
});
}
loadCSV(csvFilePath).then(pages => {
createFlipBook(pages);
});
});

23
MultiAgentEbook/book_communication_index.html Executable file
View File

@ -0,0 +1,23 @@
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Flip Book</title>
<link rel="stylesheet" href="./book_style.css">
<style>
body {
background-color: transparent;
}
</style>
</head>
<body>
<div id="flip_book_container"></div>
<script src="https://cdnjs.cloudflare.com/ajax/libs/PapaParse/5.3.0/papaparse.min.js"></script>
<script src="https://cdn.jsdelivr.net/npm/markdown-it/dist/markdown-it.min.js"></script>
<script src="./book_communication/script.js"></script>
</body>
</html>

11
MultiAgentEbook/book_evolution/data.csv Executable file
View File

@ -0,0 +1,11 @@
,image_path,title,author,summary,affiliation
0,./images/3d.png,360°REA: Towards A Reusable Experience Accumulation with 360° Assessment for Multi-Agent System,"Shen Gao, Hao Li, Zhengliang Shi, Chengrui Huang, Quan Tu, Zhiliang Tian, Minlie Huang, Shuo Shang","Largelanguagemodelagentshavedemonstratedremarkableadvancementsacross various complex tasks. Recent worksfocus on optimizing the agent team oremploying self-reflection to iteratively solvecomplex tasks.Since these agents are allbased on the same LLM, only conductingself-evaluation or removing underperformingagents does not substantively enhance thecapability of the agents.We argue that acomprehensive evaluation and accumulatingexperience from evaluation feedback is aneffectiveapproachtoimprovingsystemperformance.In this paper, we proposeReusableExperienceAccumulationwith360◦ Assessment (360◦REA), a hierarchicalmulti-agent framework inspired by corporateorganizational practices.The frameworkemploys a novel 360◦ performance assessmentmethod for multi-perspective performanceevaluation with fine-grained assessment. Toenhance the capability of agents in addressingcomplextasks,weintroducedual-levelexperience pool for agents to accumulateexperience through fine-grained assessment.Extensiveexperimentsoncomplextaskdatasets demonstrate the effectiveness of360◦REA.","University of Electronic Science and Technology of China, Shandong University, Renmin University of China, National University of Defense Technology, Tsinghua University"
1,./images/360°rea_towards_a_reusable_20240408.png,Affordable Generative Agents,"Yangbin Yu, Qin Zhang, Junyou Li, Qiang Fu, Deheng Ye","The emergence of large language models (LLMs)has significantly advanced the simulation ofbelievable interactive agents.However, thesubstantial cost on maintaining the prolongedagent interactions poses challenge over thedeployment of believable LLM-based agents.Therefore, in this paper, we develop AffordableGenerative Agents (AGA), a framework forenabling the generation of believable andlow-cost interactions on both agent-environmentand inter-agents levels. Specifically, for agent-environment interactions, we substitute repetitiveLLM inferences with learned policies; while forinter-agent interactions, we model the social rela-tionships between agents and compress auxiliarydialogue information. Extensive experiments onmultiple environments show the effectivenessand efficiency of our proposed framework. Also,we delve into the mechanisms of emergentbelievable behaviors lying in LLM agents,demonstrating that agents can only generatefinite behaviors in fixed environments, basedupon which, we understand ways to facilitateemergent interaction behaviors.Our code ispublicly available at:https://github.com/AffordableGenerativeAgents/Affordable-Generative-Agents.",Tencent Inc.
2,./images/affordable_generative_agents_20240203.png,Agent Hospital: A Simulacrum of Hospital with Evolvable Medical Agents,"Junkai Li, Siyu Wang, Meng Zhang, Weitao Li, Yunghwei Lai, Xinhui Kang, Weizhi Ma, Yang Liu","In this paper, we introduce a simulacrum of hospital called Agent Hospital that simulates theentire process of treating illness. All patients, nurses, and doctors are autonomous agents powered bylarge language models (LLMs). Our central goal is to enable a doctor agent to learn how to treat illnesswithin the simulacrum. To do so, we propose a method called MedAgent-Zero. As the simulacrum cansimulate disease onset and progression based on knowledge bases and LLMs, doctor agents can keepaccumulating experience from both successful and unsuccessful cases. Simulation experiments show thatthe treatment performance of doctor agents consistently improves on various tasks. More interestingly,the knowledge the doctor agents have acquired in Agent Hospital is applicable to real-world medicarebenchmarks. After treating around ten thousand patients (real-world doctors may take over two years),the evolved doctor agent achieves a state-of-the-art accuracy of 9",Tsinghua University
3,./images/agent_hospital_a_simulacrum_20240505.png,Beyond Natural Language: LLMs Leveraging Alternative Formats for Enhanced Reasoning and Communication,"Weize Chen, Chenfei Yuan, Jiarui Yuan, Yusheng Su, Chen Qian, Cheng Yang, Ruobing Xie, Zhiyuan Liu, Maosong Sun","Natural language (NL) has long been the predominant format for human cognition and communication, and by extension, has been similarly pivotal in the development and application of Large Language Models (LLMs). Yet, besides NL, LLMs have seen various non-NL formats during pre-training, such as code and logical expression. NL's status as the optimal format for LLMs, particularly in single-LLM reasoning and multi-agent communication, has not been thoroughly examined. In this work, we challenge the default use of NL by exploring the utility of non-NL formats in these contexts. We show that allowing LLMs to autonomously select the most suitable format before reasoning or communicating leads to a 3.3 to 5.7\% improvement in reasoning efficiency for different LLMs, and up to a 72.7\% reduction in token usage in multi-agent communication, all while maintaining communicative effectiveness. Our comprehensive analysis further reveals that LLMs can devise a format from limited task instructions and that the devised format is effectively transferable across different LLMs. Intriguingly, the structured communication format decided by LLMs exhibits notable parallels with established agent communication languages, suggesting a natural evolution towards efficient, structured communication in agent communication.","Tsinghua University, Tencent, Beijing University of Posts and Telecommunications"
4,./images/beyond_natural_language_llms_20240228.png,Dynamic LLM-Agent Network: An LLM-agent Collaboration Framework with Agent Team Optimization,"Zijun Liu, Yanzhe Zhang, Peng Li, Yang Liu, Diyi Yang","Large language model (LLM) agents have been shown effective on a wide rangeof tasks, and by ensembling multiple LLM agents, their performances could befurther improved. Existing approaches employ a fixed set of agents to interactwith each other in a static architecture, which limits their generalizability to vari-ous tasks and requires strong human prior in designing these agents. In this work,we propose to construct a strategic team of agents communicating in a dynamicinteraction architecture based on the task query. Specifically, we build a frame-work named Dynamic LLM-Agent Network (DyLAN) for LLM-agent collabora-tion on complicated tasks like reasoning and code generation. DyLAN enablesagents to interact for multiple rounds in a dynamic architecture with inference-time agent selection and an early-stopping mechanism to improve performanceand efficiency. We further design an automatic agent team optimization algorithmbased on an unsupervised metric termed Agent Importance Score, enabling theselection of best agents based on the contribution each agent makes. Empirically,we demonstrate that DyLAN performs well in both reasoning and code generationtasks with reasonable computational cost. DyLAN achieves 1","Tsinghua University, Georgia Tech, Stanford University"
5,./images/dynamic_llm-agent_network_an_20231003.png,Experiential Co-Learning of Software-Developing Agents,"Chen Qian, Yufan Dang, Jiahao Li, Wei Liu, Zihao Xie, Yifei Wang, Weize Chen, Cheng Yang, Xin Cong, Xiaoyin Che, Zhiyuan Liu, Maosong Sun","Recent advancements in large language mod-els (LLMs) have brought significant changesto various domains, especially through LLM-driven autonomous agents. A representativescenario is in software development, whereLLM agents demonstrate efficient collabora-tion, task division, and assurance of softwarequality, markedly reducing the need for man-ual involvement. However, these agents fre-quently perform a variety of tasks indepen-dently, without benefiting from past experi-ences, which leads to repeated mistakes andinefficient attempts in multi-step task execu-tion. To this end, we introduce Experiential Co-Learning, a novel LLM-agent learning frame-work in which instructor and assistant agentsgather shortcut-oriented experiences from theirhistorical trajectories and use these past expe-riences for future task execution. The exten-sive experiments demonstrate that the frame-work enables agents to tackle unseen software-developing tasks more effectively. We antici-pate that our insights will guide LLM agentstowards enhanced autonomy and contributeto their evolutionary growth in cooperativelearning. The code and data are available athttps://github.com/OpenBMB/ChatDev.","Tsinghua University, Dalian University of Technology, Beijing University of Posts and Telecommunications, Siemens"
6,./images/experiential_co-learning_of_software-developing_20231228.png,Iterative Experience Refinement of Software-Developing Agents,"Chen Qian, Jiahao Li, Yufan Dang, Wei Liu, YiFei Wang, Zihao Xie, Weize Chen, Cheng Yang, Yingli Zhang, Zhiyuan Liu, Maosong Sun","Autonomous agents powered by large languagemodels (LLMs) show significant potential forachieving high autonomy in various scenar-ios such as software development. Recent re-search has shown that LLM agents can lever-age past experiences to reduce errors and en-hance efficiency. However, the static experi-ence paradigm, reliant on a fixed collection ofpast experiences acquired heuristically, lacksiterative refinement and thus hampers agents’adaptability. In this paper, we introduce the It-erative Experience Refinement framework, en-abling LLM agents to refine experiences itera-tively during task execution. We propose twofundamental patterns: the successive pattern,refining based on nearest experiences within atask batch, and the cumulative pattern, acquir-ing experiences across all previous task batches.Augmented with our heuristic experience elim-ination, the method prioritizes high-quality andfrequently-used experiences, effectively man-aging the experience space and enhancing effi-ciency. Extensive experiments show that whilethe successive pattern may yield superior re-sults, the cumulative pattern provides more sta-ble performance......","Tsinghua University, Dalian University of Technology, Beijing University of Posts and Telecommunications, Siemens"
7,./images/iterative_experience_refinement_of_20240507.png,Language Agents as Optimizable Graphs,"Mingchen Zhuge, Wenyi Wang, Louis Kirsch, Francesco Faccio, Dmitrii Khizbullin, Jürgen Schmidhuber","Various human-designed prompt engineering techniques have been proposed to improve problem solvers based on Large Language Models (LLMs), yielding many disparate code bases. We unify these approaches by describing LLM-based agents as computational graphs. The nodes implement functions to process multimodal data or query LLMs, and the edges describe the information flow between operations. Graphs can be recursively combined into larger composite graphs representing hierarchies of inter-agent collaboration (where edges connect operations of different agents). Our novel automatic graph optimizers (1) refine node-level LLM prompts (node optimization) and (2) improve agent orchestration by changing graph connectivity (edge optimization). Experiments demonstrate that our framework can be used to efficiently develop, integrate, and automatically improve various LLM agents. ","King Abdullah University of Science and Technology, The Swiss AI Lab IDSIA, USI, SUPSI"
8,./images/language_agents_as_optimizable_20240226.png,Lyfe Agents: Generative agents for low-cost real-time social interactions,"Zhao Kaiya, Michelangelo Naim, Jovana Kondic, Manuel Cortes, Jiaxin Ge, Shuying Luo, Guangyu Robert Yang, Andrew Ahn","Highly autonomous generative agents powered by large language models promise to simulate intricate social behaviors in virtual societies. However, achieving real-time interactions with humans at a low computational cost remains challenging. Here, we introduce Lyfe Agents. They combine low-cost with real-time responsiveness, all while remaining intelligent and goal-oriented. Key innovations include: (1) an option-action framework, reducing the cost of high-level decisions; (2) asynchronous self-monitoring for better self-consistency; and (3) a Summarize-and-Forget memory mechanism, prioritizing critical memory items at a low cost. We evaluate Lyfe Agents' self-motivation and sociability across several multi-agent scenarios in our custom LyfeGame 3D virtual environment platform. When equipped with our brain-inspired techniques, Lyfe Agents can exhibit human-like self-motivated social reasoning. For example, the agents can solve a crime (a murder mystery) through autonomous collaboration and information exchange. Meanwhile, our techniques enabled Lyfe Agents to operate at a computational cost 10-100 times lower than existing alternatives. Our findings underscore the transformative potential of autonomous generative agents to enrich human social experiences in virtual worlds.","Massachusetts Institute of Technology, Peking University, LyfeAL"
9,./images/lyfe_agents_generative_agents_20231003.png,To be Continued...,Your Contributions are Welcome!,,
1 image_path title author summary affiliation
2 0 ./images/3d.png 360°REA: Towards A Reusable Experience Accumulation with 360° Assessment for Multi-Agent System Shen Gao, Hao Li, Zhengliang Shi, Chengrui Huang, Quan Tu, Zhiliang Tian, Minlie Huang, Shuo Shang Largelanguagemodelagentshavedemonstratedremarkableadvancementsacross various complex tasks. Recent worksfocus on optimizing the agent team oremploying self-reflection to iteratively solvecomplex tasks.Since these agents are allbased on the same LLM, only conductingself-evaluation or removing underperformingagents does not substantively enhance thecapability of the agents.We argue that acomprehensive evaluation and accumulatingexperience from evaluation feedback is aneffectiveapproachtoimprovingsystemperformance.In this paper, we proposeReusableExperienceAccumulationwith360◦ Assessment (360◦REA), a hierarchicalmulti-agent framework inspired by corporateorganizational practices.The frameworkemploys a novel 360◦ performance assessmentmethod for multi-perspective performanceevaluation with fine-grained assessment. Toenhance the capability of agents in addressingcomplextasks,weintroducedual-levelexperience pool for agents to accumulateexperience through fine-grained assessment.Extensiveexperimentsoncomplextaskdatasets demonstrate the effectiveness of360◦REA. University of Electronic Science and Technology of China, Shandong University, Renmin University of China, National University of Defense Technology, Tsinghua University
3 1 ./images/360°rea_towards_a_reusable_20240408.png Affordable Generative Agents Yangbin Yu, Qin Zhang, Junyou Li, Qiang Fu, Deheng Ye The emergence of large language models (LLMs)has significantly advanced the simulation ofbelievable interactive agents.However, thesubstantial cost on maintaining the prolongedagent interactions poses challenge over thedeployment of believable LLM-based agents.Therefore, in this paper, we develop AffordableGenerative Agents (AGA), a framework forenabling the generation of believable andlow-cost interactions on both agent-environmentand inter-agents levels. Specifically, for agent-environment interactions, we substitute repetitiveLLM inferences with learned policies; while forinter-agent interactions, we model the social rela-tionships between agents and compress auxiliarydialogue information. Extensive experiments onmultiple environments show the effectivenessand efficiency of our proposed framework. Also,we delve into the mechanisms of emergentbelievable behaviors lying in LLM agents,demonstrating that agents can only generatefinite behaviors in fixed environments, basedupon which, we understand ways to facilitateemergent interaction behaviors.Our code ispublicly available at:https://github.com/AffordableGenerativeAgents/Affordable-Generative-Agents. Tencent Inc.
4 2 ./images/affordable_generative_agents_20240203.png Agent Hospital: A Simulacrum of Hospital with Evolvable Medical Agents Junkai Li, Siyu Wang, Meng Zhang, Weitao Li, Yunghwei Lai, Xinhui Kang, Weizhi Ma, Yang Liu In this paper, we introduce a simulacrum of hospital called Agent Hospital that simulates theentire process of treating illness. All patients, nurses, and doctors are autonomous agents powered bylarge language models (LLMs). Our central goal is to enable a doctor agent to learn how to treat illnesswithin the simulacrum. To do so, we propose a method called MedAgent-Zero. As the simulacrum cansimulate disease onset and progression based on knowledge bases and LLMs, doctor agents can keepaccumulating experience from both successful and unsuccessful cases. Simulation experiments show thatthe treatment performance of doctor agents consistently improves on various tasks. More interestingly,the knowledge the doctor agents have acquired in Agent Hospital is applicable to real-world medicarebenchmarks. After treating around ten thousand patients (real-world doctors may take over two years),the evolved doctor agent achieves a state-of-the-art accuracy of 9 Tsinghua University
5 3 ./images/agent_hospital_a_simulacrum_20240505.png Beyond Natural Language: LLMs Leveraging Alternative Formats for Enhanced Reasoning and Communication Weize Chen, Chenfei Yuan, Jiarui Yuan, Yusheng Su, Chen Qian, Cheng Yang, Ruobing Xie, Zhiyuan Liu, Maosong Sun Natural language (NL) has long been the predominant format for human cognition and communication, and by extension, has been similarly pivotal in the development and application of Large Language Models (LLMs). Yet, besides NL, LLMs have seen various non-NL formats during pre-training, such as code and logical expression. NL's status as the optimal format for LLMs, particularly in single-LLM reasoning and multi-agent communication, has not been thoroughly examined. In this work, we challenge the default use of NL by exploring the utility of non-NL formats in these contexts. We show that allowing LLMs to autonomously select the most suitable format before reasoning or communicating leads to a 3.3 to 5.7\% improvement in reasoning efficiency for different LLMs, and up to a 72.7\% reduction in token usage in multi-agent communication, all while maintaining communicative effectiveness. Our comprehensive analysis further reveals that LLMs can devise a format from limited task instructions and that the devised format is effectively transferable across different LLMs. Intriguingly, the structured communication format decided by LLMs exhibits notable parallels with established agent communication languages, suggesting a natural evolution towards efficient, structured communication in agent communication. Tsinghua University, Tencent, Beijing University of Posts and Telecommunications
6 4 ./images/beyond_natural_language_llms_20240228.png Dynamic LLM-Agent Network: An LLM-agent Collaboration Framework with Agent Team Optimization Zijun Liu, Yanzhe Zhang, Peng Li, Yang Liu, Diyi Yang Large language model (LLM) agents have been shown effective on a wide rangeof tasks, and by ensembling multiple LLM agents, their performances could befurther improved. Existing approaches employ a fixed set of agents to interactwith each other in a static architecture, which limits their generalizability to vari-ous tasks and requires strong human prior in designing these agents. In this work,we propose to construct a strategic team of agents communicating in a dynamicinteraction architecture based on the task query. Specifically, we build a frame-work named Dynamic LLM-Agent Network (DyLAN) for LLM-agent collabora-tion on complicated tasks like reasoning and code generation. DyLAN enablesagents to interact for multiple rounds in a dynamic architecture with inference-time agent selection and an early-stopping mechanism to improve performanceand efficiency. We further design an automatic agent team optimization algorithmbased on an unsupervised metric termed Agent Importance Score, enabling theselection of best agents based on the contribution each agent makes. Empirically,we demonstrate that DyLAN performs well in both reasoning and code generationtasks with reasonable computational cost. DyLAN achieves 1 Tsinghua University, Georgia Tech, Stanford University
7 5 ./images/dynamic_llm-agent_network_an_20231003.png Experiential Co-Learning of Software-Developing Agents Chen Qian, Yufan Dang, Jiahao Li, Wei Liu, Zihao Xie, Yifei Wang, Weize Chen, Cheng Yang, Xin Cong, Xiaoyin Che, Zhiyuan Liu, Maosong Sun Recent advancements in large language mod-els (LLMs) have brought significant changesto various domains, especially through LLM-driven autonomous agents. A representativescenario is in software development, whereLLM agents demonstrate efficient collabora-tion, task division, and assurance of softwarequality, markedly reducing the need for man-ual involvement. However, these agents fre-quently perform a variety of tasks indepen-dently, without benefiting from past experi-ences, which leads to repeated mistakes andinefficient attempts in multi-step task execu-tion. To this end, we introduce Experiential Co-Learning, a novel LLM-agent learning frame-work in which instructor and assistant agentsgather shortcut-oriented experiences from theirhistorical trajectories and use these past expe-riences for future task execution. The exten-sive experiments demonstrate that the frame-work enables agents to tackle unseen software-developing tasks more effectively. We antici-pate that our insights will guide LLM agentstowards enhanced autonomy and contributeto their evolutionary growth in cooperativelearning. The code and data are available athttps://github.com/OpenBMB/ChatDev. Tsinghua University, Dalian University of Technology, Beijing University of Posts and Telecommunications, Siemens
8 6 ./images/experiential_co-learning_of_software-developing_20231228.png Iterative Experience Refinement of Software-Developing Agents Chen Qian, Jiahao Li, Yufan Dang, Wei Liu, YiFei Wang, Zihao Xie, Weize Chen, Cheng Yang, Yingli Zhang, Zhiyuan Liu, Maosong Sun Autonomous agents powered by large languagemodels (LLMs) show significant potential forachieving high autonomy in various scenar-ios such as software development. Recent re-search has shown that LLM agents can lever-age past experiences to reduce errors and en-hance efficiency. However, the static experi-ence paradigm, reliant on a fixed collection ofpast experiences acquired heuristically, lacksiterative refinement and thus hampers agents’adaptability. In this paper, we introduce the It-erative Experience Refinement framework, en-abling LLM agents to refine experiences itera-tively during task execution. We propose twofundamental patterns: the successive pattern,refining based on nearest experiences within atask batch, and the cumulative pattern, acquir-ing experiences across all previous task batches.Augmented with our heuristic experience elim-ination, the method prioritizes high-quality andfrequently-used experiences, effectively man-aging the experience space and enhancing effi-ciency. Extensive experiments show that whilethe successive pattern may yield superior re-sults, the cumulative pattern provides more sta-ble performance...... Tsinghua University, Dalian University of Technology, Beijing University of Posts and Telecommunications, Siemens
9 7 ./images/iterative_experience_refinement_of_20240507.png Language Agents as Optimizable Graphs Mingchen Zhuge, Wenyi Wang, Louis Kirsch, Francesco Faccio, Dmitrii Khizbullin, Jürgen Schmidhuber Various human-designed prompt engineering techniques have been proposed to improve problem solvers based on Large Language Models (LLMs), yielding many disparate code bases. We unify these approaches by describing LLM-based agents as computational graphs. The nodes implement functions to process multimodal data or query LLMs, and the edges describe the information flow between operations. Graphs can be recursively combined into larger composite graphs representing hierarchies of inter-agent collaboration (where edges connect operations of different agents). Our novel automatic graph optimizers (1) refine node-level LLM prompts (node optimization) and (2) improve agent orchestration by changing graph connectivity (edge optimization). Experiments demonstrate that our framework can be used to efficiently develop, integrate, and automatically improve various LLM agents. King Abdullah University of Science and Technology, The Swiss AI Lab IDSIA, USI, SUPSI
10 8 ./images/language_agents_as_optimizable_20240226.png Lyfe Agents: Generative agents for low-cost real-time social interactions Zhao Kaiya, Michelangelo Naim, Jovana Kondic, Manuel Cortes, Jiaxin Ge, Shuying Luo, Guangyu Robert Yang, Andrew Ahn Highly autonomous generative agents powered by large language models promise to simulate intricate social behaviors in virtual societies. However, achieving real-time interactions with humans at a low computational cost remains challenging. Here, we introduce Lyfe Agents. They combine low-cost with real-time responsiveness, all while remaining intelligent and goal-oriented. Key innovations include: (1) an option-action framework, reducing the cost of high-level decisions; (2) asynchronous self-monitoring for better self-consistency; and (3) a Summarize-and-Forget memory mechanism, prioritizing critical memory items at a low cost. We evaluate Lyfe Agents' self-motivation and sociability across several multi-agent scenarios in our custom LyfeGame 3D virtual environment platform. When equipped with our brain-inspired techniques, Lyfe Agents can exhibit human-like self-motivated social reasoning. For example, the agents can solve a crime (a murder mystery) through autonomous collaboration and information exchange. Meanwhile, our techniques enabled Lyfe Agents to operate at a computational cost 10-100 times lower than existing alternatives. Our findings underscore the transformative potential of autonomous generative agents to enrich human social experiences in virtual worlds. Massachusetts Institute of Technology, Peking University, LyfeAL
11 9 ./images/lyfe_agents_generative_agents_20231003.png To be Continued... Your Contributions are Welcome!

94
MultiAgentEbook/book_evolution/script.js Executable file
View File

@ -0,0 +1,94 @@
document.addEventListener("DOMContentLoaded", function() {
const csvFilePath = './book_evolution/data.csv';
function loadCSV(filePath) {
return fetch(filePath)
.then(response => response.text())
.then(text => Papa.parse(text, { header: true }).data);
}
function createFlipBook(pages) {
const container = document.getElementById('flip_book_container');
const numPages = pages.length;
let flipBookHTML = '';
let style = document.createElement('style');
let css = '';
flipBookHTML += `<input type="checkbox" id="cover_checkbox">\n`;
for (let i = 0; i < numPages - 1; i++) {
flipBookHTML += `<input type="checkbox" id="page${i + 1}_checkbox">\n`;
}
flipBookHTML += `<div id="flip_book">\n`;
flipBookHTML += `<div class="front_cover">
<label for="cover_checkbox" id="cover">
<img src="./images/3.png" alt="Book Cover" class="cover_image">
</label>
</div>`
for (let i = 0; i < numPages - 1; i++) {
console.log(i)
const page = pages[i];
const pageIndex = i + 1;
flipBookHTML += `
<div class="page" id="page${pageIndex}">
<div class="front_page">
<label for="page${pageIndex}_checkbox"></label>
<img class="back_content" src="${page.image_path}" alt="Back content">
</div>
<div class="back_page">
<label for="page${pageIndex}_checkbox"></label>
<img class="edge_shading" src="./images/back_page_edge_shading.png" alt="Back page edge shading">
<div class="text_content">
<h1>${page.title}</h1>
<p class="author">${page.author}</p>
<p class="author">${page.affiliation}</p>
<div class="text_content_summary"><p class="summary">${page.summary}</p></div>
</div>
</div>
</div>\n`;
css += `
#page${pageIndex} {
z-index: ${numPages - i};
}
#page${pageIndex}_checkbox:checked~#flip_book #page${pageIndex} {
transform: rotateY(-180deg);
z-index: ${i + 1};
}\n`;
}
flipBookHTML += `<div class="back_cover">
<img src="./images/3a.png" alt="Back Cover" class="cover_image">
</div>`;
container.innerHTML = flipBookHTML;
style.innerHTML = css;
document.head.appendChild(style);
const md = window.markdownit();
const summaryElements = document.querySelectorAll('.summary');
summaryElements.forEach(el => {
el.innerHTML = md.render(el.textContent);
});
}
loadCSV(csvFilePath).then(pages => {
createFlipBook(pages);
});
});

23
MultiAgentEbook/book_evolution_index.html Executable file
View File

@ -0,0 +1,23 @@
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Flip Book</title>
<link rel="stylesheet" href="./book_style.css">
<style>
body {
background-color: transparent;
}
</style>
</head>
<body>
<div id="flip_book_container"></div>
<script src="https://cdnjs.cloudflare.com/ajax/libs/PapaParse/5.3.0/papaparse.min.js"></script>
<script src="https://cdn.jsdelivr.net/npm/markdown-it/dist/markdown-it.min.js"></script>
<script src="./book_evolution/script.js"></script>
</body>
</html>

42
MultiAgentEbook/book_organization/data.csv Executable file
View File

@ -0,0 +1,42 @@
,image_path,title,author,summary,affiliation
0,./images/2d.png,(Perhaps) Beyond Human Translation: Harnessing Multi-Agent Collaboration for Translating Ultra-Long Literary Texts,"Minghao Wu, Yulin Yuan, Gholamreza Haffari, Longyue Wang","Recent advancements in machine translation (MT) have significantly enhancedtranslation quality across various domains. However, the translation of literarytexts remains a formidable challenge due to their complex language, figurative ex-pressions, and cultural nuances. In this work, we introduce a novel multi-agentframework based on large language models (LLMs) for literary translation, im-plemented as a company called TRANSAGENTS, which mirrors traditional trans-lation publication process by leveraging the collective capabilities of multipleagents, to address the intricate demands of translating literary works. To evaluatethe effectiveness of our system, we propose two innovative evaluation strategies:Monolingual Human Preference (MHP) and Bilingual LLM Preference (BLP).MHP assesses translations from the perspective of monolingual readers of the tar-get language, while BLP uses advanced LLMs to compare translations directlywith the original texts. Empirical findings indicate that despite lower d-BLEUscores, translations from TRANSAGENTS are preferred by both human evalua-tors and LLMs over human-written references, particularly in genres requiringdomain-specific knowledge. We also highlight the strengths and limitations ofTRANSAGENTS through case studies and suggests directions for future research.","Monash University, University of Macau, Tencent AI Lab"
1,./images/(perhaps)_beyond_human_translation_20240520.png,Agent Hospital: A Simulacrum of Hospital with Evolvable Medical Agents,"Junkai Li, Siyu Wang, Meng Zhang, Weitao Li, Yunghwei Lai, Xinhui Kang, Weizhi Ma, Yang Liu","In this paper, we introduce a simulacrum of hospital called Agent Hospital that simulates theentire process of treating illness. All patients, nurses, and doctors are autonomous agents powered bylarge language models (LLMs). Our central goal is to enable a doctor agent to learn how to treat illnesswithin the simulacrum. To do so, we propose a method called MedAgent-Zero. As the simulacrum cansimulate disease onset and progression based on knowledge bases and LLMs, doctor agents can keepaccumulating experience from both successful and unsuccessful cases. Simulation experiments show thatthe treatment performance of doctor agents consistently improves on various tasks. More interestingly,the knowledge the doctor agents have acquired in Agent Hospital is applicable to real-world medicarebenchmarks. After treating around ten thousand patients (real-world doctors may take over two years),the evolved doctor agent achieves a state-of-the-art accuracy of 9",Tsinghua University
2,./images/agent_hospital_a_simulacrum_20240505.png,AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation,"Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li, Erkang Zhu, Li Jiang, Xiaoyun Zhang, Shaokun Zhang, Jiale Liu, Ahmed Hassan Awadallah, Ryen W White, Doug Burger, Chi Wang","AutoGen2 is an open-source framework that allows developers to build LLM ap-plications via multiple agents that can converse with each other to accomplishtasks. AutoGen agents are customizable, conversable, and can operate in vari-ous modes that employ combinations of LLMs, human inputs, and tools. UsingAutoGen, developers can also flexibly define agent interaction behaviors. Bothnatural language and computer code can be used to program flexible conversationpatterns for different applications. AutoGen serves as a generic framework forbuilding diverse applications of various complexities and LLM capacities. Em-pirical studies demonstrate the effectiveness of the framework in many exampleapplications, with domains ranging from mathematics, coding, question answer-ing, operations research, online decision-making, entertainment, etc.","Microsoft Research, Pennsylvania State University, University of Washington, Xidian University"
3,./images/autogen_enabling_next-gen_llm_20230816.png,Avalon's Game of Thoughts: Battle Against Deception through Recursive Contemplation,"Shenzhi Wang, Chang Liu, Zilong Zheng, Siyuan Qi, Shuo Chen, Qisen Yang, Andrew Zhao, Chaofei Wang, Shiji Song, Gao Huang","Recent breakthroughs in large language models (LLMs) have brought remark-able success in the field of LLM-as-Agent. Nevertheless, a prevalent assumptionis that the information processed by LLMs is consistently honest, neglecting thepervasive deceptive or misleading information in human society and AI-generatedcontent.This oversight makes LLMs susceptible to malicious manipulations,potentially resulting in detrimental outcomes. This study utilizes the intricateAvalon game as a testbed to explore LLMs’ potential in deceptive environments.Avalon, full of misinformation and requiring sophisticated logic, manifests as a“Game-of-Thoughts”. Inspired by the efficacy of humans’ recursive thinking andperspective-taking in the Avalon game, we introduce a novel framework, Recur-sive Contemplation (ReCon), to enhance LLMs’ ability to identify and counteractdeceptive information. ReCon combines formulation and refinement contempla-tion processes; formulation contemplation produces initial thoughts and speech,while refinement contemplation further polishes them. Additionally, we incor-porate first-order and second-order perspective transitions into these processesrespectively. Specifically, the first-order allows an LLM agent to infer others’mental states, and the second-order involves understanding how others perceivethe agent’s mental state.......","Tsinghua University, BIGAI, Technical University of Munich"
4,./images/avalon's_game_of_thoughts_20231002.png,Chain of Agents: Large Language Models Collaborating on Long-Context Tasks,"Yusen Zhang, Ruoxi Sun, Yanfei Chen, Tomas Pfister, Rui Zhang, Sercan Ö. Arik","Addressing the challenge of effectively processing long contexts has become a critical issue for Large Language Models (LLMs). Two common strategies have emerged: 1) reducing the input length, such as retrieving relevant chunks by Retrieval-Augmented Generation (RAG), and 2) expanding the context window limit of LLMs. However, both strategies have drawbacks: input reduction has no guarantee of covering the part with needed information, while window extension struggles with focusing on the pertinent information for solving the task. To mitigate these limitations, we propose Chain-of-Agents (CoA), a novel framework that harnesses multi-agent collaboration through natural language to enable information aggregation and context reasoning across various LLMs over long-context tasks. CoA consists of multiple worker agents who sequentially communicate to handle different segmented portions of the text, followed by a manager agent who synthesizes these contributions into a coherent final output. CoA processes the entire input by interleaving reading and reasoning, and it mitigates long context focus issues by assigning each agent a short context. We perform comprehensive evaluation of CoA on a wide range of long-context tasks in question answering, summarization, and code completion, demonstrating significant improvements by up to 10% over strong baselines of RAG, Full-Context, and multi-agent LLMs.","Penn State University, Google Cloud AI Research"
5,./images/chain_of_agents_large_20240604.png,ChatCoder: Chat-based Refine Requirement Improves LLMs' Code Generation,"Zejun Wang, Jia Li, Ge Li, Zhi Jin","Large language models have shown good performances in generat-ing code to meet human requirements. However, human require-ments expressed in natural languages can be vague, incomplete,and ambiguous, leading large language models to misunderstandhuman requirements and make mistakes. Worse, it is difficult for ahuman user to refine the requirement. To help human users refinetheir requirements and improve large language models’ code gen-eration performances, we propose ChatCoder: a method to refinethe requirements via chatting with large language models. We de-sign a chat scheme in which the large language models will guidethe human users to refine their expression of requirements to bemore precise, unambiguous, and complete than before. Experimentsshow that ChatCoder has improved existing large language models’performance by a large margin. Besides, ChatCoder has the advan-tage over refine-based methods and LLMs fine-tuned via humanresponse.",Peking University
6,./images/chatcoder_chat-based_refine_requirement_20231101.png,ChatDev: Communicative Agents for Software Development,"Chen Qian, Wei Liu, Hongzhang Liu, Nuo Chen, Yufan Dang, Jiahao Li, Cheng Yang, Weize Chen, Yusheng Su, Xin Cong, Juyuan Xu, Dahai Li, Zhiyuan Liu, Maosong Sun","Software development is a complex task thatnecessitates cooperation among multiple mem-bers with diverse skills. Numerous studies useddeep learning to improve specific phases in awaterfall model, such as design, coding, andtesting.However, the deep learning modelin each phase requires unique designs, lead-ing to technical inconsistencies across variousphases, which results in a fragmented and in-effective development process. In this paper,we introduce ChatDev, a chat-powered soft-ware development framework in which special-ized agents driven by large language models(LLMs) are guided in what to communicate(via chat chain) and how to communicate (viacommunicative dehallucination). These agentsactively contribute to the design, coding, andtesting phases through unified language-basedcommunication, with solutions derived fromtheir multi-turn dialogues. We found their uti-lization of natural language is advantageousfor system design, and communicating in pro-gramming language proves helpful in debug-ging. This paradigm demonstrates how linguis-tic communication facilitates multi-agent col-laboration, establishing language as a unify-ing bridge for autonomous task-solving amongLLM agents. The code and data are availableat https://github.com/OpenBMB/ChatDev.","Tsinghua University, The University of Sydney, BUPT, Modelbest Inc."
7,./images/chatdev_communicative_agents_for_20230716.png,ChatEval: Towards Better LLM-based Evaluators through Multi-Agent Debate,"Chi-Min Chan, Weize Chen, Yusheng Su, Jianxuan Yu, Wei Xue, Shanghang Zhang, Jie Fu, Zhiyuan Liu","Text evaluation has historically posed significant challenges, often demandingsubstantial labor and time cost. With the emergence of large language models(LLMs), researchers have explored LLMs’ potential as alternatives for humanevaluation. While these single-agent-based approaches show promise, experi-mental results suggest that further advancements are needed to bridge the gapbetween their current effectiveness and human-level evaluation quality. Recog-nizing that best practices of human evaluation processes often involve multiplehuman annotators collaborating in the evaluation, we resort to a multi-agent debateframework, moving beyond single-agent prompting strategies. The multi-agent-based approach enables a group of LLMs to synergize with an array of intelli-gent counterparts, harnessing their distinct capabilities and expertise to enhanceefficiency and effectiveness in handling intricate tasks. In this paper, we con-struct a multi-agent referee team called ChatEval to autonomously discuss andevaluate the quality of generated responses from different models on open-endedquestions and traditional natural language generation (NLG) tasks. We deriveinsights and lessons from practical scenarios where humans instigate group dis-cussions for brainstorming and propose different communication strategies withinChatEval......","Tsinghua University, Hong Kong University of Science and Technology, Peking University"
8,./images/chateval_towards_better_llm-based_20230814.png,"CoMM: Collaborative Multi-Agent, Multi-Reasoning-Path Prompting for Complex Problem Solving","Pei Chen, Boran Han, Shuai Zhang","Large Language Models (LLMs) have showngreat ability in solving traditional natural lan-guage tasks and elementary reasoning taskswith appropriate prompting techniques. How-ever, their ability is still limited in solving com-plicated science problems. In this work, weaim to push the upper bound of the reason-ing capability of LLMs by proposing a col-laborative multi-agent, multi-reasoning-path(CoMM) prompting framework. Specifically,we prompt LLMs to play different roles in aproblem-solving team, and encourage differ-ent role-play agents to collaboratively solvethe target task. In particular, we discover thatapplying different reasoning paths for differ-ent roles is an effective strategy to implementfew-shot prompting approaches in the multi-agent scenarios. Empirical results demonstratethe effectiveness of the proposed methods ontwo college-level science problems over com-petitive baselines. Our further analysis showsthe necessity of prompting LLMs to play dif-ferent roles or experts independently. We re-lease the code at: https://github.com/amazon-science/comm-prompt.","Texas A&M University, Amazon Web Services"
9,"./images/comm_collaborative_multi-agent,_multi-reasoning-path_20240426.png","Describe, Explain, Plan and Select: Interactive Planning with Large Language Models Enables Open-World Multi-Task Agents","Zihao Wang, Shaofei Cai, Guanzhou Chen, Anji Liu, Xiaojian Ma, Yitao Liang","We investigate the challenge of task planning for multi-task embodied agents in open-world environments. Two main difficulties are identified: 1) executing plans in an open-world environment (e.g., Minecraft) necessitates accurate and multi-step reasoning due to the long-term nature of tasks, and 2) as vanilla planners do not consider how easy the current agent can achieve a given sub-task when ordering parallel sub-goals within a complicated plan, the resulting plan could be inefficient or even infeasible. To this end, we propose ""Describe, Explain, Plan and Select"" (DEPS), an interactive planning approach based on Large Language Models (LLMs). DEPS facilitates better error correction on initial LLM-generated plan by integrating description of the plan execution process and providing self-explanation of feedback when encountering failures during the extended planning phases. Furthermore, it includes a goal selector, which is a trainable module that ranks parallel candidate sub-goals based on the estimated steps of completion, consequently refining the initial plan. Our experiments mark the milestone of the first zero-shot multi-task agent that can robustly accomplish 70+ Minecraft tasks and nearly double the overall performances. Further testing reveals our method's general effectiveness in popularly adopted non-open-ended domains as well (i.e., ALFWorld and tabletop manipulation). The ablation and exploratory studies detail how our design beats the counterparts and provide a promising update on the 𝙾𝚋𝚝𝚊𝚒𝚗𝙳𝚒𝚊𝚖𝚘𝚗𝚍 grand challenge with our approach.","Peking University, University of California Los Angeles, Beijing Institute for General Artificial Intelligence"
10,"./images/describe,_explain,_plan_and_20230203.png",Dynamic LLM-Agent Network: An LLM-agent Collaboration Framework with Agent Team Optimization,"Zijun Liu, Yanzhe Zhang, Peng Li, Yang Liu, Diyi Yang","Large language model (LLM) agents have been shown effective on a wide rangeof tasks, and by ensembling multiple LLM agents, their performances could befurther improved. Existing approaches employ a fixed set of agents to interactwith each other in a static architecture, which limits their generalizability to vari-ous tasks and requires strong human prior in designing these agents. In this work,we propose to construct a strategic team of agents communicating in a dynamicinteraction architecture based on the task query. Specifically, we build a frame-work named Dynamic LLM-Agent Network (DyLAN) for LLM-agent collabora-tion on complicated tasks like reasoning and code generation. DyLAN enablesagents to interact for multiple rounds in a dynamic architecture with inference-time agent selection and an early-stopping mechanism to improve performanceand efficiency. We further design an automatic agent team optimization algorithmbased on an unsupervised metric termed Agent Importance Score, enabling theselection of best agents based on the contribution each agent makes. Empirically,we demonstrate that DyLAN performs well in both reasoning and code generationtasks with reasonable computational cost. DyLAN achieves 1","Tsinghua University, Georgia Tech, Stanford University"
11,./images/dynamic_llm-agent_network_an_20231003.png,EconAgent: Large Language Model-Empowered Agents for Simulating Macroeconomic Activities,"Nian Li, Chen Gao, Mingyu Li, Yong Li, Qingmin Liao","The advent of artificial intelligence has led to agrowing emphasis on data-driven modeling inmacroeconomics, with agent-based modeling(ABM) emerging as a prominent bottom-upsimulation paradigm. In ABM, agents (e.g.,households, firms) interact within a macroe-conomic environment, collectively generatingmarket dynamics. Existing agent modeling typ-ically employs predetermined rules or learning-based neural networks for decision-making.However, customizing each agent presents sig-nificant challenges, complicating the modelingof agent heterogeneity. Additionally, the in-fluence of multi-period market dynamics andmultifaceted macroeconomic factors are oftenoverlooked in decision-making processes. Inthis work, we introduce EconAgent, a largelanguage model-empowered agent with human-like characteristics for macroeconomic simu-lation. We first construct a simulation envi-ronment that incorporates various market dy-namics driven by agents’ decisions regardingwork and consumption. Through the perceptionmodule, we create heterogeneous agents withdistinct decision-making mechanisms.Fur-thermore, we model the impact of macroeco-nomic trends using a memory module, whichallows agents to reflect on past individual ex-periences and market dynamics. Simulationexperiments show that EconAgent can makerealistic decisions, leading to more reasonablemacroeconomic phenomena compared to exist-ing rule-based or learning-based agents. Ourcodes are released at https://github.com/tsinghua-fib-lab/ACL24-EconAgent.",Tsinghua University
12,./images/econagent_large_language_model-empowered_20231016.png,Experiential Co-Learning of Software-Developing Agents,"Chen Qian, Yufan Dang, Jiahao Li, Wei Liu, Zihao Xie, Yifei Wang, Weize Chen, Cheng Yang, Xin Cong, Xiaoyin Che, Zhiyuan Liu, Maosong Sun","Recent advancements in large language mod-els (LLMs) have brought significant changesto various domains, especially through LLM-driven autonomous agents. A representativescenario is in software development, whereLLM agents demonstrate efficient collabora-tion, task division, and assurance of softwarequality, markedly reducing the need for man-ual involvement. However, these agents fre-quently perform a variety of tasks indepen-dently, without benefiting from past experi-ences, which leads to repeated mistakes andinefficient attempts in multi-step task execu-tion. To this end, we introduce Experiential Co-Learning, a novel LLM-agent learning frame-work in which instructor and assistant agentsgather shortcut-oriented experiences from theirhistorical trajectories and use these past expe-riences for future task execution. The exten-sive experiments demonstrate that the frame-work enables agents to tackle unseen software-developing tasks more effectively. We antici-pate that our insights will guide LLM agentstowards enhanced autonomy and contributeto their evolutionary growth in cooperativelearning. The code and data are available athttps://github.com/OpenBMB/ChatDev.","Tsinghua University, Dalian University of Technology, Beijing University of Posts and Telecommunications, Siemens"
13,./images/experiential_co-learning_of_software-developing_20231228.png,Exploring Large Language Models for Communication Games: An Empirical Study on Werewolf,"Yuzhuang Xu, Shuo Wang, Peng Li, Fuwen Luo, Xiaolong Wang, Weidong Liu, Yang Liu","Communication games, which we refer to asincomplete information games that heavily de-pend on natural language communication, holdsignificant research value in fields such as eco-nomics, social science, and artificial intelli-gence. In this work, we explore the problem ofhow to engage large language models (LLMs)in communication games, and in response, pro-pose a tuning-free framework. Our approachkeeps LLMs frozen, and relies on the retrievaland reflection on past communications and ex-periences for improvement. An empirical studyon the representative and widely-studied com-munication game, “Werewolf”, demonstratesthat our framework can effectively play Were-wolf game without tuning the parameters of theLLMs. More importantly, strategic behaviorsbegin to emerge in our experiments, suggest-ing that it will be a fruitful journey to engageLLMs in communication games and associateddomains.","Tsinghua University, Zhongguancun Laboratory"
14,./images/exploring_large_language_models_20230909.png,Facilitating Multi-Role and Multi-Behavior Collaboration of Large Language Models for Online Job Seeking and Recruiting,"Hongda Sun, Hongzhan Lin, Haiyu Yan, Chen Zhu, Yang Song, Xin Gao, Shuo Shang, Rui Yan","The emergence of online recruitment services has revolutionizedthe traditional landscape of job seeking and recruitment, neces-sitating the development of high-quality industrial applicationsto improve person-job fitting. Existing methods generally rely onmodeling the latent semantics of resumes and job descriptions andlearning a matching function between them. Inspired by the pow-erful role-playing capabilities of Large Language Models (LLMs),we propose to introduce a mock interview process between LLM-played interviewers and candidates. The mock interview conver-sations can provide additional evidence for candidate evaluation,thereby augmenting traditional person-job fitting based solely onresumes and job descriptions. However, characterizing these tworoles in online recruitment still presents several challenges, suchas developing the skills to raise interview questions, formulatingappropriate answers, and evaluating two-sided fitness.To this end, we propose MockLLM, a novel applicable frameworkthat divides the person-job matching process into two modules:mock interview generation and two-sided evaluation in handshakeprotocol, jointly enhancing their performance through collaborativebehaviors between interviewers and candidates. We design a role-playing framework as a multi-role and multi-behavior paradigmto enable a single LLM agent to effectively behave with multiplefunctions for both parties......","Renmin University of China, BOSS Zhipin, King Abdullah University of Science and Technology, University of Electronic Science and Technology of China"
15,./images/facilitating_multi-role_and_multi-behavior_20240528.png,GameGPT: Multi-agent Collaborative Framework for Game Development,"Dake Chen, Hanbin Wang, Yunhao Huo, Yuzhao Li, Haoyang Zhang","The large language model (LLM) based agents have demonstrated their capacityto automate and expedite software development processes. In this paper, wefocus on game development and propose a multi-agent collaborative framework,dubbed GameGPT, to automate game development. While many studies havepinpointed hallucination as a primary roadblock for deploying LLMs in production,we identify another concern: redundancy. Our framework presents a series ofmethods to mitigate both concerns. These methods include dual collaboration andlayered approaches with several in-house lexicons, to mitigate the hallucinationand redundancy in the planning, task identification, and implementation phases.Furthermore, a decoupling approach is also introduced to achieve code generationwith better precision.","AutoGame Research, X-Institute, University of Southern California"
16,./images/gamegpt_multi-agent_collaborative_framework_20231012.png,Generative Agents: Interactive Simulacra of Human Behavior,"Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein","Believable proxies of human behavior can empower interactive applications ranging from immersive environments to rehearsal spaces for interpersonal communication to prototyping tools. In this paper, we introduce generative agents--computational software agents that simulate believable human behavior. Generative agents wake up, cook breakfast, and head to work; artists paint, while authors write; they form opinions, notice each other, and initiate conversations; they remember and reflect on days past as they plan the next day. To enable generative agents, we describe an architecture that extends a large language model to store a complete record of the agent's experiences using natural language, synthesize those memories over time into higher-level reflections, and retrieve them dynamically to plan behavior. We instantiate generative agents to populate an interactive sandbox environment inspired by The Sims, where end users can interact with a small town of twenty five agents using natural language. In an evaluation, these generative agents produce believable individual and emergent social behaviors: for example, starting with only a single user-specified notion that one agent wants to throw a Valentine's Day party, the agents autonomously spread invitations to the party over the next two days, make new acquaintances, ask each other out on dates to the party, and coordinate to show up for the party together at the right time. We demonstrate through ablation that the components of our agent architecture--observation, planning, and reflection--each contribute critically to the believability of agent behavior. By fusing large language models with computational, interactive agents, this work introduces architectural and interaction patterns for enabling believable simulations of human behavior.","Stanford University, Google Research, Google DeepMind"
17,./images/generative_agents_interactive_simulacra_20230407.png,Improving Multi-Agent Debate with Sparse Communication Topology,"Yunxuan Li, Yibing Du, Jiageng Zhang, Le Hou, Peter Grabowski, Yeqing Li, Eugene Ie","Multi-agent debate has proven effective in im-proving large language models quality for rea-soning and factuality tasks. While various role-playing strategies in multi-agent debates havebeen explored, in terms of the communica-tion among agents, existing approaches adopta brute force algorithm – each agent can com-municate with all other agents. In this paper,we systematically investigate the effect of com-munication connectivity in multi-agent systems.Our experiments on GPT and Mistral models re-veal that multi-agent debates leveraging sparsecommunication topology can achieve compara-ble or superior performance while significantlyreducing computational costs. Furthermore, weextend the multi-agent debate framework tomultimodal reasoning and alignment labelingtasks, showcasing its broad applicability andeffectiveness. Our findings underscore the im-portance of communication connectivity on en-hancing the efficiency and effectiveness of the“society of minds” approach.","Google, Google DeepMind"
18,./images/improving_multi-agent_debate_with_20240617.png,Iterative Experience Refinement of Software-Developing Agents,"Chen Qian, Jiahao Li, Yufan Dang, Wei Liu, YiFei Wang, Zihao Xie, Weize Chen, Cheng Yang, Yingli Zhang, Zhiyuan Liu, Maosong Sun","Autonomous agents powered by large languagemodels (LLMs) show significant potential forachieving high autonomy in various scenar-ios such as software development. Recent re-search has shown that LLM agents can lever-age past experiences to reduce errors and en-hance efficiency. However, the static experi-ence paradigm, reliant on a fixed collection ofpast experiences acquired heuristically, lacksiterative refinement and thus hampers agents’adaptability. In this paper, we introduce the It-erative Experience Refinement framework, en-abling LLM agents to refine experiences itera-tively during task execution. We propose twofundamental patterns: the successive pattern,refining based on nearest experiences within atask batch, and the cumulative pattern, acquir-ing experiences across all previous task batches.Augmented with our heuristic experience elim-ination, the method prioritizes high-quality andfrequently-used experiences, effectively man-aging the experience space and enhancing effi-ciency. Extensive experiments show that whilethe successive pattern may yield superior re-sults, the cumulative pattern provides more sta-ble performance......","Tsinghua University, Dalian University of Technology, Beijing University of Posts and Telecommunications, Siemens"
19,./images/iterative_experience_refinement_of_20240507.png,Language Agents as Optimizable Graphs,"Mingchen Zhuge, Wenyi Wang, Louis Kirsch, Francesco Faccio, Dmitrii Khizbullin, Jürgen Schmidhuber","Various human-designed prompt engineering techniques have been proposed to improve problem solvers based on Large Language Models (LLMs), yielding many disparate code bases. We unify these approaches by describing LLM-based agents as computational graphs. The nodes implement functions to process multimodal data or query LLMs, and the edges describe the information flow between operations. Graphs can be recursively combined into larger composite graphs representing hierarchies of inter-agent collaboration (where edges connect operations of different agents). Our novel automatic graph optimizers (1) refine node-level LLM prompts (node optimization) and (2) improve agent orchestration by changing graph connectivity (edge optimization). Experiments demonstrate that our framework can be used to efficiently develop, integrate, and automatically improve various LLM agents. ","King Abdullah University of Science and Technology, The Swiss AI Lab IDSIA, USI, SUPSI"
20,./images/language_agents_as_optimizable_20240226.png,Large Language Models are Diverse Role-Players for Summarization Evaluation,"Ning Wu, Ming Gong, Linjun Shou, Shining Liang, Daxin Jiang",". Text summarization has a wide range of applications in many scenarios.The evaluation of the quality of the generated text is a complex problem. A bigchallenge to language evaluation is that there is a clear divergence between existingmetrics and human evaluation. A document summary’s quality can be assessedby human annotators on various criteria, both objective ones like grammar andcorrectness, and subjective ones like informativeness, succinctness, and appeal.Most of the automatic evaluation methods like BLUE/ROUGE may be not ableto adequately capture the above dimensions. In this paper, we propose a newevaluation framework based on LLMs, which provides a comprehensive evaluationframework by comparing generated text and reference text from both objective andsubjective aspects. First, we propose to model objective and subjective dimensionsof generated text based on roleplayers prompting mechanism. Furthermore, weintroduce a context-based prompting mechanism that is able to generate dynamicroleplayer profiles based on input context. Finally, we design a multi-roleplayerprompting technology based on batch prompting and integrate multiple outputsinto the final evaluation results. Experimental results on three real datasets forsummarization show that our model is highly competitive and has a very highconsistency with human annotators.",Microsoft
21,./images/large_language_models_are_20230327.png,Learn to Disguise: Avoid Refusal Responses in LLM's Defense via a Multi-agent Attacker-Disguiser Game,"Qianqiao Xu, Zhiliang Tian, Hongyan Wu, Zhen Huang, Yiping Song, Feng Liu, Dongsheng Li","With the enhanced performance of large models on natural language processingtasks, potential moral and ethical issues of large models arise. There exist ma-licious attackers who induce large models to jailbreak and generate informationcontaining illegal, privacy-invasive information through techniques such as promptengineering. As a result, large models counter malicious attackers’ attacks usingtechniques such as safety alignment. However, the strong defense mechanismof the large model through rejection replies is easily identified by attackers andused to strengthen attackers’ capabilities. In this paper, we propose a multi-agentattacker-disguiser game approach to achieve a weak defense mechanism that allowsthe large model to both safely reply to the attacker and hide the defense intent. First,we construct a multi-agent framework to simulate attack and defense scenarios,playing different roles to be responsible for attack, disguise, safety evaluation,and disguise evaluation tasks. After that, we design attack and disguise gamealgorithms to optimize the game strategies of the attacker and the disguiser and usethe curriculum learning process to strengthen the capabilities of the agents. Theexperiments verify that the method in this paper is more effective in strengtheningthe model’s ability to disguise the defense intent compared with other methods.Moreover, our approach can adapt any black-box large model to assist the model indefense and does not suffer from model version iterations.","National University of Defense Technology, Guangdong University of Foreign Studies, "
22,./images/learn_to_disguise_avoid_20240403.png,Leveraging Large Language Models for Collective Decision-Making,"Marios Papachristou, Longqi Yang, Chin-Chia Hsu","In various work contexts, such as meeting scheduling, collaborating, and project planning, collective decision-making is essential but often challenging due to diverse individual preferences, varying work focuses, and power dynamics among members. To address this, we propose a system leveraging Large Language Models (LLMs) to facilitate group decision-making by managing conversations and balancing preferences among individuals. Our system aims to extract individual preferences from conversations and suggest options that satisfy the preferences of the members. We specifically apply this system to corporate meeting scheduling. We create synthetic employee profiles and simulate conversations at scale, leveraging LLMs to evaluate the system performance as a novel approach to conducting a user study. Our results indicate efficient coordination with reduced interactions between the members and the LLM-based system. The system refines and improves its proposed options over time, ensuring that many of the members' individual preferences are satisfied in an equitable way. Finally, we conduct a survey study involving human participants to assess our system's ability to aggregate preferences and reasoning about them. Our findings show that the system exhibits strong performance in both dimensions","Cornell University, Microsoft"
23,./images/leveraging_large_language_models_20231103.png,LLM-Based Agent Society Investigation: Collaboration and Confrontation in Avalon Gameplay,"Yihuai Lan, Zhiqiang Hu, Lei Wang, Yang Wang, Deheng Ye, Peilin Zhao, Ee-Peng Lim, Hui Xiong, Hao Wang","This paper explores the open research prob-lem of understanding the social behaviors ofLLM-based agents. Using Avalon as a testbed,we employ system prompts to guide LLMagents in gameplay. While previous studieshave touched on gameplay with LLM agents,research on their social behaviors is lacking.We propose a novel framework, tailored forAvalon, features a multi-agent system facil-itating efficient communication and interac-tion. We evaluate its performance based ongame success and analyze LLM agents’ so-cial behaviors. Results affirm the framework’seffectiveness in creating adaptive agents andsuggest LLM-based agents’ potential in nav-igating dynamic social interactions. By ex-amining collaboration and confrontation be-haviors, we offer insights into this field’s re-search and applications.Our code is pub-licly available at https://github.com/3DAgentWorld/LLM-Game-Agent","The Hong Kong University of Science and Technology (Guangzhou), Singapore University of Technology and Design, Singapore Management University, Verily Life Sciences, Tencent"
24,./images/llm-based_agent_society_investigation_20231023.png,LongAgent: Scaling Language Models to 128k Context through Multi-Agent Collaboration,"Jun Zhao, Can Zu, Hao Xu, Yi Lu, Wei He, Yiwen Ding, Tao Gui, Qi Zhang, Xuanjing Huang","Large language models (LLMs) have demon-strated impressive performance in understand-ing language and executing complex reasoningtasks. However, LLMs with long context win-dows have been notorious for their expensivetraining costs and high inference latency. Eventhe most advanced models such as GPT-4 andClaude2 often make mistakes when processinginputs of over 100k tokens, a phenomenon alsoknown as lost in the middle. In this paper,we propose LONGAGENT, a method basedon multi-agent collaboration, which scalesLLMs (e.g., LLaMA) to a context of 128K anddemonstrates potential superiority in long-textprocessing compared to GPT-",Fudan University
25,./images/longagent_scaling_language_models_20240218.png,MetaAgents: Simulating Interactions of Human Behaviors for LLM-based Task-oriented Coordination via Collaborative Generative Agents,"Yuan Li, Yixuan Zhang, Lichao Sun","Significant advancements have occurred in the application of Large LanguageModels (LLMs) for various tasks and social simulations. Despite this, their capac-ities to coordinate within task-oriented social contexts are under-explored. Suchcapabilities are crucial if LLMs are to effectively mimic human-like social be-havior and produce meaningful results. To bridge this gap, we introduce collab-orative generative agents, endowing LLM-based Agents with consistent behaviorpatterns and task-solving abilities. We situate these agents in a simulated job fairenvironment as a case study to scrutinize their coordination skills. We proposea novel framework that equips collaborative generative agents with human-likereasoning abilities and specialized skills. Our evaluation demonstrates that theseagents show promising performance. However, we also uncover limitations thathinder their effectiveness in more complex coordination tasks. Our work providesvaluable insights into the role and evolution of LLMs in task-oriented social sim-ulations.","University of Cambridge, William & Mary, Lehigh University"
26,./images/metaagents_simulating_interactions_of_20231010.png,MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework,"Sirui Hong, Mingchen Zhuge, Jonathan Chen, Xiawu Zheng, Yuheng Cheng, Ceyao Zhang, Jinlin Wang, Zili Wang, Steven Ka Shing Yau, Zijuan Lin, Liyang Zhou, Chenyu Ran, Lingfeng Xiao, Chenglin Wu, Jürgen Schmidhuber","Remarkable progress has been made on automated problem solving through so-cieties of agents based on large language models (LLMs). Existing LLM-basedmulti-agent systems can already solve simple dialogue tasks. Solutions to morecomplex tasks, however, are complicated through logic inconsistencies due tocascading hallucinations caused by naively chaining LLMs. Here we introduceMetaGPT, an innovative meta-programming framework incorporating efficienthuman workflows into LLM-based multi-agent collaborations.MetaGPT en-codes Standardized Operating Procedures (SOPs) into prompt sequences for morestreamlined workflows, thus allowing agents with human-like domain expertiseto verify intermediate results and reduce errors. MetaGPT utilizes an assemblyline paradigm to assign diverse roles to various agents, efficiently breaking downcomplex tasks into subtasks involving many agents working together. On col-laborative software engineering benchmarks, MetaGPT generates more coherentsolutions than previous chat-based multi-agent systems. Our project can be foundat https://github.com/geekan/MetaGPT","DeepWisdom, King Abdullah University of Science and Technology, Xiamen University, The Chinese University of Hong Kong (Shenzhen), Nanjing University, University of Pennsylvania University of California, Berkeley, The Swiss AI Lab IDSIA/USI/SUPSI"
27,./images/metagpt_meta_programming_for_20230801.png,Mora: Enabling Generalist Video Generation via A Multi-Agent Framework,"Zhengqing Yuan, Ruoxi Chen, Zhaoxu Li, Haolong Jia, Lifang He, Chi Wang, Lichao Sun","Sora is the first large-scale generalist video generation model that garnered significant attention across society. Since its launch by OpenAI in February 2024, no other video generation models have paralleled {Sora}'s performance or its capacity to support a broad spectrum of video generation tasks. Additionally, there are only a few fully published video generation models, with the majority being closed-source. To address this gap, this paper proposes a new multi-agent framework Mora, which incorporates several advanced visual AI agents to replicate generalist video generation demonstrated by Sora. In particular, Mora can utilize multiple visual agents and successfully mimic Sora's video generation capabilities in various tasks, such as (1) text-to-video generation, (2) text-conditional image-to-video generation, (3) extend generated videos, (4) video-to-video editing, (5) connect videos and (6) simulate digital worlds. Our extensive experimental results show that Mora achieves performance that is proximate to that of Sora in various tasks. However, there exists an obvious performance gap between our work and Sora when assessed holistically. In summary, we hope this project can guide the future trajectory of video generation through collaborative AI agents.","Lehigh University, Microsoft Research"
28,./images/mora_enabling_generalist_video_20240320.png,Multi-Agent Software Development through Cross-Team Collaboration,"Zhuoyun Du, Chen Qian, Wei Liu, Zihao Xie, Yifei Wang, Yufan Dang, Weize Chen, Cheng Yang","The latest breakthroughs in Large LanguageModels (LLMs), e.g., ChatDev, have catalyzedprofound transformations, particularly throughmulti-agent collaboration for software devel-opment. LLM agents can collaborate in teamslike humans, and follow the waterfall modelto sequentially work on requirements analysis,development, review, testing, and other phasesto perform autonomous software generation.However, for an agent team, each phase in asingle development process yields only one pos-sible outcome. This results in the completionof only one development chain, thereby losingthe opportunity to explore multiple potentialdecision paths within the solution space. Con-sequently, this may lead to obtaining subop-timal results. To address this challenge, weintroduce Cross-Team Collaboration (CTC),a scalable multi-team framework that enablesorchestrated teams to jointly propose variousdecisions and communicate with their insightsin a cross-team collaboration environment forsuperior content generation. Experimental re-sults in software development reveal a notableincrease in quality compared to state-of-the-art baselines, underscoring the efficacy of ourframework. The significant improvements instory generation demonstrate the promisinggeneralization ability of our framework acrossvarious domains. We anticipate that our workwill guide LLM agents towards a cross-teamparadigm and contribute to their significantgrowth in but not limited to software devel-opment. The code and data will be available athttps://github.com/OpenBMB/ChatDev.","Zhejiang University, Tsinghua University, Beijing University of Posts and Telecommunications"
29,./images/multi-agent_software_development_through_20240613.png,MultiAgent Collaboration Attack: Investigating Adversarial Attacks in Large Language Model Collaborations via Debate,"Alfonso Amayuelas, Xianjun Yang, Antonis Antoniades, Wenyue Hua, Liangming Pan, William Wang","Large Language Models (LLMs) have shownexceptional results on current benchmarkswhen working individually. The advancementin their capabilities, along with a reduction inparameter size and inference times, has facil-itated the use of these models as agents, en-abling interactions among multiple models toexecute complex tasks. Such collaborationsoffer several advantages, including the use ofspecialized models (e.g. coding), improvedconfidence through multiple computations, andenhanced divergent thinking, leading to morediverse outputs. Thus, the collaborative use oflanguage models is expected to grow signifi-cantly in the coming years. In this work, weevaluate the behavior of a network of modelscollaborating through debate under the influ-ence of an adversary. We introduce pertinentmetrics to assess the adversary’s effectiveness,focusing on system accuracy and model agree-ment. Our findings highlight the importanceof a model’s persuasive ability in influencingothers. Additionally, we explore inference-timemethods to generate more compelling argu-ments and evaluate the potential of prompt-based mitigation as a defensive strategy.","UC Santa Barbara, Rutgers University"
30,./images/multiagent_collaboration_attack_investigating_20240620.png,ReConcile: Round-Table Conference Improves Reasoning via Consensus among Diverse LLMs,"Justin Chih-Yao Chen, Swarnadeep Saha, Mohit Bansal","Large Language Models (LLMs) still struggle with natural language reasoning tasks. Motivated by the society of minds (Minsky, 1988), we propose ReConcile, a multi-model multi-agent framework designed as a round table conference among diverse LLM agents. ReConcile enhances collaborative reasoning between LLM agents via multiple rounds of discussion, learning to convince other agents to improve their answers, and employing a confidence-weighted voting mechanism that leads to a better consensus. In each round, ReConcile initiates discussion between agents via a 'discussion prompt' that consists of (a) grouped answers and explanations generated by each agent in the previous round, (b) their confidence scores, and (c) demonstrations of answer-rectifying human explanations, used for convincing other agents. Experiments on seven benchmarks demonstrate that ReConcile significantly improves LLMs' reasoning -- both individually and as a team -- surpassing prior single-agent and multi-agent baselines by up to 11.4% and even outperforming GPT-4 on three datasets. ReConcile also flexibly incorporates different combinations of agents, including API-based, open-source, and domain-specific models, leading to an 8% improvement on MATH. Finally, we analyze the individual components of ReConcile, demonstrating that the diversity originating from different models is critical to its superior performance.",UNC Chapel Hill
31,./images/reconcile_round-table_conference_improves_20230922.png,Rethinking the Bounds of LLM Reasoning: Are Multi-Agent Discussions the Key?,"Qineng Wang, Zihao Wang, Ying Su, Hanghang Tong, Yangqiu Song","Recent progress in LLMs discussion suggeststhat multi-agent discussion improves the rea-soning abilities of LLMs. In this work, wereevaluate this claim through systematic experi-ments, where we propose a novel group discus-sion framework to enrich the set of discussionmechanisms. Interestingly, our results showthat a single-agent LLM with strong promptscan achieve almost the same performance asthe best existing discussion approach on a widerange of reasoning tasks and backbone LLMs.We observe that the multi-agent discussion per-forms better than a single agent only when thereis no demonstration in the prompt. Furtherstudy reveals the common interaction mecha-nisms of LLMs during the discussion.","Zhejiang University, HKUST, UIUC"
32,./images/rethinking_the_bounds_of_20240228.png,Scalable Multi-Robot Collaboration with Large Language Models: Centralized or Decentralized Systems?,"Yongchao Chen, Jacob Arkin, Yang Zhang, Nicholas Roy, Chuchu Fan","— A flurry of recent work has demonstrated thatpre-trained large language models (LLMs) can be effectivetask planners for a variety of single-robot tasks. The planningperformance of LLMs is significantly improved via promptingtechniques, such as in-context learning or re-prompting withstate feedback, placing new importance on the token budgetfor the context window. An under-explored but natural nextdirection is to investigate LLMs as multi-robot task planners.However, long-horizon, heterogeneous multi-robot planningintroduces new challenges of coordination while also pushingup against the limits of context window length. It is thereforecritical to find token-efficient LLM planning frameworks thatare also able to reason about the complexities of multi-robotcoordination. In this work, we compare the task success rate andtoken efficiency of four multi-agent communication frameworks(centralized, decentralized, and two hybrid) as applied tofour coordination-dependent multi-agent 2D task scenarios forincreasing numbers of agents. We find that a hybrid frameworkachieves better task success rates across all four tasks andscales better to more agents. We further demonstrate the hybridframeworks in 3D simulations where the vision-to-text problemand dynamical errors are considered. ","Massachusetts Institute of Technology, Harvard University, MIT-IBM Watson AI Lab. "
33,./images/scalable_multi-robot_collaboration_with_20230927.png,Scaling Large-Language-Model-based Multi-Agent Collaboration,"Chen Qian, Zihao Xie, Yifei Wang, Wei Liu, Yufan Dang, Zhuoyun Du, Weize Chen, Cheng Yang, Zhiyuan Liu, Maosong Sun","Pioneering advancements in large languagemodel-powered agents have underscored thedesign pattern of multi-agent collaboration,demonstrating that collective intelligence cansurpass the capabilities of each individual. In-spired by the neural scaling law, which positsthat increasing neurons leads to emergent abil-ities, this study investigates whether a simi-lar principle applies to increasing agents inmulti-agent collaboration.Technically, wepropose ::multi-agent:collaboration::networks(MACNET), which utilize directed acyclicgraphs to organize agents and streamline theirinteractive reasoning via topological ordering,with solutions derived from their dialogues.Extensive experiments show that MACNETconsistently outperforms baseline models, en-abling effective agent collaboration across var-ious network topologies and supporting coop-eration among more than a thousand agents.Notably, we observed a small-world collabo-ration phenomenon, where topologies resem-bling small-world properties achieved supe-rior performance. Additionally, we identifieda collaborative scaling law, indicating thatnormalized solution quality follows a logisticgrowth pattern as scaling agents, with collabo-rative emergence occurring much earlier thanpreviously observed instances of neural emer-gence. The code and data will be available athttps://github.com/OpenBMB/ChatDev.","Tsinghua University, Beijing University of Posts and Telecommunications"
34,./images/scaling_large-language-model-based_multi-agent_collaboration_20240611.png,Self-Organized Agents: A LLM Multi-Agent Framework toward Ultra Large-Scale Code Generation and Optimization,"Yoichi Ishibashi, Yoshimasa Nishimura","Recent advancements in automatic code gener-ation using large language model (LLM) agenthave brought us closer to the future of auto-mated software development. However, exist-ing single-agent approaches face limitationsin generating and improving large-scale, com-plex codebases due to constraints in contextlength. To tackle this challenge, we proposeSelf-Organized multi-Agent framework (SoA),a novel multi-agent framework that enables thescalable and efficient generation and optimiza-tion of large-scale code. In SoA, self-organizedagents operate independently to generate andmodify code components while seamlessly col-laborating to construct the overall codebase. Akey feature of our framework is the automaticmultiplication of agents based on problem com-plexity, allowing for dynamic scalability. Thisenables the overall code volume to be increasedindefinitely according to the number of agents,while the amount of code managed by eachagent remains constant. We evaluate SoA onthe HumanEval benchmark and demonstratethat, compared to a single-agent system, eachagent in SoA handles significantly less code,yet the overall generated code is substantiallygreater. Moreover, SoA surpasses the powerfulsingle-agent baseline by 5%......",TsukushiAI
35,./images/self-organized_agents_a_llm_20240402.png,"StrategyLLM: Large Language Models as Strategy Generators, Executors, Optimizers, and Evaluators for Problem Solving","Chang Gao, Haiyun Jiang, Deng Cai, Shuming Shi, Wai Lam","Most existing prompting methods suffer from the issues of generalizability and consistency, as they often rely on instance-specific solutions that may not be applicable to other instances and lack task-level consistency across the selected few-shot examples. To address these limitations, we propose a comprehensive framework, StrategyLLM, allowing LLMs to perform inductive reasoning, deriving general strategies from specific task instances, and deductive reasoning, applying these general strategies to particular task examples, for constructing generalizable and consistent few-shot prompts. It employs four LLM-based agents: strategy generator, executor, optimizer, and evaluator, working together to generate, evaluate, and select promising strategies for a given task. Experimental results demonstrate that StrategyLLM outperforms the competitive baseline CoT-SC that requires human-annotated solutions on 13 datasets across 4 challenging tasks without human involvement, including math reasoning (34.2\% → 38.8\%), commonsense reasoning (70.3\% → 72.5\%), algorithmic reasoning (73.7\% → 85.0\%), and symbolic reasoning (30.0\% → 79.2\%). Further analysis reveals that StrategyLLM is applicable to various LLMs and demonstrates advantages across numerous scenarios.","The Chinese University of Hong Kong, Sun Yat-sen University, Tencent AI Lab"
36,./images/strategyllm_large_language_models_20231115.png,TraveLER: A Multi-LMM Agent Framework for Video Question-Answering,"Chuyi Shang, Amos You, Sanjay Subramanian, Trevor Darrell, Roei Herzig","Recently, Large Multimodal Models (LMMs) have made significant progressin video question-answering using a frame-wise approach by leveraginglarge-scale, image-based pretraining in a zero-shot manner. While image-based methods for videos have shown impressive performance, a currentlimitation is that they often overlook how key timestamps are selected andcannot adjust when incorrect timestamps are identified. Moreover, they areunable to extract details relevant to the question, instead providing generaldescriptions of the frame. To overcome this, we design a multi-LMM agentframework that travels along the video, iteratively collecting relevant in-formation from keyframes through interactive question-asking until thereis sufficient information to answer the question. Specifically, we proposeTraveLER, a model that can create a plan to “Traverse” through the video,ask questions about individual frames to “Locate” and store key informa-tion, and then “Evaluate” if there is enough information to answer thequestion. Finally, if there is not enough information, our method is able to“Replan” based on its collected knowledge. Through extensive experiments,we find that the proposed TraveLER approach improves performance onseveral video question-answering benchmarks, such as NExT-QA, STAR,and Perception Test, without the need to fine-tune on specific datasets.","University of California, Berkeley"
37,./images/traveler_a_multi-lmm_agent_20240401.png,Unleashing the Emergent Cognitive Synergy in Large Language Models: A Task-Solving Agent through Multi-Persona Self-Collaboration,"Zhenhailong Wang, Shaoguang Mao, Wenshan Wu, Tao Ge, Furu Wei, Heng Ji","Human intelligence thrives on cognitive syn-ergy, where collaboration among differentminds yield superior outcomes compared to iso-lated individuals. In this work, we propose SoloPerformance Prompting (SPP), which trans-forms a single LLM into a cognitive synergistby engaging in multi-turn self-collaborationwith multiple personas.A cognitive syner-gist is an intelligent agent that collaborativelycombines multiple minds’ strengths and knowl-edge to enhance problem-solving in complextasks. By dynamically identifying and simu-lating different personas based on task inputs,SPP unleashes the potential of cognitive syn-ergy in LLMs. Our in-depth analysis showsthat assigning multiple fine-grained personasin LLMs improves problem-solving abilitiescompared to using a single or fixed numberof personas. We evaluate SPP on three chal-lenging tasks: Trivia Creative Writing, Code-names Collaborative, and Logic Grid Puzzle,encompassing both knowledge-intensive andreasoning-intensive types. Unlike previousworks, such as Chain-of-Thought, that solelyenhance the reasoning abilities in LLMs, ex-perimental results demonstrate that SPP effec-tively reduces factual hallucination, and main-tains strong reasoning capabilities. Addition-ally, comparative experiments show that cog-nitive synergy only emerges in GPT-4 anddoes not appear in less capable models, suchas GPT-","University of Illinois Urbana-Champaign, Microsoft Research Asia"
38,./images/unleashing_the_emergent_cognitive_20230711.png,User Behavior Simulation with Large Language Model based Agents,"Lei Wang, Jingsen Zhang, Hao Yang, Zhiyuan Chen, Jiakai Tang, Zeyu Zhang, Xu Chen, Yankai Lin, Ruihua Song, Wayne Xin Zhao, Jun Xu, Zhicheng Dou, Jun Wang, Ji-Rong Wen","Simulating high quality user behavior data has always been a fundamental problem in human-centered applications, where the major difficulty originates from the intricate mechanism of human decision process. Recently, substantial evidences have suggested that by learning huge amounts of web knowledge, large language models (LLMs) can achieve human-like intelligence. We believe these models can provide significant opportunities to more believable user behavior simulation. To inspire such direction, we propose an LLM-based agent framework and design a sandbox environment to simulate real user behaviors. Based on extensive experiments, we find that the simulated behaviors of our method are very close to the ones of real humans. Concerning potential applications, we simulate and study two social phenomenons including (1) information cocoons and (2) user conformity behaviors. This research provides novel simulation paradigms for human-centered applications.","Renmin University of China, Beijing Key Laboratory of Big Data Management and Analysis Methods, University College London"
39,./images/user_behavior_simulation_with_20230605.png,War and Peace (WarAgent): Large Language Model-based Multi-Agent Simulation of World Wars,"Wenyue Hua, Lizhou Fan, Lingyao Li, Kai Mei, Jianchao Ji, Yingqiang Ge, Libby Hemphill, Yongfeng Zhang","Can we avoid wars at the crossroads of history? This question has been pursued byindividuals, scholars, policymakers, and organizations throughout human history.In this research, we attempt to answer the question based on the recent advancesof Artificial Intelligence (AI) and Large Language Models (LLMs). We proposeWarAgent, an LLM-powered multi-agent AI system, to simulate the participatingcountries, their decisions, and the consequences, in historical international conflicts,including the World War I (WWI), the World War II (WWII), and the WarringStates Period (WSP) in Ancient China. By evaluating the simulation effectiveness,we examine the advancements and limitations of cutting-edge AI systems’ abilitiesin studying complex collective human behaviors such as international conflictsunder diverse settings. In these simulations, the emergent interactions amongagents also offer a novel perspective for examining the triggers and conditions thatlead to war. Our findings offer data-driven and AI-augmented insights that canredefine how we approach conflict resolution and peacekeeping strategies. Theimplications stretch beyond historical analysis, offering a blueprint for using AI tounderstand human history and possibly prevent future international conflicts. Codeand data are available at https://github.com/agiresearch/WarAgent.",Rutgers University
40,./images/war_and_peace_(waragent)_20231128.png,To be Continued...,Your Contributions are Welcome!,,
1 image_path title author summary affiliation
2 0 ./images/2d.png (Perhaps) Beyond Human Translation: Harnessing Multi-Agent Collaboration for Translating Ultra-Long Literary Texts Minghao Wu, Yulin Yuan, Gholamreza Haffari, Longyue Wang Recent advancements in machine translation (MT) have significantly enhancedtranslation quality across various domains. However, the translation of literarytexts remains a formidable challenge due to their complex language, figurative ex-pressions, and cultural nuances. In this work, we introduce a novel multi-agentframework based on large language models (LLMs) for literary translation, im-plemented as a company called TRANSAGENTS, which mirrors traditional trans-lation publication process by leveraging the collective capabilities of multipleagents, to address the intricate demands of translating literary works. To evaluatethe effectiveness of our system, we propose two innovative evaluation strategies:Monolingual Human Preference (MHP) and Bilingual LLM Preference (BLP).MHP assesses translations from the perspective of monolingual readers of the tar-get language, while BLP uses advanced LLMs to compare translations directlywith the original texts. Empirical findings indicate that despite lower d-BLEUscores, translations from TRANSAGENTS are preferred by both human evalua-tors and LLMs over human-written references, particularly in genres requiringdomain-specific knowledge. We also highlight the strengths and limitations ofTRANSAGENTS through case studies and suggests directions for future research. Monash University, University of Macau, Tencent AI Lab
3 1 ./images/(perhaps)_beyond_human_translation_20240520.png Agent Hospital: A Simulacrum of Hospital with Evolvable Medical Agents Junkai Li, Siyu Wang, Meng Zhang, Weitao Li, Yunghwei Lai, Xinhui Kang, Weizhi Ma, Yang Liu In this paper, we introduce a simulacrum of hospital called Agent Hospital that simulates theentire process of treating illness. All patients, nurses, and doctors are autonomous agents powered bylarge language models (LLMs). Our central goal is to enable a doctor agent to learn how to treat illnesswithin the simulacrum. To do so, we propose a method called MedAgent-Zero. As the simulacrum cansimulate disease onset and progression based on knowledge bases and LLMs, doctor agents can keepaccumulating experience from both successful and unsuccessful cases. Simulation experiments show thatthe treatment performance of doctor agents consistently improves on various tasks. More interestingly,the knowledge the doctor agents have acquired in Agent Hospital is applicable to real-world medicarebenchmarks. After treating around ten thousand patients (real-world doctors may take over two years),the evolved doctor agent achieves a state-of-the-art accuracy of 9 Tsinghua University
4 2 ./images/agent_hospital_a_simulacrum_20240505.png AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li, Erkang Zhu, Li Jiang, Xiaoyun Zhang, Shaokun Zhang, Jiale Liu, Ahmed Hassan Awadallah, Ryen W White, Doug Burger, Chi Wang AutoGen2 is an open-source framework that allows developers to build LLM ap-plications via multiple agents that can converse with each other to accomplishtasks. AutoGen agents are customizable, conversable, and can operate in vari-ous modes that employ combinations of LLMs, human inputs, and tools. UsingAutoGen, developers can also flexibly define agent interaction behaviors. Bothnatural language and computer code can be used to program flexible conversationpatterns for different applications. AutoGen serves as a generic framework forbuilding diverse applications of various complexities and LLM capacities. Em-pirical studies demonstrate the effectiveness of the framework in many exampleapplications, with domains ranging from mathematics, coding, question answer-ing, operations research, online decision-making, entertainment, etc. Microsoft Research, Pennsylvania State University, University of Washington, Xidian University
5 3 ./images/autogen_enabling_next-gen_llm_20230816.png Avalon's Game of Thoughts: Battle Against Deception through Recursive Contemplation Shenzhi Wang, Chang Liu, Zilong Zheng, Siyuan Qi, Shuo Chen, Qisen Yang, Andrew Zhao, Chaofei Wang, Shiji Song, Gao Huang Recent breakthroughs in large language models (LLMs) have brought remark-able success in the field of LLM-as-Agent. Nevertheless, a prevalent assumptionis that the information processed by LLMs is consistently honest, neglecting thepervasive deceptive or misleading information in human society and AI-generatedcontent.This oversight makes LLMs susceptible to malicious manipulations,potentially resulting in detrimental outcomes. This study utilizes the intricateAvalon game as a testbed to explore LLMs’ potential in deceptive environments.Avalon, full of misinformation and requiring sophisticated logic, manifests as a“Game-of-Thoughts”. Inspired by the efficacy of humans’ recursive thinking andperspective-taking in the Avalon game, we introduce a novel framework, Recur-sive Contemplation (ReCon), to enhance LLMs’ ability to identify and counteractdeceptive information. ReCon combines formulation and refinement contempla-tion processes; formulation contemplation produces initial thoughts and speech,while refinement contemplation further polishes them. Additionally, we incor-porate first-order and second-order perspective transitions into these processesrespectively. Specifically, the first-order allows an LLM agent to infer others’mental states, and the second-order involves understanding how others perceivethe agent’s mental state....... Tsinghua University, BIGAI, Technical University of Munich
6 4 ./images/avalon's_game_of_thoughts_20231002.png Chain of Agents: Large Language Models Collaborating on Long-Context Tasks Yusen Zhang, Ruoxi Sun, Yanfei Chen, Tomas Pfister, Rui Zhang, Sercan Ö. Arik Addressing the challenge of effectively processing long contexts has become a critical issue for Large Language Models (LLMs). Two common strategies have emerged: 1) reducing the input length, such as retrieving relevant chunks by Retrieval-Augmented Generation (RAG), and 2) expanding the context window limit of LLMs. However, both strategies have drawbacks: input reduction has no guarantee of covering the part with needed information, while window extension struggles with focusing on the pertinent information for solving the task. To mitigate these limitations, we propose Chain-of-Agents (CoA), a novel framework that harnesses multi-agent collaboration through natural language to enable information aggregation and context reasoning across various LLMs over long-context tasks. CoA consists of multiple worker agents who sequentially communicate to handle different segmented portions of the text, followed by a manager agent who synthesizes these contributions into a coherent final output. CoA processes the entire input by interleaving reading and reasoning, and it mitigates long context focus issues by assigning each agent a short context. We perform comprehensive evaluation of CoA on a wide range of long-context tasks in question answering, summarization, and code completion, demonstrating significant improvements by up to 10% over strong baselines of RAG, Full-Context, and multi-agent LLMs. Penn State University, Google Cloud AI Research
7 5 ./images/chain_of_agents_large_20240604.png ChatCoder: Chat-based Refine Requirement Improves LLMs' Code Generation Zejun Wang, Jia Li, Ge Li, Zhi Jin Large language models have shown good performances in generat-ing code to meet human requirements. However, human require-ments expressed in natural languages can be vague, incomplete,and ambiguous, leading large language models to misunderstandhuman requirements and make mistakes. Worse, it is difficult for ahuman user to refine the requirement. To help human users refinetheir requirements and improve large language models’ code gen-eration performances, we propose ChatCoder: a method to refinethe requirements via chatting with large language models. We de-sign a chat scheme in which the large language models will guidethe human users to refine their expression of requirements to bemore precise, unambiguous, and complete than before. Experimentsshow that ChatCoder has improved existing large language models’performance by a large margin. Besides, ChatCoder has the advan-tage over refine-based methods and LLMs fine-tuned via humanresponse. Peking University
8 6 ./images/chatcoder_chat-based_refine_requirement_20231101.png ChatDev: Communicative Agents for Software Development Chen Qian, Wei Liu, Hongzhang Liu, Nuo Chen, Yufan Dang, Jiahao Li, Cheng Yang, Weize Chen, Yusheng Su, Xin Cong, Juyuan Xu, Dahai Li, Zhiyuan Liu, Maosong Sun Software development is a complex task thatnecessitates cooperation among multiple mem-bers with diverse skills. Numerous studies useddeep learning to improve specific phases in awaterfall model, such as design, coding, andtesting.However, the deep learning modelin each phase requires unique designs, lead-ing to technical inconsistencies across variousphases, which results in a fragmented and in-effective development process. In this paper,we introduce ChatDev, a chat-powered soft-ware development framework in which special-ized agents driven by large language models(LLMs) are guided in what to communicate(via chat chain) and how to communicate (viacommunicative dehallucination). These agentsactively contribute to the design, coding, andtesting phases through unified language-basedcommunication, with solutions derived fromtheir multi-turn dialogues. We found their uti-lization of natural language is advantageousfor system design, and communicating in pro-gramming language proves helpful in debug-ging. This paradigm demonstrates how linguis-tic communication facilitates multi-agent col-laboration, establishing language as a unify-ing bridge for autonomous task-solving amongLLM agents. The code and data are availableat https://github.com/OpenBMB/ChatDev. Tsinghua University, The University of Sydney, BUPT, Modelbest Inc.
9 7 ./images/chatdev_communicative_agents_for_20230716.png ChatEval: Towards Better LLM-based Evaluators through Multi-Agent Debate Chi-Min Chan, Weize Chen, Yusheng Su, Jianxuan Yu, Wei Xue, Shanghang Zhang, Jie Fu, Zhiyuan Liu Text evaluation has historically posed significant challenges, often demandingsubstantial labor and time cost. With the emergence of large language models(LLMs), researchers have explored LLMs’ potential as alternatives for humanevaluation. While these single-agent-based approaches show promise, experi-mental results suggest that further advancements are needed to bridge the gapbetween their current effectiveness and human-level evaluation quality. Recog-nizing that best practices of human evaluation processes often involve multiplehuman annotators collaborating in the evaluation, we resort to a multi-agent debateframework, moving beyond single-agent prompting strategies. The multi-agent-based approach enables a group of LLMs to synergize with an array of intelli-gent counterparts, harnessing their distinct capabilities and expertise to enhanceefficiency and effectiveness in handling intricate tasks. In this paper, we con-struct a multi-agent referee team called ChatEval to autonomously discuss andevaluate the quality of generated responses from different models on open-endedquestions and traditional natural language generation (NLG) tasks. We deriveinsights and lessons from practical scenarios where humans instigate group dis-cussions for brainstorming and propose different communication strategies withinChatEval...... Tsinghua University, Hong Kong University of Science and Technology, Peking University
10 8 ./images/chateval_towards_better_llm-based_20230814.png CoMM: Collaborative Multi-Agent, Multi-Reasoning-Path Prompting for Complex Problem Solving Pei Chen, Boran Han, Shuai Zhang Large Language Models (LLMs) have showngreat ability in solving traditional natural lan-guage tasks and elementary reasoning taskswith appropriate prompting techniques. How-ever, their ability is still limited in solving com-plicated science problems. In this work, weaim to push the upper bound of the reason-ing capability of LLMs by proposing a col-laborative multi-agent, multi-reasoning-path(CoMM) prompting framework. Specifically,we prompt LLMs to play different roles in aproblem-solving team, and encourage differ-ent role-play agents to collaboratively solvethe target task. In particular, we discover thatapplying different reasoning paths for differ-ent roles is an effective strategy to implementfew-shot prompting approaches in the multi-agent scenarios. Empirical results demonstratethe effectiveness of the proposed methods ontwo college-level science problems over com-petitive baselines. Our further analysis showsthe necessity of prompting LLMs to play dif-ferent roles or experts independently. We re-lease the code at: https://github.com/amazon-science/comm-prompt. Texas A&M University, Amazon Web Services
11 9 ./images/comm_collaborative_multi-agent,_multi-reasoning-path_20240426.png Describe, Explain, Plan and Select: Interactive Planning with Large Language Models Enables Open-World Multi-Task Agents Zihao Wang, Shaofei Cai, Guanzhou Chen, Anji Liu, Xiaojian Ma, Yitao Liang We investigate the challenge of task planning for multi-task embodied agents in open-world environments. Two main difficulties are identified: 1) executing plans in an open-world environment (e.g., Minecraft) necessitates accurate and multi-step reasoning due to the long-term nature of tasks, and 2) as vanilla planners do not consider how easy the current agent can achieve a given sub-task when ordering parallel sub-goals within a complicated plan, the resulting plan could be inefficient or even infeasible. To this end, we propose "Describe, Explain, Plan and Select" (DEPS), an interactive planning approach based on Large Language Models (LLMs). DEPS facilitates better error correction on initial LLM-generated plan by integrating description of the plan execution process and providing self-explanation of feedback when encountering failures during the extended planning phases. Furthermore, it includes a goal selector, which is a trainable module that ranks parallel candidate sub-goals based on the estimated steps of completion, consequently refining the initial plan. Our experiments mark the milestone of the first zero-shot multi-task agent that can robustly accomplish 70+ Minecraft tasks and nearly double the overall performances. Further testing reveals our method's general effectiveness in popularly adopted non-open-ended domains as well (i.e., ALFWorld and tabletop manipulation). The ablation and exploratory studies detail how our design beats the counterparts and provide a promising update on the 𝙾𝚋𝚝𝚊𝚒𝚗𝙳𝚒𝚊𝚖𝚘𝚗𝚍 grand challenge with our approach. Peking University, University of California Los Angeles, Beijing Institute for General Artificial Intelligence
12 10 ./images/describe,_explain,_plan_and_20230203.png Dynamic LLM-Agent Network: An LLM-agent Collaboration Framework with Agent Team Optimization Zijun Liu, Yanzhe Zhang, Peng Li, Yang Liu, Diyi Yang Large language model (LLM) agents have been shown effective on a wide rangeof tasks, and by ensembling multiple LLM agents, their performances could befurther improved. Existing approaches employ a fixed set of agents to interactwith each other in a static architecture, which limits their generalizability to vari-ous tasks and requires strong human prior in designing these agents. In this work,we propose to construct a strategic team of agents communicating in a dynamicinteraction architecture based on the task query. Specifically, we build a frame-work named Dynamic LLM-Agent Network (DyLAN) for LLM-agent collabora-tion on complicated tasks like reasoning and code generation. DyLAN enablesagents to interact for multiple rounds in a dynamic architecture with inference-time agent selection and an early-stopping mechanism to improve performanceand efficiency. We further design an automatic agent team optimization algorithmbased on an unsupervised metric termed Agent Importance Score, enabling theselection of best agents based on the contribution each agent makes. Empirically,we demonstrate that DyLAN performs well in both reasoning and code generationtasks with reasonable computational cost. DyLAN achieves 1 Tsinghua University, Georgia Tech, Stanford University
13 11 ./images/dynamic_llm-agent_network_an_20231003.png EconAgent: Large Language Model-Empowered Agents for Simulating Macroeconomic Activities Nian Li, Chen Gao, Mingyu Li, Yong Li, Qingmin Liao The advent of artificial intelligence has led to agrowing emphasis on data-driven modeling inmacroeconomics, with agent-based modeling(ABM) emerging as a prominent bottom-upsimulation paradigm. In ABM, agents (e.g.,households, firms) interact within a macroe-conomic environment, collectively generatingmarket dynamics. Existing agent modeling typ-ically employs predetermined rules or learning-based neural networks for decision-making.However, customizing each agent presents sig-nificant challenges, complicating the modelingof agent heterogeneity. Additionally, the in-fluence of multi-period market dynamics andmultifaceted macroeconomic factors are oftenoverlooked in decision-making processes. Inthis work, we introduce EconAgent, a largelanguage model-empowered agent with human-like characteristics for macroeconomic simu-lation. We first construct a simulation envi-ronment that incorporates various market dy-namics driven by agents’ decisions regardingwork and consumption. Through the perceptionmodule, we create heterogeneous agents withdistinct decision-making mechanisms.Fur-thermore, we model the impact of macroeco-nomic trends using a memory module, whichallows agents to reflect on past individual ex-periences and market dynamics. Simulationexperiments show that EconAgent can makerealistic decisions, leading to more reasonablemacroeconomic phenomena compared to exist-ing rule-based or learning-based agents. Ourcodes are released at https://github.com/tsinghua-fib-lab/ACL24-EconAgent. Tsinghua University
14 12 ./images/econagent_large_language_model-empowered_20231016.png Experiential Co-Learning of Software-Developing Agents Chen Qian, Yufan Dang, Jiahao Li, Wei Liu, Zihao Xie, Yifei Wang, Weize Chen, Cheng Yang, Xin Cong, Xiaoyin Che, Zhiyuan Liu, Maosong Sun Recent advancements in large language mod-els (LLMs) have brought significant changesto various domains, especially through LLM-driven autonomous agents. A representativescenario is in software development, whereLLM agents demonstrate efficient collabora-tion, task division, and assurance of softwarequality, markedly reducing the need for man-ual involvement. However, these agents fre-quently perform a variety of tasks indepen-dently, without benefiting from past experi-ences, which leads to repeated mistakes andinefficient attempts in multi-step task execu-tion. To this end, we introduce Experiential Co-Learning, a novel LLM-agent learning frame-work in which instructor and assistant agentsgather shortcut-oriented experiences from theirhistorical trajectories and use these past expe-riences for future task execution. The exten-sive experiments demonstrate that the frame-work enables agents to tackle unseen software-developing tasks more effectively. We antici-pate that our insights will guide LLM agentstowards enhanced autonomy and contributeto their evolutionary growth in cooperativelearning. The code and data are available athttps://github.com/OpenBMB/ChatDev. Tsinghua University, Dalian University of Technology, Beijing University of Posts and Telecommunications, Siemens
15 13 ./images/experiential_co-learning_of_software-developing_20231228.png Exploring Large Language Models for Communication Games: An Empirical Study on Werewolf Yuzhuang Xu, Shuo Wang, Peng Li, Fuwen Luo, Xiaolong Wang, Weidong Liu, Yang Liu Communication games, which we refer to asincomplete information games that heavily de-pend on natural language communication, holdsignificant research value in fields such as eco-nomics, social science, and artificial intelli-gence. In this work, we explore the problem ofhow to engage large language models (LLMs)in communication games, and in response, pro-pose a tuning-free framework. Our approachkeeps LLMs frozen, and relies on the retrievaland reflection on past communications and ex-periences for improvement. An empirical studyon the representative and widely-studied com-munication game, “Werewolf”, demonstratesthat our framework can effectively play Were-wolf game without tuning the parameters of theLLMs. More importantly, strategic behaviorsbegin to emerge in our experiments, suggest-ing that it will be a fruitful journey to engageLLMs in communication games and associateddomains. Tsinghua University, Zhongguancun Laboratory
16 14 ./images/exploring_large_language_models_20230909.png Facilitating Multi-Role and Multi-Behavior Collaboration of Large Language Models for Online Job Seeking and Recruiting Hongda Sun, Hongzhan Lin, Haiyu Yan, Chen Zhu, Yang Song, Xin Gao, Shuo Shang, Rui Yan The emergence of online recruitment services has revolutionizedthe traditional landscape of job seeking and recruitment, neces-sitating the development of high-quality industrial applicationsto improve person-job fitting. Existing methods generally rely onmodeling the latent semantics of resumes and job descriptions andlearning a matching function between them. Inspired by the pow-erful role-playing capabilities of Large Language Models (LLMs),we propose to introduce a mock interview process between LLM-played interviewers and candidates. The mock interview conver-sations can provide additional evidence for candidate evaluation,thereby augmenting traditional person-job fitting based solely onresumes and job descriptions. However, characterizing these tworoles in online recruitment still presents several challenges, suchas developing the skills to raise interview questions, formulatingappropriate answers, and evaluating two-sided fitness.To this end, we propose MockLLM, a novel applicable frameworkthat divides the person-job matching process into two modules:mock interview generation and two-sided evaluation in handshakeprotocol, jointly enhancing their performance through collaborativebehaviors between interviewers and candidates. We design a role-playing framework as a multi-role and multi-behavior paradigmto enable a single LLM agent to effectively behave with multiplefunctions for both parties...... Renmin University of China, BOSS Zhipin, King Abdullah University of Science and Technology, University of Electronic Science and Technology of China
17 15 ./images/facilitating_multi-role_and_multi-behavior_20240528.png GameGPT: Multi-agent Collaborative Framework for Game Development Dake Chen, Hanbin Wang, Yunhao Huo, Yuzhao Li, Haoyang Zhang The large language model (LLM) based agents have demonstrated their capacityto automate and expedite software development processes. In this paper, wefocus on game development and propose a multi-agent collaborative framework,dubbed GameGPT, to automate game development. While many studies havepinpointed hallucination as a primary roadblock for deploying LLMs in production,we identify another concern: redundancy. Our framework presents a series ofmethods to mitigate both concerns. These methods include dual collaboration andlayered approaches with several in-house lexicons, to mitigate the hallucinationand redundancy in the planning, task identification, and implementation phases.Furthermore, a decoupling approach is also introduced to achieve code generationwith better precision. AutoGame Research, X-Institute, University of Southern California
18 16 ./images/gamegpt_multi-agent_collaborative_framework_20231012.png Generative Agents: Interactive Simulacra of Human Behavior Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein Believable proxies of human behavior can empower interactive applications ranging from immersive environments to rehearsal spaces for interpersonal communication to prototyping tools. In this paper, we introduce generative agents--computational software agents that simulate believable human behavior. Generative agents wake up, cook breakfast, and head to work; artists paint, while authors write; they form opinions, notice each other, and initiate conversations; they remember and reflect on days past as they plan the next day. To enable generative agents, we describe an architecture that extends a large language model to store a complete record of the agent's experiences using natural language, synthesize those memories over time into higher-level reflections, and retrieve them dynamically to plan behavior. We instantiate generative agents to populate an interactive sandbox environment inspired by The Sims, where end users can interact with a small town of twenty five agents using natural language. In an evaluation, these generative agents produce believable individual and emergent social behaviors: for example, starting with only a single user-specified notion that one agent wants to throw a Valentine's Day party, the agents autonomously spread invitations to the party over the next two days, make new acquaintances, ask each other out on dates to the party, and coordinate to show up for the party together at the right time. We demonstrate through ablation that the components of our agent architecture--observation, planning, and reflection--each contribute critically to the believability of agent behavior. By fusing large language models with computational, interactive agents, this work introduces architectural and interaction patterns for enabling believable simulations of human behavior. Stanford University, Google Research, Google DeepMind
19 17 ./images/generative_agents_interactive_simulacra_20230407.png Improving Multi-Agent Debate with Sparse Communication Topology Yunxuan Li, Yibing Du, Jiageng Zhang, Le Hou, Peter Grabowski, Yeqing Li, Eugene Ie Multi-agent debate has proven effective in im-proving large language models quality for rea-soning and factuality tasks. While various role-playing strategies in multi-agent debates havebeen explored, in terms of the communica-tion among agents, existing approaches adopta brute force algorithm – each agent can com-municate with all other agents. In this paper,we systematically investigate the effect of com-munication connectivity in multi-agent systems.Our experiments on GPT and Mistral models re-veal that multi-agent debates leveraging sparsecommunication topology can achieve compara-ble or superior performance while significantlyreducing computational costs. Furthermore, weextend the multi-agent debate framework tomultimodal reasoning and alignment labelingtasks, showcasing its broad applicability andeffectiveness. Our findings underscore the im-portance of communication connectivity on en-hancing the efficiency and effectiveness of the“society of minds” approach. Google, Google DeepMind
20 18 ./images/improving_multi-agent_debate_with_20240617.png Iterative Experience Refinement of Software-Developing Agents Chen Qian, Jiahao Li, Yufan Dang, Wei Liu, YiFei Wang, Zihao Xie, Weize Chen, Cheng Yang, Yingli Zhang, Zhiyuan Liu, Maosong Sun Autonomous agents powered by large languagemodels (LLMs) show significant potential forachieving high autonomy in various scenar-ios such as software development. Recent re-search has shown that LLM agents can lever-age past experiences to reduce errors and en-hance efficiency. However, the static experi-ence paradigm, reliant on a fixed collection ofpast experiences acquired heuristically, lacksiterative refinement and thus hampers agents’adaptability. In this paper, we introduce the It-erative Experience Refinement framework, en-abling LLM agents to refine experiences itera-tively during task execution. We propose twofundamental patterns: the successive pattern,refining based on nearest experiences within atask batch, and the cumulative pattern, acquir-ing experiences across all previous task batches.Augmented with our heuristic experience elim-ination, the method prioritizes high-quality andfrequently-used experiences, effectively man-aging the experience space and enhancing effi-ciency. Extensive experiments show that whilethe successive pattern may yield superior re-sults, the cumulative pattern provides more sta-ble performance...... Tsinghua University, Dalian University of Technology, Beijing University of Posts and Telecommunications, Siemens
21 19 ./images/iterative_experience_refinement_of_20240507.png Language Agents as Optimizable Graphs Mingchen Zhuge, Wenyi Wang, Louis Kirsch, Francesco Faccio, Dmitrii Khizbullin, Jürgen Schmidhuber Various human-designed prompt engineering techniques have been proposed to improve problem solvers based on Large Language Models (LLMs), yielding many disparate code bases. We unify these approaches by describing LLM-based agents as computational graphs. The nodes implement functions to process multimodal data or query LLMs, and the edges describe the information flow between operations. Graphs can be recursively combined into larger composite graphs representing hierarchies of inter-agent collaboration (where edges connect operations of different agents). Our novel automatic graph optimizers (1) refine node-level LLM prompts (node optimization) and (2) improve agent orchestration by changing graph connectivity (edge optimization). Experiments demonstrate that our framework can be used to efficiently develop, integrate, and automatically improve various LLM agents. King Abdullah University of Science and Technology, The Swiss AI Lab IDSIA, USI, SUPSI
22 20 ./images/language_agents_as_optimizable_20240226.png Large Language Models are Diverse Role-Players for Summarization Evaluation Ning Wu, Ming Gong, Linjun Shou, Shining Liang, Daxin Jiang . Text summarization has a wide range of applications in many scenarios.The evaluation of the quality of the generated text is a complex problem. A bigchallenge to language evaluation is that there is a clear divergence between existingmetrics and human evaluation. A document summary’s quality can be assessedby human annotators on various criteria, both objective ones like grammar andcorrectness, and subjective ones like informativeness, succinctness, and appeal.Most of the automatic evaluation methods like BLUE/ROUGE may be not ableto adequately capture the above dimensions. In this paper, we propose a newevaluation framework based on LLMs, which provides a comprehensive evaluationframework by comparing generated text and reference text from both objective andsubjective aspects. First, we propose to model objective and subjective dimensionsof generated text based on roleplayers prompting mechanism. Furthermore, weintroduce a context-based prompting mechanism that is able to generate dynamicroleplayer profiles based on input context. Finally, we design a multi-roleplayerprompting technology based on batch prompting and integrate multiple outputsinto the final evaluation results. Experimental results on three real datasets forsummarization show that our model is highly competitive and has a very highconsistency with human annotators. Microsoft
23 21 ./images/large_language_models_are_20230327.png Learn to Disguise: Avoid Refusal Responses in LLM's Defense via a Multi-agent Attacker-Disguiser Game Qianqiao Xu, Zhiliang Tian, Hongyan Wu, Zhen Huang, Yiping Song, Feng Liu, Dongsheng Li With the enhanced performance of large models on natural language processingtasks, potential moral and ethical issues of large models arise. There exist ma-licious attackers who induce large models to jailbreak and generate informationcontaining illegal, privacy-invasive information through techniques such as promptengineering. As a result, large models counter malicious attackers’ attacks usingtechniques such as safety alignment. However, the strong defense mechanismof the large model through rejection replies is easily identified by attackers andused to strengthen attackers’ capabilities. In this paper, we propose a multi-agentattacker-disguiser game approach to achieve a weak defense mechanism that allowsthe large model to both safely reply to the attacker and hide the defense intent. First,we construct a multi-agent framework to simulate attack and defense scenarios,playing different roles to be responsible for attack, disguise, safety evaluation,and disguise evaluation tasks. After that, we design attack and disguise gamealgorithms to optimize the game strategies of the attacker and the disguiser and usethe curriculum learning process to strengthen the capabilities of the agents. Theexperiments verify that the method in this paper is more effective in strengtheningthe model’s ability to disguise the defense intent compared with other methods.Moreover, our approach can adapt any black-box large model to assist the model indefense and does not suffer from model version iterations. National University of Defense Technology, Guangdong University of Foreign Studies,
24 22 ./images/learn_to_disguise_avoid_20240403.png Leveraging Large Language Models for Collective Decision-Making Marios Papachristou, Longqi Yang, Chin-Chia Hsu In various work contexts, such as meeting scheduling, collaborating, and project planning, collective decision-making is essential but often challenging due to diverse individual preferences, varying work focuses, and power dynamics among members. To address this, we propose a system leveraging Large Language Models (LLMs) to facilitate group decision-making by managing conversations and balancing preferences among individuals. Our system aims to extract individual preferences from conversations and suggest options that satisfy the preferences of the members. We specifically apply this system to corporate meeting scheduling. We create synthetic employee profiles and simulate conversations at scale, leveraging LLMs to evaluate the system performance as a novel approach to conducting a user study. Our results indicate efficient coordination with reduced interactions between the members and the LLM-based system. The system refines and improves its proposed options over time, ensuring that many of the members' individual preferences are satisfied in an equitable way. Finally, we conduct a survey study involving human participants to assess our system's ability to aggregate preferences and reasoning about them. Our findings show that the system exhibits strong performance in both dimensions Cornell University, Microsoft
25 23 ./images/leveraging_large_language_models_20231103.png LLM-Based Agent Society Investigation: Collaboration and Confrontation in Avalon Gameplay Yihuai Lan, Zhiqiang Hu, Lei Wang, Yang Wang, Deheng Ye, Peilin Zhao, Ee-Peng Lim, Hui Xiong, Hao Wang This paper explores the open research prob-lem of understanding the social behaviors ofLLM-based agents. Using Avalon as a testbed,we employ system prompts to guide LLMagents in gameplay. While previous studieshave touched on gameplay with LLM agents,research on their social behaviors is lacking.We propose a novel framework, tailored forAvalon, features a multi-agent system facil-itating efficient communication and interac-tion. We evaluate its performance based ongame success and analyze LLM agents’ so-cial behaviors. Results affirm the framework’seffectiveness in creating adaptive agents andsuggest LLM-based agents’ potential in nav-igating dynamic social interactions. By ex-amining collaboration and confrontation be-haviors, we offer insights into this field’s re-search and applications.Our code is pub-licly available at https://github.com/3DAgentWorld/LLM-Game-Agent The Hong Kong University of Science and Technology (Guangzhou), Singapore University of Technology and Design, Singapore Management University, Verily Life Sciences, Tencent
26 24 ./images/llm-based_agent_society_investigation_20231023.png LongAgent: Scaling Language Models to 128k Context through Multi-Agent Collaboration Jun Zhao, Can Zu, Hao Xu, Yi Lu, Wei He, Yiwen Ding, Tao Gui, Qi Zhang, Xuanjing Huang Large language models (LLMs) have demon-strated impressive performance in understand-ing language and executing complex reasoningtasks. However, LLMs with long context win-dows have been notorious for their expensivetraining costs and high inference latency. Eventhe most advanced models such as GPT-4 andClaude2 often make mistakes when processinginputs of over 100k tokens, a phenomenon alsoknown as lost in the middle. In this paper,we propose LONGAGENT, a method basedon multi-agent collaboration, which scalesLLMs (e.g., LLaMA) to a context of 128K anddemonstrates potential superiority in long-textprocessing compared to GPT- Fudan University
27 25 ./images/longagent_scaling_language_models_20240218.png MetaAgents: Simulating Interactions of Human Behaviors for LLM-based Task-oriented Coordination via Collaborative Generative Agents Yuan Li, Yixuan Zhang, Lichao Sun Significant advancements have occurred in the application of Large LanguageModels (LLMs) for various tasks and social simulations. Despite this, their capac-ities to coordinate within task-oriented social contexts are under-explored. Suchcapabilities are crucial if LLMs are to effectively mimic human-like social be-havior and produce meaningful results. To bridge this gap, we introduce collab-orative generative agents, endowing LLM-based Agents with consistent behaviorpatterns and task-solving abilities. We situate these agents in a simulated job fairenvironment as a case study to scrutinize their coordination skills. We proposea novel framework that equips collaborative generative agents with human-likereasoning abilities and specialized skills. Our evaluation demonstrates that theseagents show promising performance. However, we also uncover limitations thathinder their effectiveness in more complex coordination tasks. Our work providesvaluable insights into the role and evolution of LLMs in task-oriented social sim-ulations. University of Cambridge, William & Mary, Lehigh University
28 26 ./images/metaagents_simulating_interactions_of_20231010.png MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework Sirui Hong, Mingchen Zhuge, Jonathan Chen, Xiawu Zheng, Yuheng Cheng, Ceyao Zhang, Jinlin Wang, Zili Wang, Steven Ka Shing Yau, Zijuan Lin, Liyang Zhou, Chenyu Ran, Lingfeng Xiao, Chenglin Wu, Jürgen Schmidhuber Remarkable progress has been made on automated problem solving through so-cieties of agents based on large language models (LLMs). Existing LLM-basedmulti-agent systems can already solve simple dialogue tasks. Solutions to morecomplex tasks, however, are complicated through logic inconsistencies due tocascading hallucinations caused by naively chaining LLMs. Here we introduceMetaGPT, an innovative meta-programming framework incorporating efficienthuman workflows into LLM-based multi-agent collaborations.MetaGPT en-codes Standardized Operating Procedures (SOPs) into prompt sequences for morestreamlined workflows, thus allowing agents with human-like domain expertiseto verify intermediate results and reduce errors. MetaGPT utilizes an assemblyline paradigm to assign diverse roles to various agents, efficiently breaking downcomplex tasks into subtasks involving many agents working together. On col-laborative software engineering benchmarks, MetaGPT generates more coherentsolutions than previous chat-based multi-agent systems. Our project can be foundat https://github.com/geekan/MetaGPT DeepWisdom, King Abdullah University of Science and Technology, Xiamen University, The Chinese University of Hong Kong (Shenzhen), Nanjing University, University of Pennsylvania University of California, Berkeley, The Swiss AI Lab IDSIA/USI/SUPSI
29 27 ./images/metagpt_meta_programming_for_20230801.png Mora: Enabling Generalist Video Generation via A Multi-Agent Framework Zhengqing Yuan, Ruoxi Chen, Zhaoxu Li, Haolong Jia, Lifang He, Chi Wang, Lichao Sun Sora is the first large-scale generalist video generation model that garnered significant attention across society. Since its launch by OpenAI in February 2024, no other video generation models have paralleled {Sora}'s performance or its capacity to support a broad spectrum of video generation tasks. Additionally, there are only a few fully published video generation models, with the majority being closed-source. To address this gap, this paper proposes a new multi-agent framework Mora, which incorporates several advanced visual AI agents to replicate generalist video generation demonstrated by Sora. In particular, Mora can utilize multiple visual agents and successfully mimic Sora's video generation capabilities in various tasks, such as (1) text-to-video generation, (2) text-conditional image-to-video generation, (3) extend generated videos, (4) video-to-video editing, (5) connect videos and (6) simulate digital worlds. Our extensive experimental results show that Mora achieves performance that is proximate to that of Sora in various tasks. However, there exists an obvious performance gap between our work and Sora when assessed holistically. In summary, we hope this project can guide the future trajectory of video generation through collaborative AI agents. Lehigh University, Microsoft Research
30 28 ./images/mora_enabling_generalist_video_20240320.png Multi-Agent Software Development through Cross-Team Collaboration Zhuoyun Du, Chen Qian, Wei Liu, Zihao Xie, Yifei Wang, Yufan Dang, Weize Chen, Cheng Yang The latest breakthroughs in Large LanguageModels (LLMs), e.g., ChatDev, have catalyzedprofound transformations, particularly throughmulti-agent collaboration for software devel-opment. LLM agents can collaborate in teamslike humans, and follow the waterfall modelto sequentially work on requirements analysis,development, review, testing, and other phasesto perform autonomous software generation.However, for an agent team, each phase in asingle development process yields only one pos-sible outcome. This results in the completionof only one development chain, thereby losingthe opportunity to explore multiple potentialdecision paths within the solution space. Con-sequently, this may lead to obtaining subop-timal results. To address this challenge, weintroduce Cross-Team Collaboration (CTC),a scalable multi-team framework that enablesorchestrated teams to jointly propose variousdecisions and communicate with their insightsin a cross-team collaboration environment forsuperior content generation. Experimental re-sults in software development reveal a notableincrease in quality compared to state-of-the-art baselines, underscoring the efficacy of ourframework. The significant improvements instory generation demonstrate the promisinggeneralization ability of our framework acrossvarious domains. We anticipate that our workwill guide LLM agents towards a cross-teamparadigm and contribute to their significantgrowth in but not limited to software devel-opment. The code and data will be available athttps://github.com/OpenBMB/ChatDev. Zhejiang University, Tsinghua University, Beijing University of Posts and Telecommunications
31 29 ./images/multi-agent_software_development_through_20240613.png MultiAgent Collaboration Attack: Investigating Adversarial Attacks in Large Language Model Collaborations via Debate Alfonso Amayuelas, Xianjun Yang, Antonis Antoniades, Wenyue Hua, Liangming Pan, William Wang Large Language Models (LLMs) have shownexceptional results on current benchmarkswhen working individually. The advancementin their capabilities, along with a reduction inparameter size and inference times, has facil-itated the use of these models as agents, en-abling interactions among multiple models toexecute complex tasks. Such collaborationsoffer several advantages, including the use ofspecialized models (e.g. coding), improvedconfidence through multiple computations, andenhanced divergent thinking, leading to morediverse outputs. Thus, the collaborative use oflanguage models is expected to grow signifi-cantly in the coming years. In this work, weevaluate the behavior of a network of modelscollaborating through debate under the influ-ence of an adversary. We introduce pertinentmetrics to assess the adversary’s effectiveness,focusing on system accuracy and model agree-ment. Our findings highlight the importanceof a model’s persuasive ability in influencingothers. Additionally, we explore inference-timemethods to generate more compelling argu-ments and evaluate the potential of prompt-based mitigation as a defensive strategy. UC Santa Barbara, Rutgers University
32 30 ./images/multiagent_collaboration_attack_investigating_20240620.png ReConcile: Round-Table Conference Improves Reasoning via Consensus among Diverse LLMs Justin Chih-Yao Chen, Swarnadeep Saha, Mohit Bansal Large Language Models (LLMs) still struggle with natural language reasoning tasks. Motivated by the society of minds (Minsky, 1988), we propose ReConcile, a multi-model multi-agent framework designed as a round table conference among diverse LLM agents. ReConcile enhances collaborative reasoning between LLM agents via multiple rounds of discussion, learning to convince other agents to improve their answers, and employing a confidence-weighted voting mechanism that leads to a better consensus. In each round, ReConcile initiates discussion between agents via a 'discussion prompt' that consists of (a) grouped answers and explanations generated by each agent in the previous round, (b) their confidence scores, and (c) demonstrations of answer-rectifying human explanations, used for convincing other agents. Experiments on seven benchmarks demonstrate that ReConcile significantly improves LLMs' reasoning -- both individually and as a team -- surpassing prior single-agent and multi-agent baselines by up to 11.4% and even outperforming GPT-4 on three datasets. ReConcile also flexibly incorporates different combinations of agents, including API-based, open-source, and domain-specific models, leading to an 8% improvement on MATH. Finally, we analyze the individual components of ReConcile, demonstrating that the diversity originating from different models is critical to its superior performance. UNC Chapel Hill
33 31 ./images/reconcile_round-table_conference_improves_20230922.png Rethinking the Bounds of LLM Reasoning: Are Multi-Agent Discussions the Key? Qineng Wang, Zihao Wang, Ying Su, Hanghang Tong, Yangqiu Song Recent progress in LLMs discussion suggeststhat multi-agent discussion improves the rea-soning abilities of LLMs. In this work, wereevaluate this claim through systematic experi-ments, where we propose a novel group discus-sion framework to enrich the set of discussionmechanisms. Interestingly, our results showthat a single-agent LLM with strong promptscan achieve almost the same performance asthe best existing discussion approach on a widerange of reasoning tasks and backbone LLMs.We observe that the multi-agent discussion per-forms better than a single agent only when thereis no demonstration in the prompt. Furtherstudy reveals the common interaction mecha-nisms of LLMs during the discussion. Zhejiang University, HKUST, UIUC
34 32 ./images/rethinking_the_bounds_of_20240228.png Scalable Multi-Robot Collaboration with Large Language Models: Centralized or Decentralized Systems? Yongchao Chen, Jacob Arkin, Yang Zhang, Nicholas Roy, Chuchu Fan — A flurry of recent work has demonstrated thatpre-trained large language models (LLMs) can be effectivetask planners for a variety of single-robot tasks. The planningperformance of LLMs is significantly improved via promptingtechniques, such as in-context learning or re-prompting withstate feedback, placing new importance on the token budgetfor the context window. An under-explored but natural nextdirection is to investigate LLMs as multi-robot task planners.However, long-horizon, heterogeneous multi-robot planningintroduces new challenges of coordination while also pushingup against the limits of context window length. It is thereforecritical to find token-efficient LLM planning frameworks thatare also able to reason about the complexities of multi-robotcoordination. In this work, we compare the task success rate andtoken efficiency of four multi-agent communication frameworks(centralized, decentralized, and two hybrid) as applied tofour coordination-dependent multi-agent 2D task scenarios forincreasing numbers of agents. We find that a hybrid frameworkachieves better task success rates across all four tasks andscales better to more agents. We further demonstrate the hybridframeworks in 3D simulations where the vision-to-text problemand dynamical errors are considered. Massachusetts Institute of Technology, Harvard University, MIT-IBM Watson AI Lab.
35 33 ./images/scalable_multi-robot_collaboration_with_20230927.png Scaling Large-Language-Model-based Multi-Agent Collaboration Chen Qian, Zihao Xie, Yifei Wang, Wei Liu, Yufan Dang, Zhuoyun Du, Weize Chen, Cheng Yang, Zhiyuan Liu, Maosong Sun Pioneering advancements in large languagemodel-powered agents have underscored thedesign pattern of multi-agent collaboration,demonstrating that collective intelligence cansurpass the capabilities of each individual. In-spired by the neural scaling law, which positsthat increasing neurons leads to emergent abil-ities, this study investigates whether a simi-lar principle applies to increasing agents inmulti-agent collaboration.Technically, wepropose ::multi-agent:collaboration::networks(MACNET), which utilize directed acyclicgraphs to organize agents and streamline theirinteractive reasoning via topological ordering,with solutions derived from their dialogues.Extensive experiments show that MACNETconsistently outperforms baseline models, en-abling effective agent collaboration across var-ious network topologies and supporting coop-eration among more than a thousand agents.Notably, we observed a small-world collabo-ration phenomenon, where topologies resem-bling small-world properties achieved supe-rior performance. Additionally, we identifieda collaborative scaling law, indicating thatnormalized solution quality follows a logisticgrowth pattern as scaling agents, with collabo-rative emergence occurring much earlier thanpreviously observed instances of neural emer-gence. The code and data will be available athttps://github.com/OpenBMB/ChatDev. Tsinghua University, Beijing University of Posts and Telecommunications
36 34 ./images/scaling_large-language-model-based_multi-agent_collaboration_20240611.png Self-Organized Agents: A LLM Multi-Agent Framework toward Ultra Large-Scale Code Generation and Optimization Yoichi Ishibashi, Yoshimasa Nishimura Recent advancements in automatic code gener-ation using large language model (LLM) agenthave brought us closer to the future of auto-mated software development. However, exist-ing single-agent approaches face limitationsin generating and improving large-scale, com-plex codebases due to constraints in contextlength. To tackle this challenge, we proposeSelf-Organized multi-Agent framework (SoA),a novel multi-agent framework that enables thescalable and efficient generation and optimiza-tion of large-scale code. In SoA, self-organizedagents operate independently to generate andmodify code components while seamlessly col-laborating to construct the overall codebase. Akey feature of our framework is the automaticmultiplication of agents based on problem com-plexity, allowing for dynamic scalability. Thisenables the overall code volume to be increasedindefinitely according to the number of agents,while the amount of code managed by eachagent remains constant. We evaluate SoA onthe HumanEval benchmark and demonstratethat, compared to a single-agent system, eachagent in SoA handles significantly less code,yet the overall generated code is substantiallygreater. Moreover, SoA surpasses the powerfulsingle-agent baseline by 5%...... TsukushiAI
37 35 ./images/self-organized_agents_a_llm_20240402.png StrategyLLM: Large Language Models as Strategy Generators, Executors, Optimizers, and Evaluators for Problem Solving Chang Gao, Haiyun Jiang, Deng Cai, Shuming Shi, Wai Lam Most existing prompting methods suffer from the issues of generalizability and consistency, as they often rely on instance-specific solutions that may not be applicable to other instances and lack task-level consistency across the selected few-shot examples. To address these limitations, we propose a comprehensive framework, StrategyLLM, allowing LLMs to perform inductive reasoning, deriving general strategies from specific task instances, and deductive reasoning, applying these general strategies to particular task examples, for constructing generalizable and consistent few-shot prompts. It employs four LLM-based agents: strategy generator, executor, optimizer, and evaluator, working together to generate, evaluate, and select promising strategies for a given task. Experimental results demonstrate that StrategyLLM outperforms the competitive baseline CoT-SC that requires human-annotated solutions on 13 datasets across 4 challenging tasks without human involvement, including math reasoning (34.2\% → 38.8\%), commonsense reasoning (70.3\% → 72.5\%), algorithmic reasoning (73.7\% → 85.0\%), and symbolic reasoning (30.0\% → 79.2\%). Further analysis reveals that StrategyLLM is applicable to various LLMs and demonstrates advantages across numerous scenarios. The Chinese University of Hong Kong, Sun Yat-sen University, Tencent AI Lab
38 36 ./images/strategyllm_large_language_models_20231115.png TraveLER: A Multi-LMM Agent Framework for Video Question-Answering Chuyi Shang, Amos You, Sanjay Subramanian, Trevor Darrell, Roei Herzig Recently, Large Multimodal Models (LMMs) have made significant progressin video question-answering using a frame-wise approach by leveraginglarge-scale, image-based pretraining in a zero-shot manner. While image-based methods for videos have shown impressive performance, a currentlimitation is that they often overlook how key timestamps are selected andcannot adjust when incorrect timestamps are identified. Moreover, they areunable to extract details relevant to the question, instead providing generaldescriptions of the frame. To overcome this, we design a multi-LMM agentframework that travels along the video, iteratively collecting relevant in-formation from keyframes through interactive question-asking until thereis sufficient information to answer the question. Specifically, we proposeTraveLER, a model that can create a plan to “Traverse” through the video,ask questions about individual frames to “Locate” and store key informa-tion, and then “Evaluate” if there is enough information to answer thequestion. Finally, if there is not enough information, our method is able to“Replan” based on its collected knowledge. Through extensive experiments,we find that the proposed TraveLER approach improves performance onseveral video question-answering benchmarks, such as NExT-QA, STAR,and Perception Test, without the need to fine-tune on specific datasets. University of California, Berkeley
39 37 ./images/traveler_a_multi-lmm_agent_20240401.png Unleashing the Emergent Cognitive Synergy in Large Language Models: A Task-Solving Agent through Multi-Persona Self-Collaboration Zhenhailong Wang, Shaoguang Mao, Wenshan Wu, Tao Ge, Furu Wei, Heng Ji Human intelligence thrives on cognitive syn-ergy, where collaboration among differentminds yield superior outcomes compared to iso-lated individuals. In this work, we propose SoloPerformance Prompting (SPP), which trans-forms a single LLM into a cognitive synergistby engaging in multi-turn self-collaborationwith multiple personas.A cognitive syner-gist is an intelligent agent that collaborativelycombines multiple minds’ strengths and knowl-edge to enhance problem-solving in complextasks. By dynamically identifying and simu-lating different personas based on task inputs,SPP unleashes the potential of cognitive syn-ergy in LLMs. Our in-depth analysis showsthat assigning multiple fine-grained personasin LLMs improves problem-solving abilitiescompared to using a single or fixed numberof personas. We evaluate SPP on three chal-lenging tasks: Trivia Creative Writing, Code-names Collaborative, and Logic Grid Puzzle,encompassing both knowledge-intensive andreasoning-intensive types. Unlike previousworks, such as Chain-of-Thought, that solelyenhance the reasoning abilities in LLMs, ex-perimental results demonstrate that SPP effec-tively reduces factual hallucination, and main-tains strong reasoning capabilities. Addition-ally, comparative experiments show that cog-nitive synergy only emerges in GPT-4 anddoes not appear in less capable models, suchas GPT- University of Illinois Urbana-Champaign, Microsoft Research Asia
40 38 ./images/unleashing_the_emergent_cognitive_20230711.png User Behavior Simulation with Large Language Model based Agents Lei Wang, Jingsen Zhang, Hao Yang, Zhiyuan Chen, Jiakai Tang, Zeyu Zhang, Xu Chen, Yankai Lin, Ruihua Song, Wayne Xin Zhao, Jun Xu, Zhicheng Dou, Jun Wang, Ji-Rong Wen Simulating high quality user behavior data has always been a fundamental problem in human-centered applications, where the major difficulty originates from the intricate mechanism of human decision process. Recently, substantial evidences have suggested that by learning huge amounts of web knowledge, large language models (LLMs) can achieve human-like intelligence. We believe these models can provide significant opportunities to more believable user behavior simulation. To inspire such direction, we propose an LLM-based agent framework and design a sandbox environment to simulate real user behaviors. Based on extensive experiments, we find that the simulated behaviors of our method are very close to the ones of real humans. Concerning potential applications, we simulate and study two social phenomenons including (1) information cocoons and (2) user conformity behaviors. This research provides novel simulation paradigms for human-centered applications. Renmin University of China, Beijing Key Laboratory of Big Data Management and Analysis Methods, University College London
41 39 ./images/user_behavior_simulation_with_20230605.png War and Peace (WarAgent): Large Language Model-based Multi-Agent Simulation of World Wars Wenyue Hua, Lizhou Fan, Lingyao Li, Kai Mei, Jianchao Ji, Yingqiang Ge, Libby Hemphill, Yongfeng Zhang Can we avoid wars at the crossroads of history? This question has been pursued byindividuals, scholars, policymakers, and organizations throughout human history.In this research, we attempt to answer the question based on the recent advancesof Artificial Intelligence (AI) and Large Language Models (LLMs). We proposeWarAgent, an LLM-powered multi-agent AI system, to simulate the participatingcountries, their decisions, and the consequences, in historical international conflicts,including the World War I (WWI), the World War II (WWII), and the WarringStates Period (WSP) in Ancient China. By evaluating the simulation effectiveness,we examine the advancements and limitations of cutting-edge AI systems’ abilitiesin studying complex collective human behaviors such as international conflictsunder diverse settings. In these simulations, the emergent interactions amongagents also offer a novel perspective for examining the triggers and conditions thatlead to war. Our findings offer data-driven and AI-augmented insights that canredefine how we approach conflict resolution and peacekeeping strategies. Theimplications stretch beyond historical analysis, offering a blueprint for using AI tounderstand human history and possibly prevent future international conflicts. Codeand data are available at https://github.com/agiresearch/WarAgent. Rutgers University
42 40 ./images/war_and_peace_(waragent)_20231128.png To be Continued... Your Contributions are Welcome!

94
MultiAgentEbook/book_organization/script.js Executable file
View File

@ -0,0 +1,94 @@
document.addEventListener("DOMContentLoaded", function() {
const csvFilePath = './book_organization/data.csv';
function loadCSV(filePath) {
return fetch(filePath)
.then(response => response.text())
.then(text => Papa.parse(text, { header: true }).data);
}
function createFlipBook(pages) {
const container = document.getElementById('flip_book_container');
const numPages = pages.length;
let flipBookHTML = '';
let style = document.createElement('style');
let css = '';
flipBookHTML += `<input type="checkbox" id="cover_checkbox">\n`;
for (let i = 0; i < numPages - 1; i++) {
flipBookHTML += `<input type="checkbox" id="page${i + 1}_checkbox">\n`;
}
flipBookHTML += `<div id="flip_book">\n`;
flipBookHTML += `<div class="front_cover">
<label for="cover_checkbox" id="cover">
<img src="./images/2.png" alt="Book Cover" class="cover_image">
</label>
</div>`
for (let i = 0; i < numPages - 1; i++) {
console.log(i)
const page = pages[i];
const pageIndex = i + 1;
flipBookHTML += `
<div class="page" id="page${pageIndex}">
<div class="front_page">
<label for="page${pageIndex}_checkbox"></label>
<img class="back_content" src="${page.image_path}" alt="Back content">
</div>
<div class="back_page">
<label for="page${pageIndex}_checkbox"></label>
<img class="edge_shading" src="./images/back_page_edge_shading.png" alt="Back page edge shading">
<div class="text_content">
<h1>${page.title}</h1>
<p class="author">${page.author}</p>
<p class="author">${page.affiliation}</p>
<div class="text_content_summary"><p class="summary">${page.summary}</p></div>
</div>
</div>
</div>\n`;
css += `
#page${pageIndex} {
z-index: ${numPages - i};
}
#page${pageIndex}_checkbox:checked~#flip_book #page${pageIndex} {
transform: rotateY(-180deg);
z-index: ${i + 1};
}\n`;
}
flipBookHTML += `<div class="back_cover">
<img src="./images/2a.png" alt="Back Cover" class="cover_image">
</div>`;
container.innerHTML = flipBookHTML;
style.innerHTML = css;
document.head.appendChild(style);
const md = window.markdownit();
const summaryElements = document.querySelectorAll('.summary');
summaryElements.forEach(el => {
el.innerHTML = md.render(el.textContent);
});
}
loadCSV(csvFilePath).then(pages => {
createFlipBook(pages);
});
});

23
MultiAgentEbook/book_organization_index.html Executable file
View File

@ -0,0 +1,23 @@
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Flip Book</title>
<link rel="stylesheet" href="./book_style.css">
<style>
body {
background-color: transparent;
}
</style>
</head>
<body>
<div id="flip_book_container"></div>
<script src="https://cdnjs.cloudflare.com/ajax/libs/PapaParse/5.3.0/papaparse.min.js"></script>
<script src="https://cdn.jsdelivr.net/npm/markdown-it/dist/markdown-it.min.js"></script>
<script src="./book_organization/script.js"></script>
</body>
</html>

34
MultiAgentEbook/book_simulation/data.csv Executable file
View File

@ -0,0 +1,34 @@
,image_path,title,author,summary,affiliation
0,./images/4d.png,(Perhaps) Beyond Human Translation: Harnessing Multi-Agent Collaboration for Translating Ultra-Long Literary Texts,"Minghao Wu, Yulin Yuan, Gholamreza Haffari, Longyue Wang","Recent advancements in machine translation (MT) have significantly enhancedtranslation quality across various domains. However, the translation of literarytexts remains a formidable challenge due to their complex language, figurative ex-pressions, and cultural nuances. In this work, we introduce a novel multi-agentframework based on large language models (LLMs) for literary translation, im-plemented as a company called TRANSAGENTS, which mirrors traditional trans-lation publication process by leveraging the collective capabilities of multipleagents, to address the intricate demands of translating literary works. To evaluatethe effectiveness of our system, we propose two innovative evaluation strategies:Monolingual Human Preference (MHP) and Bilingual LLM Preference (BLP).MHP assesses translations from the perspective of monolingual readers of the tar-get language, while BLP uses advanced LLMs to compare translations directlywith the original texts. Empirical findings indicate that despite lower d-BLEUscores, translations from TRANSAGENTS are preferred by both human evalua-tors and LLMs over human-written references, particularly in genres requiringdomain-specific knowledge. We also highlight the strengths and limitations ofTRANSAGENTS through case studies and suggests directions for future research.","Monash University, University of Macau, Tencent AI Lab"
1,./images/(perhaps)_beyond_human_translation_20240520.png,Agent Hospital: A Simulacrum of Hospital with Evolvable Medical Agents,"Junkai Li, Siyu Wang, Meng Zhang, Weitao Li, Yunghwei Lai, Xinhui Kang, Weizhi Ma, Yang Liu","In this paper, we introduce a simulacrum of hospital called Agent Hospital that simulates theentire process of treating illness. All patients, nurses, and doctors are autonomous agents powered bylarge language models (LLMs). Our central goal is to enable a doctor agent to learn how to treat illnesswithin the simulacrum. To do so, we propose a method called MedAgent-Zero. As the simulacrum cansimulate disease onset and progression based on knowledge bases and LLMs, doctor agents can keepaccumulating experience from both successful and unsuccessful cases. Simulation experiments show thatthe treatment performance of doctor agents consistently improves on various tasks. More interestingly,the knowledge the doctor agents have acquired in Agent Hospital is applicable to real-world medicarebenchmarks. After treating around ten thousand patients (real-world doctors may take over two years),the evolved doctor agent achieves a state-of-the-art accuracy of 9",Tsinghua University
2,./images/agent_hospital_a_simulacrum_20240505.png,AgentCF: Collaborative Learning with Autonomous Language Agents for Recommender Systems,"Junjie Zhang, Yupeng Hou, Ruobing Xie, Wenqi Sun, Julian McAuley, Wayne Xin Zhao, Leyu Lin, Ji-Rong Wen","Recently, there has been an emergence of employing LLM-poweredagents as believable human proxies, based on their remarkabledecision-making capability. However, existing studies mainly focuson simulating human dialogue. Human non-verbal behaviors, suchas item clicking in recommender systems, although implicitly ex-hibiting user preferences and could enhance the modeling of users,have not been deeply explored. The main reasons lie in the gapbetween language modeling and behavior modeling, as well as theincomprehension of LLMs about user-item relations.To address this issue, we propose AgentCF for simulating user-item interactions in recommender systems through agent-basedcollaborative filtering. We creatively consider not only users butalso items as agents, and develop a collaborative learning approachthat optimizes both kinds of agents together. Specifically, at eachtime step, we first prompt the user and item agents to interact au-tonomously. Then, based on the disparities between the agents’decisions and real-world interaction records, user and item agentsare prompted to reflect on and adjust the misleading simulationscollaboratively, thereby modeling their two-sided relations. The op-timized agents can also propagate their preferences to other agentsin subsequent interactions, implicitly capturing the collaborative fil-tering idea. Overall, the optimized agents exhibit diverse interactionbehaviors within our framework, including user-item, user-user,item-item, and collective interactions. The results show that theseagents can demonstrate personalized behaviors akin to those of real-world individuals, sparking the development of next-generationuser behavior simulation.","Renmin University of China, UC San Diego, Tencent"
3,./images/agentcf_collaborative_learning_with_20231013.png,AgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors,"Weize Chen, Yusheng Su, Jingwei Zuo, Cheng Yang, Chenfei Yuan, Chi-Min Chan, Heyang Yu, Yaxi Lu, Yi-Hsin Hung, Chen Qian, Yujia Qin, Xin Cong, Ruobing Xie, Zhiyuan Liu, Maosong Sun, Jie Zhou","Autonomous agents empowered by Large Language Models (LLMs) have under-gone significant improvements, enabling them to generalize across a broad spec-trum of tasks. However, in real-world scenarios, cooperation among individuals isoften required to enhance the efficiency and effectiveness of task accomplishment.Hence, inspired by human group dynamics, we propose a multi-agent frameworkAGENTVERSE that can effectively orchestrate a collaborative group of expert agentsas a greater-than-the-sum-of-its-parts system. Our experiments demonstrate thatAGENTVERSE can proficiently deploy multi-agent groups that outperform a singleagent. Extensive experiments on text understanding, reasoning, coding, tool utiliza-tion, and embodied AI confirm the effectiveness of AGENTVERSE. Moreover, ouranalysis of agent interactions within AGENTVERSE reveals the emergence of spe-cific collaborative behaviors, contributing to heightened group efficiency. Our codehas been released at https://github.com/OpenBMB/AgentVerse/.","Tsinghua University, Beijing University of Posts and Telecommunications, Tencent Inc."
4,./images/agentverse_facilitating_multi-agent_collaboration_20230821.png,AI Hospital: Interactive Evaluation and Collaboration of LLMs as Intern Doctors for Clinical Diagnosis,"Zhihao Fan, Jialong Tang, Wei Chen, Siyuan Wang, Zhongyu Wei, Jun Xi, Fei Huang, Jingren Zhou","The incorporation of Large Language Models(LLMs) in healthcare marks a significant ad-vancement. However, the application has pre-dominantly been limited to discriminative andquestion-answering tasks, which does not fullyleverage their interactive potential. To addressthis limitation, our paper presents AI Hospital,a framework designed to build a real-time in-teractive diagnosis environment. To simulatethe procedure, we collect high-quality medicalrecords to create patient, examiner, and medicaldirector agents. AI Hospital is then utilized forthe interactive evaluation and collaboration ofLLMs. Initially, we create a Multi-View Medi-cal Evaluation (MVME) benchmark where vari-ous LLMs serve as intern doctors for interactivediagnosis. Subsequently, to improve diagnosticaccuracy, we introduce a collaborative mech-anism that involves iterative discussions anda dispute resolution process under the supervi-sion of the medical director. In our experiments,we validate the reliability of AI Hospital. Theresults not only explore the feasibility of applyLLMs in clinical consultation but also confirmthe effectiveness of the dispute resolution fo-cused collaboration method.","Alibaba Inc., Huazhong University of Science and Technology, Fudan University"
5,./images/ai_hospital_interactive_evaluation_20240215.png,Are you in a Masquerade? Exploring the Behavior and Impact of Large Language Model Driven Social Bots in Online Social Networks,"Siyu Li, Jin Yang, Kui Zhao","As the capabilities of Large Language Models (LLMs) emerge, they not only assist in accomplishing traditional tasks within more efficient paradigms but also stimulate the evolution of social bots. Researchers have begun exploring the implementation of LLMs as the driving core of social bots, enabling more efficient and user-friendly completion of tasks like profile completion, social behavior decision-making, and social content generation. However, there is currently a lack of systematic research on the behavioral characteristics of LLMs-driven social bots and their impact on social networks. We have curated data from Chirper, a Twitter-like social network populated by LLMs-driven social bots and embarked on an exploratory study. Our findings indicate that: (1) LLMs-driven social bots possess enhanced individual-level camouflage while exhibiting certain collective characteristics; (2) these bots have the ability to exert influence on online communities through toxic behaviors; (3) existing detection methods are applicable to the activity environment of LLMs-driven social bots but may be subject to certain limitations in effectiveness. Moreover, we have organized the data collected in our study into the Masquerade-23 dataset, which we have publicly released, thus addressing the data void in the subfield of LLMs-driven social bots behavior datasets. Our research outcomes provide primary insights for the research and governance of LLMs-driven social bots within the research community.",Sichuan University
6,./images/are_you_in_a_20230719.png,BattleAgent: Multi-modal Dynamic Emulation on Historical Battles to Complement Historical Analysis,"Shuhang Lin, Wenyue Hua, Lingyao Li, Che-Jui Chang, Lizhou Fan, Jianchao Ji, Hang Hua, Mingyu Jin, Jiebo Luo, Yongfeng Zhang","This paper presents BattleAgent, a detailed emulation demonstration system thatcombines the Large Vision-Language Model (VLM) and Multi-Agent System(MAS). This novel system aims to simulate complex dynamic interactions amongmultiple agents, as well as between agents and their environments, over a period oftime. It emulates both the decision-making processes of leaders and the viewpointsof ordinary participants, such as soldiers. The emulation showcases the currentcapabilities of agents, featuring fine-grained multi-modal interactions betweenagents and landscapes. It develops customizable agent structures to meet specificsituational requirements, for example, a variety of battle-related activities likescouting and trench digging. These components collaborate to recreate historicalevents in a lively and comprehensive manner while offering insights into thethoughts and feelings of individuals from diverse viewpoints. The technologicalfoundations of BattleAgent establish detailed and immersive settings for historicalbattles, enabling individual agents to partake in, observe, and dynamically respondto evolving battle scenarios. This methodology holds the potential to substantiallydeepen our understanding of historical events, particularly through individualaccounts. Such initiatives can also aid historical research, as conventional historicalnarratives often lack documentation and prioritize the perspectives of decision-makers, thereby overlooking the experiences of ordinary individuals. This biaseddocumentation results in a considerable gap in our historical understanding, as manystories remain untold......","Rutgers University, University of Michigan, University of Rochester"
7,./images/battleagent_multi-modal_dynamic_emulation_20240423.png,Can Large Language Model Agents Simulate Human Trust Behaviors?,"Chengxing Xie, Canyu Chen, Feiran Jia, Ziyu Ye, Kai Shu, Adel Bibi, Ziniu Hu, Philip Torr, Bernard Ghanem, Guohao Li","Large Language Model (LLM) agents have beenincreasingly adopted as simulation tools to modelhumans in applications such as social science.However, one fundamental question remains: canLLM agents really simulate human behaviors? Inthis paper, we focus on one of the most criticalbehaviors in human interactions, trust, and aim toinvestigate whether or not LLM agents can sim-ulate human trust behaviors. We first find thatLLM agents generally exhibit trust behaviors, re-ferred to as agent trust, under the framework ofTrust Games, which are widely recognized in be-havioral economics. Then, we discover that LLMagents can have high behavioral alignment withhumans regarding trust behaviors, particularly forGPT-4, indicating the feasibility to simulate hu-man trust behaviors with LLM agents. In addition,we probe into the biases in agent trust and thedifferences in agent trust towards agents and hu-mans. We also explore the intrinsic properties ofagent trust under conditions including advancedreasoning strategies and external manipulations.We further offer important implications of ourdiscoveries for various scenarios where trust isparamount. Our study provides new insights intothe behaviors of LLM agents and the fundamentalanalogy between LLMs and humans.","KAUST, Illinois Institute of Technology, Pennsylvania State University, The University of Chicago, University of Oxford, California Institute of Technology"
8,./images/can_large_language_model_20240207.png,ChatDev: Communicative Agents for Software Development,"Chen Qian, Wei Liu, Hongzhang Liu, Nuo Chen, Yufan Dang, Jiahao Li, Cheng Yang, Weize Chen, Yusheng Su, Xin Cong, Juyuan Xu, Dahai Li, Zhiyuan Liu, Maosong Sun","Software development is a complex task thatnecessitates cooperation among multiple mem-bers with diverse skills. Numerous studies useddeep learning to improve specific phases in awaterfall model, such as design, coding, andtesting.However, the deep learning modelin each phase requires unique designs, lead-ing to technical inconsistencies across variousphases, which results in a fragmented and in-effective development process. In this paper,we introduce ChatDev, a chat-powered soft-ware development framework in which special-ized agents driven by large language models(LLMs) are guided in what to communicate(via chat chain) and how to communicate (viacommunicative dehallucination). These agentsactively contribute to the design, coding, andtesting phases through unified language-basedcommunication, with solutions derived fromtheir multi-turn dialogues. We found their uti-lization of natural language is advantageousfor system design, and communicating in pro-gramming language proves helpful in debug-ging. This paradigm demonstrates how linguis-tic communication facilitates multi-agent col-laboration, establishing language as a unify-ing bridge for autonomous task-solving amongLLM agents. The code and data are availableat https://github.com/OpenBMB/ChatDev.","Tsinghua University, The University of Sydney, BUPT, Modelbest Inc."
9,./images/chatdev_communicative_agents_for_20230716.png,CompeteAI: Understanding the Competition Dynamics in Large Language Model-based Agents,"Qinlin Zhao, Jindong Wang, Yixuan Zhang, Yiqiao Jin, Kaijie Zhu, Hao Chen, Xing Xie","Large language models (LLMs) have been widelyused as agents to complete different tasks, suchas personal assistance or event planning. Whilemost of the work has focused on cooperationand collaboration between agents, little workexplores competition, another important mech-anism that promotes the development of soci-ety and economy. In this paper, we seek to ex-amine the competition dynamics in LLM-basedagents. We first propose a general framework forstudying the competition between agents. Then,we implement a practical competitive environ-ment using GPT-4 to simulate a virtual town withtwo types of agents, including restaurant agentsand customer agents. Specifically, the restaurantagents compete with each other to attract morecustomers, where competition encourages themto transform, such as cultivating new operatingstrategies. Simulation experiments reveal severalinteresting findings at the micro and macro lev-els, which align well with existing market andsociological theories. We hope that the frame-work and environment can be a promising testbedto study the competition that fosters understand-ing of society. Code is available at: https://github.com/microsoft/competeai.","University of Science and Technology of China, Microsoft Research, William & Mary, Georgia Institute of Technology, Carnegie Mellon University"
10,./images/competeai_understanding_the_competition_20231026.png,EconAgent: Large Language Model-Empowered Agents for Simulating Macroeconomic Activities,"Nian Li, Chen Gao, Mingyu Li, Yong Li, Qingmin Liao","The advent of artificial intelligence has led to agrowing emphasis on data-driven modeling inmacroeconomics, with agent-based modeling(ABM) emerging as a prominent bottom-upsimulation paradigm. In ABM, agents (e.g.,households, firms) interact within a macroe-conomic environment, collectively generatingmarket dynamics. Existing agent modeling typ-ically employs predetermined rules or learning-based neural networks for decision-making.However, customizing each agent presents sig-nificant challenges, complicating the modelingof agent heterogeneity. Additionally, the in-fluence of multi-period market dynamics andmultifaceted macroeconomic factors are oftenoverlooked in decision-making processes. Inthis work, we introduce EconAgent, a largelanguage model-empowered agent with human-like characteristics for macroeconomic simu-lation. We first construct a simulation envi-ronment that incorporates various market dy-namics driven by agents’ decisions regardingwork and consumption. Through the perceptionmodule, we create heterogeneous agents withdistinct decision-making mechanisms.Fur-thermore, we model the impact of macroeco-nomic trends using a memory module, whichallows agents to reflect on past individual ex-periences and market dynamics. Simulationexperiments show that EconAgent can makerealistic decisions, leading to more reasonablemacroeconomic phenomena compared to exist-ing rule-based or learning-based agents. Ourcodes are released at https://github.com/tsinghua-fib-lab/ACL24-EconAgent.",Tsinghua University
11,./images/econagent_large_language_model-empowered_20231016.png,Epidemic Modeling with Generative Agents,"Ross Williams, Niyousha Hosseinichimeh, Aritra Majumdar, Navid Ghaffarzadegan","This study offers a new paradigm of individual-level modeling to address the grand challenge of incorporating human behavior in epidemic models. Using generative artificial intelligence in an agent-based epidemic model, each agent is empowered to make its own reasonings and decisions via connecting to a large language model such as ChatGPT. Through various simulation experiments, we present compelling evidence that generative agents mimic real-world behaviors such as quarantining when sick and self-isolation when cases rise. Collectively, the agents demonstrate patterns akin to multiple waves observed in recent pandemics followed by an endemic period. Moreover, the agents successfully flatten the epidemic curve. This study creates potential to improve dynamic system modeling by offering a way to represent human brain, reasoning, and decision making.",Virginia Tech
12,./images/epidemic_modeling_with_generative_20230711.png,Exploring Collaboration Mechanisms for LLM Agents: A Social Psychology View,"Jintian Zhang, Xin Xu, Ningyu Zhang, Ruibo Liu, Bryan Hooi, Shumin Deng","As Natural Language Processing (NLP) sys-tems are increasingly employed in intricate so-cial environments, a pressing query emerges:Can these NLP systems mirror human-esquecollaborative intelligence, in a multi-agent so-ciety consisting of multiple large language mod-els (LLMs)? This paper probes the collabora-tion mechanisms among contemporary NLPsystems by melding practical experiments withtheoretical insights. We fabricate four unique‘societies’ comprised of LLM agents, whereeach agent is characterized by a specific ‘trait’(easy-going or overconfident) and engages incollaboration with a distinct ‘thinking pattern’(debate or reflection).Through evaluatingthese multi-agent societies on three benchmarkdatasets, we discern that certain collaborativestrategies not only outshine previous top-tierapproaches but also optimize efficiency (usingfewer API tokens). Moreover, our results fur-ther illustrate that LLM agents manifest human-like social behaviors, such as conformity andconsensus reaching, mirroring foundational so-cial psychology theories. In conclusion, weintegrate insights from social psychology tocontextualize the collaboration of LLM agents,inspiring further investigations into the collab-oration mechanism for LLMs. We have sharedour code and datasets1, hoping to catalyze fur-ther research in this promising avenue.","Zhejiang University, National University of Singapore, NUS-NCS Joint Lab, Google DeepMind"
13,./images/exploring_collaboration_mechanisms_for_20231003.png,Generative Agents: Interactive Simulacra of Human Behavior,"Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein","Believable proxies of human behavior can empower interactive applications ranging from immersive environments to rehearsal spaces for interpersonal communication to prototyping tools. In this paper, we introduce generative agents--computational software agents that simulate believable human behavior. Generative agents wake up, cook breakfast, and head to work; artists paint, while authors write; they form opinions, notice each other, and initiate conversations; they remember and reflect on days past as they plan the next day. To enable generative agents, we describe an architecture that extends a large language model to store a complete record of the agent's experiences using natural language, synthesize those memories over time into higher-level reflections, and retrieve them dynamically to plan behavior. We instantiate generative agents to populate an interactive sandbox environment inspired by The Sims, where end users can interact with a small town of twenty five agents using natural language. In an evaluation, these generative agents produce believable individual and emergent social behaviors: for example, starting with only a single user-specified notion that one agent wants to throw a Valentine's Day party, the agents autonomously spread invitations to the party over the next two days, make new acquaintances, ask each other out on dates to the party, and coordinate to show up for the party together at the right time. We demonstrate through ablation that the components of our agent architecture--observation, planning, and reflection--each contribute critically to the believability of agent behavior. By fusing large language models with computational, interactive agents, this work introduces architectural and interaction patterns for enabling believable simulations of human behavior.","Stanford University, Google Research, Google DeepMind"
14,./images/generative_agents_interactive_simulacra_20230407.png,Humanoid Agents: Platform for Simulating Human-like Generative Agents,"Zhilin Wang, Yu Ying Chiu, Yu Cheung Chiu","Just as computational simulations of atoms, molecules and cells have shaped the way we study the sciences, true-to-life simulations of human-like agents can be valuable tools for studying human behavior. We propose Humanoid Agents, a system that guides Generative Agents to behave more like humans by introducing three elements of System 1 processing: Basic needs (e.g. hunger, health and energy), Emotion and Closeness in Relationships. Humanoid Agents are able to use these dynamic elements to adapt their daily activities and conversations with other agents, as supported with empirical experiments. Our system is designed to be extensible to various settings, three of which we demonstrate, as well as to other elements influencing human behavior (e.g. empathy, moral values and cultural background). Our platform also includes a Unity WebGL game interface for visualization and an interactive analytics dashboard to show agent statuses over time.","University of Washington, NVIDIA, The University of Hong Kong"
15,./images/humanoid_agents_platform_for_20231009.png,Language Agents as Digital Representatives in Collective Decision-Making,"Jarrett, Daniel and Pislar, Miruna and Bakker, Michiel A and Tessler, Michael Henry and Koster, Raphael and Balaguer, Jan and Elie, Romuald and Summerfield, Christopher and Tacchetti, Andrea","Consider the process of collective decision-making, in which a group of individualsinteractively select a preferred outcome from among a universe of alternatives. Inthis context, “representation” is the activity of making an individual’s preferencespresent in the process via participation by a proxy agent—i.e. their “representative”.To this end, learned models of human behavior have the potential to fill this role,with practical implications for multi-agent scenario studies and mechanism design.In this work, we investigate the possibility of training language agents to behavein the capacity of representatives of human agents, appropriately expressing thepreferences of those individuals whom they stand for. First, we formalize the settingof collective decision-making—as the episodic process of interaction between agroup of agents and a decision mechanism. On this basis, we then formalize theproblem of digital representation—as the simulation of an agent’s behavior to yieldequivalent outcomes from the mechanism. Finally, we conduct an empirical casestudy in the setting of consensus-finding among diverse humans, and demonstratethe feasibility of fine-tuning large language models to act as digital representatives.",Google DeepMind
16,./images/language_agents_as_digital_20231108.png,LLM-Driven Agents for Influencer Selection in Digital Advertising Campaigns,"Xiaoqing Zhang, Xiuying Chen, Yuhan Liu, Jianzhou Wang, Zhenxing Hu, Rui Yan","In the digital world, influencers are pivotal as opinion leaders, shap-ing the views and choices of their influencees. Modern advertisingoften follows this trend, where marketers choose appropriate in-fluencers for product endorsements, based on thorough marketanalysis. Previous studies on influencer selection have typicallyrelied on numerical representations of individual opinions andinteractions, a method that simplifies the intricacies of social dy-namics. With the development of large language models (LLMs),we now have the opportunity to capture the nuanced exchangesof information within social networks. Hence, in this work, wefirst introduce an Influencer Dynamics Simulator (IDS), helpingpromoters identify and select the right influencers to market theirproducts, based on LLM simulation. Concretely, we first propose aninfluencer-influencee engagement-based pre-selection module toscreen potential influencer candidates. Subsequently, a simulation isconstructed for these candidates and their influencees. Each user isrepresented as an LLM-based agent, drawing from their interactionhistory to deduce their profile and interests. The influencee agentswill predict their behavior in response to influencer advertising. Fi-nally, we develop a ranking metric designed to pinpoint influencerswho are most likely to drive product purchases based on feedbackfrom their influencees. To evaluate our framework, we collect areal-world advertising network dataset, including social relations,post and comment content, and user behaviors.......","Renmin University of China, King Abdullah University of Science and Technology, Moonshot AI"
17,./images/llm-driven_agents_for_influencer_20240322.png,Lyfe Agents: Generative agents for low-cost real-time social interactions,"Zhao Kaiya, Michelangelo Naim, Jovana Kondic, Manuel Cortes, Jiaxin Ge, Shuying Luo, Guangyu Robert Yang, Andrew Ahn","Highly autonomous generative agents powered by large language models promise to simulate intricate social behaviors in virtual societies. However, achieving real-time interactions with humans at a low computational cost remains challenging. Here, we introduce Lyfe Agents. They combine low-cost with real-time responsiveness, all while remaining intelligent and goal-oriented. Key innovations include: (1) an option-action framework, reducing the cost of high-level decisions; (2) asynchronous self-monitoring for better self-consistency; and (3) a Summarize-and-Forget memory mechanism, prioritizing critical memory items at a low cost. We evaluate Lyfe Agents' self-motivation and sociability across several multi-agent scenarios in our custom LyfeGame 3D virtual environment platform. When equipped with our brain-inspired techniques, Lyfe Agents can exhibit human-like self-motivated social reasoning. For example, the agents can solve a crime (a murder mystery) through autonomous collaboration and information exchange. Meanwhile, our techniques enabled Lyfe Agents to operate at a computational cost 10-100 times lower than existing alternatives. Our findings underscore the transformative potential of autonomous generative agents to enrich human social experiences in virtual worlds.","Massachusetts Institute of Technology, Peking University, LyfeAL"
18,./images/lyfe_agents_generative_agents_20231003.png,MetaAgents: Simulating Interactions of Human Behaviors for LLM-based Task-oriented Coordination via Collaborative Generative Agents,"Yuan Li, Yixuan Zhang, Lichao Sun","Significant advancements have occurred in the application of Large LanguageModels (LLMs) for various tasks and social simulations. Despite this, their capac-ities to coordinate within task-oriented social contexts are under-explored. Suchcapabilities are crucial if LLMs are to effectively mimic human-like social be-havior and produce meaningful results. To bridge this gap, we introduce collab-orative generative agents, endowing LLM-based Agents with consistent behaviorpatterns and task-solving abilities. We situate these agents in a simulated job fairenvironment as a case study to scrutinize their coordination skills. We proposea novel framework that equips collaborative generative agents with human-likereasoning abilities and specialized skills. Our evaluation demonstrates that theseagents show promising performance. However, we also uncover limitations thathinder their effectiveness in more complex coordination tasks. Our work providesvaluable insights into the role and evolution of LLMs in task-oriented social sim-ulations.","University of Cambridge, William & Mary, Lehigh University"
19,./images/metaagents_simulating_interactions_of_20231010.png,On Generative Agents in Recommendation,"An Zhang, Yuxin Chen, Leheng Sheng, Xiang Wang, Tat-Seng Chua","Recommender systems are the cornerstone of today's information dissemination, yet a disconnect between offline metrics and online performance greatly hinders their development. Addressing this challenge, we envision a recommendation simulator, capitalizing on recent breakthroughs in human-level intelligence exhibited by Large Language Models (LLMs). We propose Agent4Rec, a user simulator in recommendation, leveraging LLM-empowered generative agents equipped with user profile, memory, and actions modules specifically tailored for the recommender system. In particular, these agents' profile modules are initialized using real-world datasets (e.g. MovieLens, Steam, Amazon-Book), capturing users' unique tastes and social traits; memory modules log both factual and emotional memories and are integrated with an emotion-driven reflection mechanism; action modules support a wide variety of behaviors, spanning both taste-driven and emotion-driven actions. Each agent interacts with personalized recommender models in a page-by-page manner, relying on a pre-implemented collaborative filtering-based recommendation algorithm. We delve into both the capabilities and limitations of Agent4Rec, aiming to explore an essential research question: ``To what extent can LLM-empowered generative agents faithfully simulate the behavior of real, autonomous humans in recommender systems?'' Extensive and multi-faceted evaluations of Agent4Rec highlight both the alignment and deviation between agents and user-personalized preferences. Beyond mere performance comparison, we explore insightful experiments, such as emulating the filter bubble effect and discovering the underlying causal relationships in recommendation tasks.","National University of Singapore, Tsinghua University, University of Science and Technology of China"
20,./images/on_generative_agents_in_20231016.png,"Out of One, Many: Using Language Models to Simulate Human Samples","Lisa P. Argyle, Ethan C. Busby, Nancy Fulda, Joshua Gubler, Christopher Rytting, David Wingate","We propose and explore the possibility that language models can be studied as effective proxies for specific human sub-populations in social science research. Practical and research applications of artificial intelligence tools have sometimes been limited by problematic biases (such as racism or sexism), which are often treated as uniform properties of the models. We show that the ""algorithmic bias"" within one such tool -- the GPT-3 language model -- is instead both fine-grained and demographically correlated, meaning that proper conditioning will cause it to accurately emulate response distributions from a wide variety of human subgroups. We term this property ""algorithmic fidelity"" and explore its extent in GPT-3. We create ""silicon samples"" by conditioning the model on thousands of socio-demographic backstories from real human participants in multiple large surveys conducted in the United States. We then compare the silicon and human samples to demonstrate that the information contained in GPT-3 goes far beyond surface similarity. It is nuanced, multifaceted, and reflects the complex interplay between ideas, attitudes, and socio-cultural context that characterize human attitudes. We suggest that language models with sufficient algorithmic fidelity thus constitute a novel and powerful tool to advance understanding of humans and society across a variety of disciplines.",Brigham Young University
21,./images/out_of_one_many_20220914.png,Quantifying the Impact of Large Language Models on Collective Opinion Dynamics,"Chao Li, Xing Su, Haoying Han, Cong Xue, Chunmo Zheng, Chao Fan","The process of opinion expression and exchange is a critical component of democratic societies. As people interact with large language models (LLMs) in the opinion shaping process different from traditional media, the impacts of LLMs are increasingly recognized and being concerned. However, the knowledge about how LLMs affect the process of opinion expression and exchange of social opinion networks is very limited. Here, we create an opinion network dynamics model to encode the opinions of LLMs, cognitive acceptability and usage strategies of individuals, and simulate the impact of LLMs on opinion dynamics in a variety of scenarios. The outcomes of the simulations inform about effective demand-oriented opinion network interventions. The results from this study suggested that the output opinion of LLMs has a unique and positive effect on the collective opinion difference. The marginal effect of cognitive acceptability on collective opinion formation is nonlinear and shows a decreasing trend. When people partially rely on LLMs, the exchange process of opinion becomes more intense and the diversity of opinion becomes more favorable. In fact, there is 38.6% more opinion diversity when people all partially rely on LLMs, compared to prohibiting the use of LLMs entirely. The optimal diversity of opinion was found when the fractions of people who do not use, partially rely on, and fully rely on LLMs reached roughly 4:12:1. Our experiments also find that introducing extra agents with opposite/neutral/random opinions, we can effectively mitigate the impact of biased/toxic output from LLMs. Our findings provide valuable insights into opinion dynamics in the age of LLMs, highlighting the need for customized interventions tailored to specific scenarios to address the drawbacks of improper output and use of LLMs."," Zhejiang University, Clemson University, "
22,./images/quantifying_the_impact_of_20230807.png,S3: Social-network Simulation System with Large Language Model-Empowered Agents,"Chen Gao, Xiaochong Lan, Zhihong Lu, Jinzhu Mao, Jinghua Piao, Huandong Wang, Depeng Jin, Yong Li","Simulation plays a crucial role in addressing various challenges within socialscience. It offers extensive applications such as state prediction, phenomena ex-planation, and policy-making support, among others. In this work, we harness thehuman-like capabilities of large language models (LLMs) in sensing, reasoning,and behaving, and utilize these qualities to construct the S3 system (short forSocial network Simulation System). Adhering to the widely employed agent-basedsimulation paradigm, we employ fine-tuning and prompt engineering techniques toensure that the agent’s behavior closely emulates that of a genuine human withinthe social network. Specifically, we simulate three pivotal aspects: emotion, at-titude, and interaction behaviors. By endowing the agent in the system with theability to perceive the informational environment and emulate human actions, weobserve the emergence of population-level phenomena, including the propagationof information, attitudes, and emotions. We conduct an evaluation encompassingtwo levels of simulation, employing real-world social network data. Encouragingly,the results demonstrate promising accuracy. This work represents an initial step inthe realm of social network simulation empowered by LLM-based agents. We an-ticipate that our endeavors will serve as a source of inspiration for the developmentof simulation systems within, but not limited to, social science.",Tsinghua University
23,./images/s3_social-network_simulation_system_20230727.png,Simulating Opinion Dynamics with Networks of LLM-based Agents,"Yun-Shiuan Chuang, Agam Goyal, Nikunj Harlalka, Siddharth Suresh, Robert Hawkins, Sijia Yang, Dhavan Shah, Junjie Hu, Timothy T. Rogers","Accurately simulating human opinion dynam-ics is crucial for understanding a variety of soci-etal phenomena, including polarization and thespread of misinformation. However, the agent-based models (ABMs) commonly used for suchsimulations often over-simplify human behav-ior. We propose a new approach to simulat-ing opinion dynamics based on populations ofLarge Language Models (LLMs). Our findingsreveal a strong inherent bias in LLM agents to-wards producing accurate information, leadingsimulated agents to consensus in line with sci-entific reality. This bias limits their utility forunderstanding resistance to consensus viewson issues like climate change. After induc-ing confirmation bias through prompt engineer-ing, however, we observed opinion fragmenta-tion in line with existing agent-based modelingand opinion dynamics research. These insightshighlight the promise and limitations of LLMagents in this domain and suggest a path for-ward: refining LLMs with real-world discourseto better simulate the evolution of human be-liefs.",University of Wisconsin-Madison
24,./images/simulating_opinion_dynamics_with_20231116.png,Simulating Social Media Using Large Language Models to Evaluate Alternative News Feed Algorithms,"Petter Törnberg, Diliara Valeeva, Justus Uitermark, Christopher Bail",". Social media is often criticized for amplifyingtoxic discourse and discouraging constructive conversa-tions. But designing social media platforms to promotebetter conversations is inherently challenging. This paperasks whether simulating social media through a combina-tion of Large Language Models (LLM) and Agent-BasedModeling can help researchers study how different newsfeed algorithms shape the quality of online conversations.We create realistic personas using data from the Ameri-can National Election Study to populate simulated socialmedia platforms. Next, we prompt the agents to readand share news articles — and like or comment uponeach other’s messages — within three platforms that usedifferent news feed algorithms. In the first platform, userssee the most liked and commented posts from users whomthey follow. In the second, they see posts from all users —even those outside their own network. The third platformemploys a novel “bridging” algorithm that highlights poststhat are liked by people with opposing political views. Wefind this bridging algorithm promotes more constructive,non-toxic, conversation across political divides than theother two models. Though further research is needed toevaluate these findings, we argue that LLMs hold consid-erable potential to improve simulation research on socialmedia and many other complex social settings.","University of Amsterdam, Duke University"
25,./images/simulating_social_media_using_20231005.png,Social Simulacra: Creating Populated Prototypes for Social Computing Systems,"Joon Sung Park, Lindsay Popowski, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein","Social computing prototypes probe the social behaviors that mayarise in an envisioned system design. This prototyping practiceis currently limited to recruiting small groups of people. Unfortu-nately, many challenges do not arise until a system is populatedat a larger scale. Can a designer understand how a social systemmight behave when populated, and make adjustments to the de-sign before the system falls prey to such challenges? We intro-duce social simulacra, a prototyping technique that generates abreadth of realistic social interactions that may emerge when a so-cial computing system is populated. Social simulacra take as inputthe designer’s description of a community’s design—goal, rules, andmember personas—and produce as output an instance of that designwith simulated behavior, including posts, replies, and anti-socialbehaviors. We demonstrate that social simulacra shift the behaviorsthat they generate appropriately in response to design changes, andthat they enable exploration of “what if?” scenarios where commu-nity members or moderators intervene. To power social simulacra,we contribute techniques for prompting a large language modelto generate thousands of distinct community members and theirsocial interactions with each other; these techniques are enabled bythe observation that large language models’ training data alreadyincludes a wide variety of positive and negative behavior on socialmedia platforms. In evaluations, we show that participants are of-ten unable to distinguish social simulacra from actual communitybehavior and that social computing designers successfully refinetheir social computing designs when using social simulacra.","Stanford University, Google Research"
26,./images/social_simulacra_creating_populated_20220808.png,The Wisdom of Partisan Crowds: Comparing Collective Intelligence in Humans and LLM-based Agents,"Yun-Shiuan Chuang, Siddharth Suresh, Nikunj Harlalka, Agam Goyal, Robert Hawkins, Sijia Yang, Dhavan Shah, Junjie Hu, Timothy T. Rogers","Human groups are able to converge on more accurate beliefs through deliberation,even in the presence of polarization and partisan bias — a phenomenon known asthe “wisdom of partisan crowds.” Generated agents powered by Large LanguageModels (LLMs) are increasingly used to simulate human collective behavior, yetfew benchmarks exist for evaluating their dynamics against the behavior of hu-man groups. In this paper, we examine the extent to which the wisdom of partisancrowds emerges in groups of LLM-based agents that are prompted to role-playas partisan personas (e.g., Democrat or Republican). We find that they not onlydisplay human-like partisan biases, but also converge to more accurate beliefsthrough deliberation as humans do. We then identify several factors that interferewith convergence, including the use of chain-of-thought prompt and lack of detailsin personas. Conversely, fine-tuning on human data appears to enhance conver-gence. These findings show the potential and limitations of LLM-based agents asa model of human collective intelligence.",University of Wisconsin-Madison
27,./images/the_wisdom_of_partisan_20231116.png,To Infinity and Beyond- SHOW-1 and Showrunner Agents in Multi-Agent Simulations,"Philipp Maas, Frank Carey, Chris Wheeler, Edward Saatchi, Pete Billington, Jessica Yaffa Shamash","In this work we present our approach to generating high-quality episodic content for IP’s (Intellectual Property) using large language models (LLMs), custom state-of- the art diffusion models and our multi-agent simulation for contextualization, story progression and behavioral control. Powerful LLMs such as GPT-4 were trained on a large corpus of TV show data which lets us believe that with the right guidance users will be able to rewrite entire seasons.""That Is What Entertainment Will Look Like. Maybe people are still upset about the last season of Game of Thrones. Imagine if you could ask your A.I. to make a new ending that goes a different way and maybe even put yourself in there as a main character or something.”. ",Fable Studio
28,./images/to_infinity_and_beyond_20230724.png,Unveiling the Truth and Facilitating Change: Towards Agent-based Large-scale Social Movement Simulation,"Xinyi Mou, Zhongyu Wei, Xuanjing Huang","Social media has emerged as a cornerstone ofsocial movements, wielding significant influ-ence in driving societal change. Simulatingthe response of the public and forecasting thepotential impact has become increasingly im-portant. However, existing methods for simu-lating such phenomena encounter challengesconcerning their efficacy and efficiency in cap-turing the behaviors of social movement par-ticipants. In this paper, we introduce a hybridframework HiSim for social media user simu-lation, wherein users are categorized into twotypes. Core users are driven by Large Lan-guage Models, while numerous ordinary usersare modeled by deductive agent-based models.We further construct a Twitter-like environmentto replicate their response dynamics followingtrigger events. Subsequently, we develop amulti-faceted benchmark SoMoSiMu-Benchfor evaluation and conduct comprehensive ex-periments across real-world datasets. Exper-imental results demonstrate the effectivenessand flexibility of our method","Fudan University, Shanghai Collaborative Innovation Center of Intelligent Visual Computing"
29,./images/unveiling_the_truth_and_20240226.png,User Behavior Simulation with Large Language Model based Agents,"Lei Wang, Jingsen Zhang, Hao Yang, Zhiyuan Chen, Jiakai Tang, Zeyu Zhang, Xu Chen, Yankai Lin, Ruihua Song, Wayne Xin Zhao, Jun Xu, Zhicheng Dou, Jun Wang, Ji-Rong Wen","Simulating high quality user behavior data has always been a fundamental problem in human-centered applications, where the major difficulty originates from the intricate mechanism of human decision process. Recently, substantial evidences have suggested that by learning huge amounts of web knowledge, large language models (LLMs) can achieve human-like intelligence. We believe these models can provide significant opportunities to more believable user behavior simulation. To inspire such direction, we propose an LLM-based agent framework and design a sandbox environment to simulate real user behaviors. Based on extensive experiments, we find that the simulated behaviors of our method are very close to the ones of real humans. Concerning potential applications, we simulate and study two social phenomenons including (1) information cocoons and (2) user conformity behaviors. This research provides novel simulation paradigms for human-centered applications.","Renmin University of China, Beijing Key Laboratory of Big Data Management and Analysis Methods, University College London"
30,./images/user_behavior_simulation_with_20230605.png,Using Large Language Models to Simulate Multiple Humans and Replicate Human Subject Studies,"Gati Aher, Rosa I. Arriaga, Adam Tauman Kalai","We introduce a new type of test, called a Turing Experiment (TE), for evaluating to what extent a given language model, such as GPT models, can simulate different aspects of human behavior. A TE can also reveal consistent distortions in a language model's simulation of a specific human behavior. Unlike the Turing Test, which involves simulating a single arbitrary individual, a TE requires simulating a representative sample of participants in human subject research. We carry out TEs that attempt to replicate well-established findings from prior studies. We design a methodology for simulating TEs and illustrate its use to compare how well different language models are able to reproduce classic economic, psycholinguistic, and social psychology experiments: Ultimatum Game, Garden Path Sentences, Milgram Shock Experiment, and Wisdom of Crowds. In the first three TEs, the existing findings were replicated using recent models, while the last TE reveals a ""hyper-accuracy distortion"" present in some language models (including ChatGPT and GPT-4), which could affect downstream applications in education and the arts.","Olin College of Engineering, Georgia Tech, Microsoft Research"
31,./images/using_large_language_models_20220818.png,War and Peace (WarAgent): Large Language Model-based Multi-Agent Simulation of World Wars,"Wenyue Hua, Lizhou Fan, Lingyao Li, Kai Mei, Jianchao Ji, Yingqiang Ge, Libby Hemphill, Yongfeng Zhang","Can we avoid wars at the crossroads of history? This question has been pursued byindividuals, scholars, policymakers, and organizations throughout human history.In this research, we attempt to answer the question based on the recent advancesof Artificial Intelligence (AI) and Large Language Models (LLMs). We proposeWarAgent, an LLM-powered multi-agent AI system, to simulate the participatingcountries, their decisions, and the consequences, in historical international conflicts,including the World War I (WWI), the World War II (WWII), and the WarringStates Period (WSP) in Ancient China. By evaluating the simulation effectiveness,we examine the advancements and limitations of cutting-edge AI systems’ abilitiesin studying complex collective human behaviors such as international conflictsunder diverse settings. In these simulations, the emergent interactions amongagents also offer a novel perspective for examining the triggers and conditions thatlead to war. Our findings offer data-driven and AI-augmented insights that canredefine how we approach conflict resolution and peacekeeping strategies. Theimplications stretch beyond historical analysis, offering a blueprint for using AI tounderstand human history and possibly prevent future international conflicts. Codeand data are available at https://github.com/agiresearch/WarAgent.",Rutgers University
32,./images/war_and_peace_(waragent)_20231128.png,To be Continued...,Your Contributions are Welcome!,,
1 image_path title author summary affiliation
2 0 ./images/4d.png (Perhaps) Beyond Human Translation: Harnessing Multi-Agent Collaboration for Translating Ultra-Long Literary Texts Minghao Wu, Yulin Yuan, Gholamreza Haffari, Longyue Wang Recent advancements in machine translation (MT) have significantly enhancedtranslation quality across various domains. However, the translation of literarytexts remains a formidable challenge due to their complex language, figurative ex-pressions, and cultural nuances. In this work, we introduce a novel multi-agentframework based on large language models (LLMs) for literary translation, im-plemented as a company called TRANSAGENTS, which mirrors traditional trans-lation publication process by leveraging the collective capabilities of multipleagents, to address the intricate demands of translating literary works. To evaluatethe effectiveness of our system, we propose two innovative evaluation strategies:Monolingual Human Preference (MHP) and Bilingual LLM Preference (BLP).MHP assesses translations from the perspective of monolingual readers of the tar-get language, while BLP uses advanced LLMs to compare translations directlywith the original texts. Empirical findings indicate that despite lower d-BLEUscores, translations from TRANSAGENTS are preferred by both human evalua-tors and LLMs over human-written references, particularly in genres requiringdomain-specific knowledge. We also highlight the strengths and limitations ofTRANSAGENTS through case studies and suggests directions for future research. Monash University, University of Macau, Tencent AI Lab
3 1 ./images/(perhaps)_beyond_human_translation_20240520.png Agent Hospital: A Simulacrum of Hospital with Evolvable Medical Agents Junkai Li, Siyu Wang, Meng Zhang, Weitao Li, Yunghwei Lai, Xinhui Kang, Weizhi Ma, Yang Liu In this paper, we introduce a simulacrum of hospital called Agent Hospital that simulates theentire process of treating illness. All patients, nurses, and doctors are autonomous agents powered bylarge language models (LLMs). Our central goal is to enable a doctor agent to learn how to treat illnesswithin the simulacrum. To do so, we propose a method called MedAgent-Zero. As the simulacrum cansimulate disease onset and progression based on knowledge bases and LLMs, doctor agents can keepaccumulating experience from both successful and unsuccessful cases. Simulation experiments show thatthe treatment performance of doctor agents consistently improves on various tasks. More interestingly,the knowledge the doctor agents have acquired in Agent Hospital is applicable to real-world medicarebenchmarks. After treating around ten thousand patients (real-world doctors may take over two years),the evolved doctor agent achieves a state-of-the-art accuracy of 9 Tsinghua University
4 2 ./images/agent_hospital_a_simulacrum_20240505.png AgentCF: Collaborative Learning with Autonomous Language Agents for Recommender Systems Junjie Zhang, Yupeng Hou, Ruobing Xie, Wenqi Sun, Julian McAuley, Wayne Xin Zhao, Leyu Lin, Ji-Rong Wen Recently, there has been an emergence of employing LLM-poweredagents as believable human proxies, based on their remarkabledecision-making capability. However, existing studies mainly focuson simulating human dialogue. Human non-verbal behaviors, suchas item clicking in recommender systems, although implicitly ex-hibiting user preferences and could enhance the modeling of users,have not been deeply explored. The main reasons lie in the gapbetween language modeling and behavior modeling, as well as theincomprehension of LLMs about user-item relations.To address this issue, we propose AgentCF for simulating user-item interactions in recommender systems through agent-basedcollaborative filtering. We creatively consider not only users butalso items as agents, and develop a collaborative learning approachthat optimizes both kinds of agents together. Specifically, at eachtime step, we first prompt the user and item agents to interact au-tonomously. Then, based on the disparities between the agents’decisions and real-world interaction records, user and item agentsare prompted to reflect on and adjust the misleading simulationscollaboratively, thereby modeling their two-sided relations. The op-timized agents can also propagate their preferences to other agentsin subsequent interactions, implicitly capturing the collaborative fil-tering idea. Overall, the optimized agents exhibit diverse interactionbehaviors within our framework, including user-item, user-user,item-item, and collective interactions. The results show that theseagents can demonstrate personalized behaviors akin to those of real-world individuals, sparking the development of next-generationuser behavior simulation. Renmin University of China, UC San Diego, Tencent
5 3 ./images/agentcf_collaborative_learning_with_20231013.png AgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors Weize Chen, Yusheng Su, Jingwei Zuo, Cheng Yang, Chenfei Yuan, Chi-Min Chan, Heyang Yu, Yaxi Lu, Yi-Hsin Hung, Chen Qian, Yujia Qin, Xin Cong, Ruobing Xie, Zhiyuan Liu, Maosong Sun, Jie Zhou Autonomous agents empowered by Large Language Models (LLMs) have under-gone significant improvements, enabling them to generalize across a broad spec-trum of tasks. However, in real-world scenarios, cooperation among individuals isoften required to enhance the efficiency and effectiveness of task accomplishment.Hence, inspired by human group dynamics, we propose a multi-agent frameworkAGENTVERSE that can effectively orchestrate a collaborative group of expert agentsas a greater-than-the-sum-of-its-parts system. Our experiments demonstrate thatAGENTVERSE can proficiently deploy multi-agent groups that outperform a singleagent. Extensive experiments on text understanding, reasoning, coding, tool utiliza-tion, and embodied AI confirm the effectiveness of AGENTVERSE. Moreover, ouranalysis of agent interactions within AGENTVERSE reveals the emergence of spe-cific collaborative behaviors, contributing to heightened group efficiency. Our codehas been released at https://github.com/OpenBMB/AgentVerse/. Tsinghua University, Beijing University of Posts and Telecommunications, Tencent Inc.
6 4 ./images/agentverse_facilitating_multi-agent_collaboration_20230821.png AI Hospital: Interactive Evaluation and Collaboration of LLMs as Intern Doctors for Clinical Diagnosis Zhihao Fan, Jialong Tang, Wei Chen, Siyuan Wang, Zhongyu Wei, Jun Xi, Fei Huang, Jingren Zhou The incorporation of Large Language Models(LLMs) in healthcare marks a significant ad-vancement. However, the application has pre-dominantly been limited to discriminative andquestion-answering tasks, which does not fullyleverage their interactive potential. To addressthis limitation, our paper presents AI Hospital,a framework designed to build a real-time in-teractive diagnosis environment. To simulatethe procedure, we collect high-quality medicalrecords to create patient, examiner, and medicaldirector agents. AI Hospital is then utilized forthe interactive evaluation and collaboration ofLLMs. Initially, we create a Multi-View Medi-cal Evaluation (MVME) benchmark where vari-ous LLMs serve as intern doctors for interactivediagnosis. Subsequently, to improve diagnosticaccuracy, we introduce a collaborative mech-anism that involves iterative discussions anda dispute resolution process under the supervi-sion of the medical director. In our experiments,we validate the reliability of AI Hospital. Theresults not only explore the feasibility of applyLLMs in clinical consultation but also confirmthe effectiveness of the dispute resolution fo-cused collaboration method. Alibaba Inc., Huazhong University of Science and Technology, Fudan University
7 5 ./images/ai_hospital_interactive_evaluation_20240215.png Are you in a Masquerade? Exploring the Behavior and Impact of Large Language Model Driven Social Bots in Online Social Networks Siyu Li, Jin Yang, Kui Zhao As the capabilities of Large Language Models (LLMs) emerge, they not only assist in accomplishing traditional tasks within more efficient paradigms but also stimulate the evolution of social bots. Researchers have begun exploring the implementation of LLMs as the driving core of social bots, enabling more efficient and user-friendly completion of tasks like profile completion, social behavior decision-making, and social content generation. However, there is currently a lack of systematic research on the behavioral characteristics of LLMs-driven social bots and their impact on social networks. We have curated data from Chirper, a Twitter-like social network populated by LLMs-driven social bots and embarked on an exploratory study. Our findings indicate that: (1) LLMs-driven social bots possess enhanced individual-level camouflage while exhibiting certain collective characteristics; (2) these bots have the ability to exert influence on online communities through toxic behaviors; (3) existing detection methods are applicable to the activity environment of LLMs-driven social bots but may be subject to certain limitations in effectiveness. Moreover, we have organized the data collected in our study into the Masquerade-23 dataset, which we have publicly released, thus addressing the data void in the subfield of LLMs-driven social bots behavior datasets. Our research outcomes provide primary insights for the research and governance of LLMs-driven social bots within the research community. Sichuan University
8 6 ./images/are_you_in_a_20230719.png BattleAgent: Multi-modal Dynamic Emulation on Historical Battles to Complement Historical Analysis Shuhang Lin, Wenyue Hua, Lingyao Li, Che-Jui Chang, Lizhou Fan, Jianchao Ji, Hang Hua, Mingyu Jin, Jiebo Luo, Yongfeng Zhang This paper presents BattleAgent, a detailed emulation demonstration system thatcombines the Large Vision-Language Model (VLM) and Multi-Agent System(MAS). This novel system aims to simulate complex dynamic interactions amongmultiple agents, as well as between agents and their environments, over a period oftime. It emulates both the decision-making processes of leaders and the viewpointsof ordinary participants, such as soldiers. The emulation showcases the currentcapabilities of agents, featuring fine-grained multi-modal interactions betweenagents and landscapes. It develops customizable agent structures to meet specificsituational requirements, for example, a variety of battle-related activities likescouting and trench digging. These components collaborate to recreate historicalevents in a lively and comprehensive manner while offering insights into thethoughts and feelings of individuals from diverse viewpoints. The technologicalfoundations of BattleAgent establish detailed and immersive settings for historicalbattles, enabling individual agents to partake in, observe, and dynamically respondto evolving battle scenarios. This methodology holds the potential to substantiallydeepen our understanding of historical events, particularly through individualaccounts. Such initiatives can also aid historical research, as conventional historicalnarratives often lack documentation and prioritize the perspectives of decision-makers, thereby overlooking the experiences of ordinary individuals. This biaseddocumentation results in a considerable gap in our historical understanding, as manystories remain untold...... Rutgers University, University of Michigan, University of Rochester
9 7 ./images/battleagent_multi-modal_dynamic_emulation_20240423.png Can Large Language Model Agents Simulate Human Trust Behaviors? Chengxing Xie, Canyu Chen, Feiran Jia, Ziyu Ye, Kai Shu, Adel Bibi, Ziniu Hu, Philip Torr, Bernard Ghanem, Guohao Li Large Language Model (LLM) agents have beenincreasingly adopted as simulation tools to modelhumans in applications such as social science.However, one fundamental question remains: canLLM agents really simulate human behaviors? Inthis paper, we focus on one of the most criticalbehaviors in human interactions, trust, and aim toinvestigate whether or not LLM agents can sim-ulate human trust behaviors. We first find thatLLM agents generally exhibit trust behaviors, re-ferred to as agent trust, under the framework ofTrust Games, which are widely recognized in be-havioral economics. Then, we discover that LLMagents can have high behavioral alignment withhumans regarding trust behaviors, particularly forGPT-4, indicating the feasibility to simulate hu-man trust behaviors with LLM agents. In addition,we probe into the biases in agent trust and thedifferences in agent trust towards agents and hu-mans. We also explore the intrinsic properties ofagent trust under conditions including advancedreasoning strategies and external manipulations.We further offer important implications of ourdiscoveries for various scenarios where trust isparamount. Our study provides new insights intothe behaviors of LLM agents and the fundamentalanalogy between LLMs and humans. KAUST, Illinois Institute of Technology, Pennsylvania State University, The University of Chicago, University of Oxford, California Institute of Technology
10 8 ./images/can_large_language_model_20240207.png ChatDev: Communicative Agents for Software Development Chen Qian, Wei Liu, Hongzhang Liu, Nuo Chen, Yufan Dang, Jiahao Li, Cheng Yang, Weize Chen, Yusheng Su, Xin Cong, Juyuan Xu, Dahai Li, Zhiyuan Liu, Maosong Sun Software development is a complex task thatnecessitates cooperation among multiple mem-bers with diverse skills. Numerous studies useddeep learning to improve specific phases in awaterfall model, such as design, coding, andtesting.However, the deep learning modelin each phase requires unique designs, lead-ing to technical inconsistencies across variousphases, which results in a fragmented and in-effective development process. In this paper,we introduce ChatDev, a chat-powered soft-ware development framework in which special-ized agents driven by large language models(LLMs) are guided in what to communicate(via chat chain) and how to communicate (viacommunicative dehallucination). These agentsactively contribute to the design, coding, andtesting phases through unified language-basedcommunication, with solutions derived fromtheir multi-turn dialogues. We found their uti-lization of natural language is advantageousfor system design, and communicating in pro-gramming language proves helpful in debug-ging. This paradigm demonstrates how linguis-tic communication facilitates multi-agent col-laboration, establishing language as a unify-ing bridge for autonomous task-solving amongLLM agents. The code and data are availableat https://github.com/OpenBMB/ChatDev. Tsinghua University, The University of Sydney, BUPT, Modelbest Inc.
11 9 ./images/chatdev_communicative_agents_for_20230716.png CompeteAI: Understanding the Competition Dynamics in Large Language Model-based Agents Qinlin Zhao, Jindong Wang, Yixuan Zhang, Yiqiao Jin, Kaijie Zhu, Hao Chen, Xing Xie Large language models (LLMs) have been widelyused as agents to complete different tasks, suchas personal assistance or event planning. Whilemost of the work has focused on cooperationand collaboration between agents, little workexplores competition, another important mech-anism that promotes the development of soci-ety and economy. In this paper, we seek to ex-amine the competition dynamics in LLM-basedagents. We first propose a general framework forstudying the competition between agents. Then,we implement a practical competitive environ-ment using GPT-4 to simulate a virtual town withtwo types of agents, including restaurant agentsand customer agents. Specifically, the restaurantagents compete with each other to attract morecustomers, where competition encourages themto transform, such as cultivating new operatingstrategies. Simulation experiments reveal severalinteresting findings at the micro and macro lev-els, which align well with existing market andsociological theories. We hope that the frame-work and environment can be a promising testbedto study the competition that fosters understand-ing of society. Code is available at: https://github.com/microsoft/competeai. University of Science and Technology of China, Microsoft Research, William & Mary, Georgia Institute of Technology, Carnegie Mellon University
12 10 ./images/competeai_understanding_the_competition_20231026.png EconAgent: Large Language Model-Empowered Agents for Simulating Macroeconomic Activities Nian Li, Chen Gao, Mingyu Li, Yong Li, Qingmin Liao The advent of artificial intelligence has led to agrowing emphasis on data-driven modeling inmacroeconomics, with agent-based modeling(ABM) emerging as a prominent bottom-upsimulation paradigm. In ABM, agents (e.g.,households, firms) interact within a macroe-conomic environment, collectively generatingmarket dynamics. Existing agent modeling typ-ically employs predetermined rules or learning-based neural networks for decision-making.However, customizing each agent presents sig-nificant challenges, complicating the modelingof agent heterogeneity. Additionally, the in-fluence of multi-period market dynamics andmultifaceted macroeconomic factors are oftenoverlooked in decision-making processes. Inthis work, we introduce EconAgent, a largelanguage model-empowered agent with human-like characteristics for macroeconomic simu-lation. We first construct a simulation envi-ronment that incorporates various market dy-namics driven by agents’ decisions regardingwork and consumption. Through the perceptionmodule, we create heterogeneous agents withdistinct decision-making mechanisms.Fur-thermore, we model the impact of macroeco-nomic trends using a memory module, whichallows agents to reflect on past individual ex-periences and market dynamics. Simulationexperiments show that EconAgent can makerealistic decisions, leading to more reasonablemacroeconomic phenomena compared to exist-ing rule-based or learning-based agents. Ourcodes are released at https://github.com/tsinghua-fib-lab/ACL24-EconAgent. Tsinghua University
13 11 ./images/econagent_large_language_model-empowered_20231016.png Epidemic Modeling with Generative Agents Ross Williams, Niyousha Hosseinichimeh, Aritra Majumdar, Navid Ghaffarzadegan This study offers a new paradigm of individual-level modeling to address the grand challenge of incorporating human behavior in epidemic models. Using generative artificial intelligence in an agent-based epidemic model, each agent is empowered to make its own reasonings and decisions via connecting to a large language model such as ChatGPT. Through various simulation experiments, we present compelling evidence that generative agents mimic real-world behaviors such as quarantining when sick and self-isolation when cases rise. Collectively, the agents demonstrate patterns akin to multiple waves observed in recent pandemics followed by an endemic period. Moreover, the agents successfully flatten the epidemic curve. This study creates potential to improve dynamic system modeling by offering a way to represent human brain, reasoning, and decision making. Virginia Tech
14 12 ./images/epidemic_modeling_with_generative_20230711.png Exploring Collaboration Mechanisms for LLM Agents: A Social Psychology View Jintian Zhang, Xin Xu, Ningyu Zhang, Ruibo Liu, Bryan Hooi, Shumin Deng As Natural Language Processing (NLP) sys-tems are increasingly employed in intricate so-cial environments, a pressing query emerges:Can these NLP systems mirror human-esquecollaborative intelligence, in a multi-agent so-ciety consisting of multiple large language mod-els (LLMs)? This paper probes the collabora-tion mechanisms among contemporary NLPsystems by melding practical experiments withtheoretical insights. We fabricate four unique‘societies’ comprised of LLM agents, whereeach agent is characterized by a specific ‘trait’(easy-going or overconfident) and engages incollaboration with a distinct ‘thinking pattern’(debate or reflection).Through evaluatingthese multi-agent societies on three benchmarkdatasets, we discern that certain collaborativestrategies not only outshine previous top-tierapproaches but also optimize efficiency (usingfewer API tokens). Moreover, our results fur-ther illustrate that LLM agents manifest human-like social behaviors, such as conformity andconsensus reaching, mirroring foundational so-cial psychology theories. In conclusion, weintegrate insights from social psychology tocontextualize the collaboration of LLM agents,inspiring further investigations into the collab-oration mechanism for LLMs. We have sharedour code and datasets1, hoping to catalyze fur-ther research in this promising avenue. Zhejiang University, National University of Singapore, NUS-NCS Joint Lab, Google DeepMind
15 13 ./images/exploring_collaboration_mechanisms_for_20231003.png Generative Agents: Interactive Simulacra of Human Behavior Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein Believable proxies of human behavior can empower interactive applications ranging from immersive environments to rehearsal spaces for interpersonal communication to prototyping tools. In this paper, we introduce generative agents--computational software agents that simulate believable human behavior. Generative agents wake up, cook breakfast, and head to work; artists paint, while authors write; they form opinions, notice each other, and initiate conversations; they remember and reflect on days past as they plan the next day. To enable generative agents, we describe an architecture that extends a large language model to store a complete record of the agent's experiences using natural language, synthesize those memories over time into higher-level reflections, and retrieve them dynamically to plan behavior. We instantiate generative agents to populate an interactive sandbox environment inspired by The Sims, where end users can interact with a small town of twenty five agents using natural language. In an evaluation, these generative agents produce believable individual and emergent social behaviors: for example, starting with only a single user-specified notion that one agent wants to throw a Valentine's Day party, the agents autonomously spread invitations to the party over the next two days, make new acquaintances, ask each other out on dates to the party, and coordinate to show up for the party together at the right time. We demonstrate through ablation that the components of our agent architecture--observation, planning, and reflection--each contribute critically to the believability of agent behavior. By fusing large language models with computational, interactive agents, this work introduces architectural and interaction patterns for enabling believable simulations of human behavior. Stanford University, Google Research, Google DeepMind
16 14 ./images/generative_agents_interactive_simulacra_20230407.png Humanoid Agents: Platform for Simulating Human-like Generative Agents Zhilin Wang, Yu Ying Chiu, Yu Cheung Chiu Just as computational simulations of atoms, molecules and cells have shaped the way we study the sciences, true-to-life simulations of human-like agents can be valuable tools for studying human behavior. We propose Humanoid Agents, a system that guides Generative Agents to behave more like humans by introducing three elements of System 1 processing: Basic needs (e.g. hunger, health and energy), Emotion and Closeness in Relationships. Humanoid Agents are able to use these dynamic elements to adapt their daily activities and conversations with other agents, as supported with empirical experiments. Our system is designed to be extensible to various settings, three of which we demonstrate, as well as to other elements influencing human behavior (e.g. empathy, moral values and cultural background). Our platform also includes a Unity WebGL game interface for visualization and an interactive analytics dashboard to show agent statuses over time. University of Washington, NVIDIA, The University of Hong Kong
17 15 ./images/humanoid_agents_platform_for_20231009.png Language Agents as Digital Representatives in Collective Decision-Making Jarrett, Daniel and Pislar, Miruna and Bakker, Michiel A and Tessler, Michael Henry and Koster, Raphael and Balaguer, Jan and Elie, Romuald and Summerfield, Christopher and Tacchetti, Andrea Consider the process of collective decision-making, in which a group of individualsinteractively select a preferred outcome from among a universe of alternatives. Inthis context, “representation” is the activity of making an individual’s preferencespresent in the process via participation by a proxy agent—i.e. their “representative”.To this end, learned models of human behavior have the potential to fill this role,with practical implications for multi-agent scenario studies and mechanism design.In this work, we investigate the possibility of training language agents to behavein the capacity of representatives of human agents, appropriately expressing thepreferences of those individuals whom they stand for. First, we formalize the settingof collective decision-making—as the episodic process of interaction between agroup of agents and a decision mechanism. On this basis, we then formalize theproblem of digital representation—as the simulation of an agent’s behavior to yieldequivalent outcomes from the mechanism. Finally, we conduct an empirical casestudy in the setting of consensus-finding among diverse humans, and demonstratethe feasibility of fine-tuning large language models to act as digital representatives. Google DeepMind
18 16 ./images/language_agents_as_digital_20231108.png LLM-Driven Agents for Influencer Selection in Digital Advertising Campaigns Xiaoqing Zhang, Xiuying Chen, Yuhan Liu, Jianzhou Wang, Zhenxing Hu, Rui Yan In the digital world, influencers are pivotal as opinion leaders, shap-ing the views and choices of their influencees. Modern advertisingoften follows this trend, where marketers choose appropriate in-fluencers for product endorsements, based on thorough marketanalysis. Previous studies on influencer selection have typicallyrelied on numerical representations of individual opinions andinteractions, a method that simplifies the intricacies of social dy-namics. With the development of large language models (LLMs),we now have the opportunity to capture the nuanced exchangesof information within social networks. Hence, in this work, wefirst introduce an Influencer Dynamics Simulator (IDS), helpingpromoters identify and select the right influencers to market theirproducts, based on LLM simulation. Concretely, we first propose aninfluencer-influencee engagement-based pre-selection module toscreen potential influencer candidates. Subsequently, a simulation isconstructed for these candidates and their influencees. Each user isrepresented as an LLM-based agent, drawing from their interactionhistory to deduce their profile and interests. The influencee agentswill predict their behavior in response to influencer advertising. Fi-nally, we develop a ranking metric designed to pinpoint influencerswho are most likely to drive product purchases based on feedbackfrom their influencees. To evaluate our framework, we collect areal-world advertising network dataset, including social relations,post and comment content, and user behaviors....... Renmin University of China, King Abdullah University of Science and Technology, Moonshot AI
19 17 ./images/llm-driven_agents_for_influencer_20240322.png Lyfe Agents: Generative agents for low-cost real-time social interactions Zhao Kaiya, Michelangelo Naim, Jovana Kondic, Manuel Cortes, Jiaxin Ge, Shuying Luo, Guangyu Robert Yang, Andrew Ahn Highly autonomous generative agents powered by large language models promise to simulate intricate social behaviors in virtual societies. However, achieving real-time interactions with humans at a low computational cost remains challenging. Here, we introduce Lyfe Agents. They combine low-cost with real-time responsiveness, all while remaining intelligent and goal-oriented. Key innovations include: (1) an option-action framework, reducing the cost of high-level decisions; (2) asynchronous self-monitoring for better self-consistency; and (3) a Summarize-and-Forget memory mechanism, prioritizing critical memory items at a low cost. We evaluate Lyfe Agents' self-motivation and sociability across several multi-agent scenarios in our custom LyfeGame 3D virtual environment platform. When equipped with our brain-inspired techniques, Lyfe Agents can exhibit human-like self-motivated social reasoning. For example, the agents can solve a crime (a murder mystery) through autonomous collaboration and information exchange. Meanwhile, our techniques enabled Lyfe Agents to operate at a computational cost 10-100 times lower than existing alternatives. Our findings underscore the transformative potential of autonomous generative agents to enrich human social experiences in virtual worlds. Massachusetts Institute of Technology, Peking University, LyfeAL
20 18 ./images/lyfe_agents_generative_agents_20231003.png MetaAgents: Simulating Interactions of Human Behaviors for LLM-based Task-oriented Coordination via Collaborative Generative Agents Yuan Li, Yixuan Zhang, Lichao Sun Significant advancements have occurred in the application of Large LanguageModels (LLMs) for various tasks and social simulations. Despite this, their capac-ities to coordinate within task-oriented social contexts are under-explored. Suchcapabilities are crucial if LLMs are to effectively mimic human-like social be-havior and produce meaningful results. To bridge this gap, we introduce collab-orative generative agents, endowing LLM-based Agents with consistent behaviorpatterns and task-solving abilities. We situate these agents in a simulated job fairenvironment as a case study to scrutinize their coordination skills. We proposea novel framework that equips collaborative generative agents with human-likereasoning abilities and specialized skills. Our evaluation demonstrates that theseagents show promising performance. However, we also uncover limitations thathinder their effectiveness in more complex coordination tasks. Our work providesvaluable insights into the role and evolution of LLMs in task-oriented social sim-ulations. University of Cambridge, William & Mary, Lehigh University
21 19 ./images/metaagents_simulating_interactions_of_20231010.png On Generative Agents in Recommendation An Zhang, Yuxin Chen, Leheng Sheng, Xiang Wang, Tat-Seng Chua Recommender systems are the cornerstone of today's information dissemination, yet a disconnect between offline metrics and online performance greatly hinders their development. Addressing this challenge, we envision a recommendation simulator, capitalizing on recent breakthroughs in human-level intelligence exhibited by Large Language Models (LLMs). We propose Agent4Rec, a user simulator in recommendation, leveraging LLM-empowered generative agents equipped with user profile, memory, and actions modules specifically tailored for the recommender system. In particular, these agents' profile modules are initialized using real-world datasets (e.g. MovieLens, Steam, Amazon-Book), capturing users' unique tastes and social traits; memory modules log both factual and emotional memories and are integrated with an emotion-driven reflection mechanism; action modules support a wide variety of behaviors, spanning both taste-driven and emotion-driven actions. Each agent interacts with personalized recommender models in a page-by-page manner, relying on a pre-implemented collaborative filtering-based recommendation algorithm. We delve into both the capabilities and limitations of Agent4Rec, aiming to explore an essential research question: ``To what extent can LLM-empowered generative agents faithfully simulate the behavior of real, autonomous humans in recommender systems?'' Extensive and multi-faceted evaluations of Agent4Rec highlight both the alignment and deviation between agents and user-personalized preferences. Beyond mere performance comparison, we explore insightful experiments, such as emulating the filter bubble effect and discovering the underlying causal relationships in recommendation tasks. National University of Singapore, Tsinghua University, University of Science and Technology of China
22 20 ./images/on_generative_agents_in_20231016.png Out of One, Many: Using Language Models to Simulate Human Samples Lisa P. Argyle, Ethan C. Busby, Nancy Fulda, Joshua Gubler, Christopher Rytting, David Wingate We propose and explore the possibility that language models can be studied as effective proxies for specific human sub-populations in social science research. Practical and research applications of artificial intelligence tools have sometimes been limited by problematic biases (such as racism or sexism), which are often treated as uniform properties of the models. We show that the "algorithmic bias" within one such tool -- the GPT-3 language model -- is instead both fine-grained and demographically correlated, meaning that proper conditioning will cause it to accurately emulate response distributions from a wide variety of human subgroups. We term this property "algorithmic fidelity" and explore its extent in GPT-3. We create "silicon samples" by conditioning the model on thousands of socio-demographic backstories from real human participants in multiple large surveys conducted in the United States. We then compare the silicon and human samples to demonstrate that the information contained in GPT-3 goes far beyond surface similarity. It is nuanced, multifaceted, and reflects the complex interplay between ideas, attitudes, and socio-cultural context that characterize human attitudes. We suggest that language models with sufficient algorithmic fidelity thus constitute a novel and powerful tool to advance understanding of humans and society across a variety of disciplines. Brigham Young University
23 21 ./images/out_of_one_many_20220914.png Quantifying the Impact of Large Language Models on Collective Opinion Dynamics Chao Li, Xing Su, Haoying Han, Cong Xue, Chunmo Zheng, Chao Fan The process of opinion expression and exchange is a critical component of democratic societies. As people interact with large language models (LLMs) in the opinion shaping process different from traditional media, the impacts of LLMs are increasingly recognized and being concerned. However, the knowledge about how LLMs affect the process of opinion expression and exchange of social opinion networks is very limited. Here, we create an opinion network dynamics model to encode the opinions of LLMs, cognitive acceptability and usage strategies of individuals, and simulate the impact of LLMs on opinion dynamics in a variety of scenarios. The outcomes of the simulations inform about effective demand-oriented opinion network interventions. The results from this study suggested that the output opinion of LLMs has a unique and positive effect on the collective opinion difference. The marginal effect of cognitive acceptability on collective opinion formation is nonlinear and shows a decreasing trend. When people partially rely on LLMs, the exchange process of opinion becomes more intense and the diversity of opinion becomes more favorable. In fact, there is 38.6% more opinion diversity when people all partially rely on LLMs, compared to prohibiting the use of LLMs entirely. The optimal diversity of opinion was found when the fractions of people who do not use, partially rely on, and fully rely on LLMs reached roughly 4:12:1. Our experiments also find that introducing extra agents with opposite/neutral/random opinions, we can effectively mitigate the impact of biased/toxic output from LLMs. Our findings provide valuable insights into opinion dynamics in the age of LLMs, highlighting the need for customized interventions tailored to specific scenarios to address the drawbacks of improper output and use of LLMs. Zhejiang University, Clemson University,
24 22 ./images/quantifying_the_impact_of_20230807.png S3: Social-network Simulation System with Large Language Model-Empowered Agents Chen Gao, Xiaochong Lan, Zhihong Lu, Jinzhu Mao, Jinghua Piao, Huandong Wang, Depeng Jin, Yong Li Simulation plays a crucial role in addressing various challenges within socialscience. It offers extensive applications such as state prediction, phenomena ex-planation, and policy-making support, among others. In this work, we harness thehuman-like capabilities of large language models (LLMs) in sensing, reasoning,and behaving, and utilize these qualities to construct the S3 system (short forSocial network Simulation System). Adhering to the widely employed agent-basedsimulation paradigm, we employ fine-tuning and prompt engineering techniques toensure that the agent’s behavior closely emulates that of a genuine human withinthe social network. Specifically, we simulate three pivotal aspects: emotion, at-titude, and interaction behaviors. By endowing the agent in the system with theability to perceive the informational environment and emulate human actions, weobserve the emergence of population-level phenomena, including the propagationof information, attitudes, and emotions. We conduct an evaluation encompassingtwo levels of simulation, employing real-world social network data. Encouragingly,the results demonstrate promising accuracy. This work represents an initial step inthe realm of social network simulation empowered by LLM-based agents. We an-ticipate that our endeavors will serve as a source of inspiration for the developmentof simulation systems within, but not limited to, social science. Tsinghua University
25 23 ./images/s3_social-network_simulation_system_20230727.png Simulating Opinion Dynamics with Networks of LLM-based Agents Yun-Shiuan Chuang, Agam Goyal, Nikunj Harlalka, Siddharth Suresh, Robert Hawkins, Sijia Yang, Dhavan Shah, Junjie Hu, Timothy T. Rogers Accurately simulating human opinion dynam-ics is crucial for understanding a variety of soci-etal phenomena, including polarization and thespread of misinformation. However, the agent-based models (ABMs) commonly used for suchsimulations often over-simplify human behav-ior. We propose a new approach to simulat-ing opinion dynamics based on populations ofLarge Language Models (LLMs). Our findingsreveal a strong inherent bias in LLM agents to-wards producing accurate information, leadingsimulated agents to consensus in line with sci-entific reality. This bias limits their utility forunderstanding resistance to consensus viewson issues like climate change. After induc-ing confirmation bias through prompt engineer-ing, however, we observed opinion fragmenta-tion in line with existing agent-based modelingand opinion dynamics research. These insightshighlight the promise and limitations of LLMagents in this domain and suggest a path for-ward: refining LLMs with real-world discourseto better simulate the evolution of human be-liefs. University of Wisconsin-Madison
26 24 ./images/simulating_opinion_dynamics_with_20231116.png Simulating Social Media Using Large Language Models to Evaluate Alternative News Feed Algorithms Petter Törnberg, Diliara Valeeva, Justus Uitermark, Christopher Bail . Social media is often criticized for amplifyingtoxic discourse and discouraging constructive conversa-tions. But designing social media platforms to promotebetter conversations is inherently challenging. This paperasks whether simulating social media through a combina-tion of Large Language Models (LLM) and Agent-BasedModeling can help researchers study how different newsfeed algorithms shape the quality of online conversations.We create realistic personas using data from the Ameri-can National Election Study to populate simulated socialmedia platforms. Next, we prompt the agents to readand share news articles — and like or comment uponeach other’s messages — within three platforms that usedifferent news feed algorithms. In the first platform, userssee the most liked and commented posts from users whomthey follow. In the second, they see posts from all users —even those outside their own network. The third platformemploys a novel “bridging” algorithm that highlights poststhat are liked by people with opposing political views. Wefind this bridging algorithm promotes more constructive,non-toxic, conversation across political divides than theother two models. Though further research is needed toevaluate these findings, we argue that LLMs hold consid-erable potential to improve simulation research on socialmedia and many other complex social settings. University of Amsterdam, Duke University
27 25 ./images/simulating_social_media_using_20231005.png Social Simulacra: Creating Populated Prototypes for Social Computing Systems Joon Sung Park, Lindsay Popowski, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein Social computing prototypes probe the social behaviors that mayarise in an envisioned system design. This prototyping practiceis currently limited to recruiting small groups of people. Unfortu-nately, many challenges do not arise until a system is populatedat a larger scale. Can a designer understand how a social systemmight behave when populated, and make adjustments to the de-sign before the system falls prey to such challenges? We intro-duce social simulacra, a prototyping technique that generates abreadth of realistic social interactions that may emerge when a so-cial computing system is populated. Social simulacra take as inputthe designer’s description of a community’s design—goal, rules, andmember personas—and produce as output an instance of that designwith simulated behavior, including posts, replies, and anti-socialbehaviors. We demonstrate that social simulacra shift the behaviorsthat they generate appropriately in response to design changes, andthat they enable exploration of “what if?” scenarios where commu-nity members or moderators intervene. To power social simulacra,we contribute techniques for prompting a large language modelto generate thousands of distinct community members and theirsocial interactions with each other; these techniques are enabled bythe observation that large language models’ training data alreadyincludes a wide variety of positive and negative behavior on socialmedia platforms. In evaluations, we show that participants are of-ten unable to distinguish social simulacra from actual communitybehavior and that social computing designers successfully refinetheir social computing designs when using social simulacra. Stanford University, Google Research
28 26 ./images/social_simulacra_creating_populated_20220808.png The Wisdom of Partisan Crowds: Comparing Collective Intelligence in Humans and LLM-based Agents Yun-Shiuan Chuang, Siddharth Suresh, Nikunj Harlalka, Agam Goyal, Robert Hawkins, Sijia Yang, Dhavan Shah, Junjie Hu, Timothy T. Rogers Human groups are able to converge on more accurate beliefs through deliberation,even in the presence of polarization and partisan bias — a phenomenon known asthe “wisdom of partisan crowds.” Generated agents powered by Large LanguageModels (LLMs) are increasingly used to simulate human collective behavior, yetfew benchmarks exist for evaluating their dynamics against the behavior of hu-man groups. In this paper, we examine the extent to which the wisdom of partisancrowds emerges in groups of LLM-based agents that are prompted to role-playas partisan personas (e.g., Democrat or Republican). We find that they not onlydisplay human-like partisan biases, but also converge to more accurate beliefsthrough deliberation as humans do. We then identify several factors that interferewith convergence, including the use of chain-of-thought prompt and lack of detailsin personas. Conversely, fine-tuning on human data appears to enhance conver-gence. These findings show the potential and limitations of LLM-based agents asa model of human collective intelligence. University of Wisconsin-Madison
29 27 ./images/the_wisdom_of_partisan_20231116.png To Infinity and Beyond- SHOW-1 and Showrunner Agents in Multi-Agent Simulations Philipp Maas, Frank Carey, Chris Wheeler, Edward Saatchi, Pete Billington, Jessica Yaffa Shamash In this work we present our approach to generating high-quality episodic content for IP’s (Intellectual Property) using large language models (LLMs), custom state-of- the art diffusion models and our multi-agent simulation for contextualization, story progression and behavioral control. Powerful LLMs such as GPT-4 were trained on a large corpus of TV show data which lets us believe that with the right guidance users will be able to rewrite entire seasons."That Is What Entertainment Will Look Like. Maybe people are still upset about the last season of Game of Thrones. Imagine if you could ask your A.I. to make a new ending that goes a different way and maybe even put yourself in there as a main character or something.”. Fable Studio
30 28 ./images/to_infinity_and_beyond_20230724.png Unveiling the Truth and Facilitating Change: Towards Agent-based Large-scale Social Movement Simulation Xinyi Mou, Zhongyu Wei, Xuanjing Huang Social media has emerged as a cornerstone ofsocial movements, wielding significant influ-ence in driving societal change. Simulatingthe response of the public and forecasting thepotential impact has become increasingly im-portant. However, existing methods for simu-lating such phenomena encounter challengesconcerning their efficacy and efficiency in cap-turing the behaviors of social movement par-ticipants. In this paper, we introduce a hybridframework HiSim for social media user simu-lation, wherein users are categorized into twotypes. Core users are driven by Large Lan-guage Models, while numerous ordinary usersare modeled by deductive agent-based models.We further construct a Twitter-like environmentto replicate their response dynamics followingtrigger events. Subsequently, we develop amulti-faceted benchmark SoMoSiMu-Benchfor evaluation and conduct comprehensive ex-periments across real-world datasets. Exper-imental results demonstrate the effectivenessand flexibility of our method Fudan University, Shanghai Collaborative Innovation Center of Intelligent Visual Computing
31 29 ./images/unveiling_the_truth_and_20240226.png User Behavior Simulation with Large Language Model based Agents Lei Wang, Jingsen Zhang, Hao Yang, Zhiyuan Chen, Jiakai Tang, Zeyu Zhang, Xu Chen, Yankai Lin, Ruihua Song, Wayne Xin Zhao, Jun Xu, Zhicheng Dou, Jun Wang, Ji-Rong Wen Simulating high quality user behavior data has always been a fundamental problem in human-centered applications, where the major difficulty originates from the intricate mechanism of human decision process. Recently, substantial evidences have suggested that by learning huge amounts of web knowledge, large language models (LLMs) can achieve human-like intelligence. We believe these models can provide significant opportunities to more believable user behavior simulation. To inspire such direction, we propose an LLM-based agent framework and design a sandbox environment to simulate real user behaviors. Based on extensive experiments, we find that the simulated behaviors of our method are very close to the ones of real humans. Concerning potential applications, we simulate and study two social phenomenons including (1) information cocoons and (2) user conformity behaviors. This research provides novel simulation paradigms for human-centered applications. Renmin University of China, Beijing Key Laboratory of Big Data Management and Analysis Methods, University College London
32 30 ./images/user_behavior_simulation_with_20230605.png Using Large Language Models to Simulate Multiple Humans and Replicate Human Subject Studies Gati Aher, Rosa I. Arriaga, Adam Tauman Kalai We introduce a new type of test, called a Turing Experiment (TE), for evaluating to what extent a given language model, such as GPT models, can simulate different aspects of human behavior. A TE can also reveal consistent distortions in a language model's simulation of a specific human behavior. Unlike the Turing Test, which involves simulating a single arbitrary individual, a TE requires simulating a representative sample of participants in human subject research. We carry out TEs that attempt to replicate well-established findings from prior studies. We design a methodology for simulating TEs and illustrate its use to compare how well different language models are able to reproduce classic economic, psycholinguistic, and social psychology experiments: Ultimatum Game, Garden Path Sentences, Milgram Shock Experiment, and Wisdom of Crowds. In the first three TEs, the existing findings were replicated using recent models, while the last TE reveals a "hyper-accuracy distortion" present in some language models (including ChatGPT and GPT-4), which could affect downstream applications in education and the arts. Olin College of Engineering, Georgia Tech, Microsoft Research
33 31 ./images/using_large_language_models_20220818.png War and Peace (WarAgent): Large Language Model-based Multi-Agent Simulation of World Wars Wenyue Hua, Lizhou Fan, Lingyao Li, Kai Mei, Jianchao Ji, Yingqiang Ge, Libby Hemphill, Yongfeng Zhang Can we avoid wars at the crossroads of history? This question has been pursued byindividuals, scholars, policymakers, and organizations throughout human history.In this research, we attempt to answer the question based on the recent advancesof Artificial Intelligence (AI) and Large Language Models (LLMs). We proposeWarAgent, an LLM-powered multi-agent AI system, to simulate the participatingcountries, their decisions, and the consequences, in historical international conflicts,including the World War I (WWI), the World War II (WWII), and the WarringStates Period (WSP) in Ancient China. By evaluating the simulation effectiveness,we examine the advancements and limitations of cutting-edge AI systems’ abilitiesin studying complex collective human behaviors such as international conflictsunder diverse settings. In these simulations, the emergent interactions amongagents also offer a novel perspective for examining the triggers and conditions thatlead to war. Our findings offer data-driven and AI-augmented insights that canredefine how we approach conflict resolution and peacekeeping strategies. Theimplications stretch beyond historical analysis, offering a blueprint for using AI tounderstand human history and possibly prevent future international conflicts. Codeand data are available at https://github.com/agiresearch/WarAgent. Rutgers University
34 32 ./images/war_and_peace_(waragent)_20231128.png To be Continued... Your Contributions are Welcome!

94
MultiAgentEbook/book_simulation/script.js Executable file
View File

@ -0,0 +1,94 @@
document.addEventListener("DOMContentLoaded", function() {
const csvFilePath = './book_simulation/data.csv';
function loadCSV(filePath) {
return fetch(filePath)
.then(response => response.text())
.then(text => Papa.parse(text, { header: true }).data);
}
function createFlipBook(pages) {
const container = document.getElementById('flip_book_container');
const numPages = pages.length;
let flipBookHTML = '';
let style = document.createElement('style');
let css = '';
flipBookHTML += `<input type="checkbox" id="cover_checkbox">\n`;
for (let i = 0; i < numPages - 1; i++) {
flipBookHTML += `<input type="checkbox" id="page${i + 1}_checkbox">\n`;
}
flipBookHTML += `<div id="flip_book">\n`;
flipBookHTML += `<div class="front_cover">
<label for="cover_checkbox" id="cover">
<img src="./images/4.png" alt="Book Cover" class="cover_image">
</label>
</div>`
for (let i = 0; i < numPages - 1; i++) {
console.log(i)
const page = pages[i];
const pageIndex = i + 1;
flipBookHTML += `
<div class="page" id="page${pageIndex}">
<div class="front_page">
<label for="page${pageIndex}_checkbox"></label>
<img class="back_content" src="${page.image_path}" alt="Back content">
</div>
<div class="back_page">
<label for="page${pageIndex}_checkbox"></label>
<img class="edge_shading" src="./images/back_page_edge_shading.png" alt="Back page edge shading">
<div class="text_content">
<h1>${page.title}</h1>
<p class="author">${page.author}</p>
<p class="author">${page.affiliation}</p>
<div class="text_content_summary"><p class="summary">${page.summary}</p></div>
</div>
</div>
</div>\n`;
css += `
#page${pageIndex} {
z-index: ${numPages - i};
}
#page${pageIndex}_checkbox:checked~#flip_book #page${pageIndex} {
transform: rotateY(-180deg);
z-index: ${i + 1};
}\n`;
}
flipBookHTML += `<div class="back_cover">
<img src="./images/4a.png" alt="Back Cover" class="cover_image">
</div>`;
container.innerHTML = flipBookHTML;
style.innerHTML = css;
document.head.appendChild(style);
const md = window.markdownit();
const summaryElements = document.querySelectorAll('.summary');
summaryElements.forEach(el => {
el.innerHTML = md.render(el.textContent);
});
}
loadCSV(csvFilePath).then(pages => {
createFlipBook(pages);
});
});

23
MultiAgentEbook/book_simulation_index.html Executable file
View File

@ -0,0 +1,23 @@
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Flip Book</title>
<link rel="stylesheet" href="./book_style.css">
<style>
body {
background-color: transparent;
}
</style>
</head>
<body>
<div id="flip_book_container"></div>
<script src="https://cdnjs.cloudflare.com/ajax/libs/PapaParse/5.3.0/papaparse.min.js"></script>
<script src="https://cdn.jsdelivr.net/npm/markdown-it/dist/markdown-it.min.js"></script>
<script src="./book_simulation/script.js"></script>
</body>
</html>

238
MultiAgentEbook/book_style.css Executable file
View File

@ -0,0 +1,238 @@
*, *::before, *::after {
box-sizing: border-box;
margin: 0;
padding: 0;
font-family: "Arial Black";
color: rgb(25, 25, 25);
}
body {
display: flex;
flex-direction: column;
justify-content: center;
align-items: center;
width: 100%;
height: 100vh;
background-color: rgb(245, 245, 245);
}
input {
display: none;
}
#flip_book {
position: relative;
width: 596px;
height: 840px;
transition-duration: 1s;
perspective: 2000px;
}
.front_cover, .back_cover {
display: flex;
justify-content: center;
align-items: center;
width: 100%;
height: 100%;
border-radius: 2.5px 5px 5px 2.5px;
background-image: url(./images/flip_book_edge_shading.png);
background-size: cover;
background-position: center;
background-repeat: no-repeat;
background-color: rgb(255, 255, 255);
box-shadow: 0 0 5px 0 rgb(25, 25, 25, 0.25);
}
.front_cover {
position: absolute;
cursor: pointer;
transform-origin: center left;
transition: transform 0.5s;
z-index: 99;
}
.front_cover label {
position: absolute;
width: 100%;
height: 100%;
cursor: pointer;
}
.page {
position: absolute;
top: 10px;
left: 1px;
width: 576px;
height: 820px;
border-radius: 0 5px 5px 0;
background-color: white;
transform-origin: left;
transform-style: preserve-3d;
transform: rotateY(0deg);
transition-duration: 0.5s;
}
.front_page {
position: absolute;
width: 100%;
height: 100%;
backface-visibility: hidden;
}
.front_page label {
position: absolute;
width: 100%;
height: 100%;
cursor: pointer;
z-index: 100;
}
.back_page {
position: absolute;
width: 100%;
height: 100%;
backface-visibility: hidden;
transform: rotateY(180deg);
z-index: 100;
}
.back_page label {
position: absolute;
width: 100%;
height: 100%;
cursor: pointer;
z-index: 100;
}
.edge_shading {
position: absolute;
width: 576px;
height: 820px;
z-index: 98;
}
.front_content {
position: absolute;
top: 1px;
width: 574px;
height: 796px;
border-radius: 0 5px 5px 0;
z-index: 97;
}
.back_content {
position: absolute;
top: 1px;
left: 1px;
width: 574px;
height: 796px;
border-radius: 5px 0 0 5px;
z-index: 97;
}
.back_cover {
position: relative;
z-index: -1;
}
.welcome_text {
display: flex;
justify-content: center;
align-items: center;
width: 100%;
height: 100%;
font-size: 2em;
color: rgb(25, 25, 25);
}
.text_content {
display: flex;
flex-direction: column;
justify-content: center;
align-items: justify;
width: 100%;
height: 100%;
padding: 20px;
box-sizing: border-box;
text-align: center;
}
.text_content h1, .text_content p.author {
margin: 0;
padding: 10px 0;
}
.text_content p.author {
font-style: italic;
color: #555;
}
.text_content p.summary {
text-align: justify;
text-align-last: left;
max-width: 800px;
margin: 20px 0;
line-height: 1.6;
overflow-wrap: break-word;
hyphens: auto;
font-size: smaller;
}
.text_content h1 {
font-size: 24px;
}
.thank_you_text {
display: flex;
justify-content: center;
align-items: center;
width: 100%;
height: 100%;
font-size: 2em;
color: rgb(25, 25, 25);
}
#cover_checkbox:checked~#flip_book {
transform: translateX(288px)
}
#cover_checkbox:checked~#flip_book .front_cover {
transform: rotateY(-180deg);
transition: transform 1.5s, z-index 0.5s 0.5s;
z-index: 0;
}
#cover_checkbox:checked~#flip_book {
transform: translateX(288px);
}
#cover_checkbox:checked~#flip_book .front_cover {
transform: rotateY(-180deg);
transition: transform 1.5s, z-index 0.5s 0.5s;
z-index: 0;
}
.cover_image {
width: 100%;
height: 100%;
object-fit: cover;
border-radius: 2.5px 5px 5px 2.5px;
}
.back_cover .cover_image {
width: 100%;
height: 100%;
object-fit: cover;
border-radius: 2.5px 5px 5px 2.5px;
}
.text_content_summary {
text-align: left;
display: inline-block;
width: 100%;
}

162
MultiAgentEbook/communication.html Normal file
View File

@ -0,0 +1,162 @@
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<meta http-equiv="Cache-Control" content="no-store">
<meta http-equiv="Pragma" content="no-cache">
<meta http-equiv="Expires" content="0">
<link rel="icon" type="image/png" sizes="32x32" href="./images/logo.png" />
<title>§1: Communication</title>
<link rel="preconnect" href="https://fonts.gstatic.com" />
<link href="https://fonts.googleapis.com/css2?family=Rubik:wght@400;500&display=swap" rel="stylesheet" />
<link rel="stylesheet" href="style.css" />
<script src="https://cdnjs.cloudflare.com/ajax/libs/PapaParse/5.3.0/papaparse.min.js"></script>
<link rel="stylesheet" href="https://cdn.datatables.net/1.11.5/css/jquery.dataTables.min.css" />
<script src="https://code.jquery.com/jquery-3.6.0.min.js"></script>
<script src="https://cdn.datatables.net/1.11.5/js/jquery.dataTables.min.js"></script>
<style>
.table-container {
overflow-x: auto;
}
.paper-table {
width: 100%;
border-collapse: collapse;
margin: 20px 0;
}
.paper-table th,
.paper-table td {
padding: 12px;
border: 1px solid #ddd;
text-align: left;
}
.paper-table th {
cursor: pointer;
background-color: #f4f4f4;
}
.paper-table th:hover {
background-color: #e0e0e0;
}
.sorting:after {
content: ' \25B2';
}
.sorting_asc:after {
content: ' \25B2';
}
.sorting_desc:after {
content: ' \25BC';
}
#interactiveBookContainer {
width: 1250px;
height: 850px;
margin: 0 auto;
}
iframe {
width: 100%;
height: 100%;
border: none;
}
</style>
</head>
<body>
<header>
<div class="container">
</div>
</header>
<section class="paper-list">
<div class="container">
<div class="container" style="display: flex; align-items: center; position: relative;">
<a href="index.html" class="btn clr3" style="position: absolute; left: 0; top: 50%; transform: translateY(-50%);">← Back Homepage</a>
<div style="flex: 1; text-align: center;">
<h2 class="section-heading" style="display: inline-block; margin: 0;">§1: Communication</h2>
</div>
</div>
<p class="section-description text-center">
Task-oriented agent communication typically focuses on <b>protocol design</b> and <b>knowledge-augmented communication</b>, ensuring more effective information interaction and consensus building. Click on the ebook below to read.
</p>
<div id="interactiveBookContainer">
<iframe src="./book_communication_index.html"></iframe>
</div>
<br>
<br>
<br>
<br>
<div class="table-container">
<table id="paperTable" class="display paper-table">
<thead>
<tr>
<th>Title</th>
<th>Authors</th>
<th>Affiliations</th>
<th>Link</th>
<th>Date</th>
</tr>
</thead>
<tbody>
</tbody>
</table>
</div>
</div>
</section>
<!-- ATTRIBUTION -->
<div class="attribution">
<p>
Initiated by the <a href="https://github.com/OpenBMB/ChatDev" target="_blank">ChatDev</a> Group, Tsinghua
University
<br>Contact us via <a href="mailto:qianc62@gmail.com">qianc62@gmail.com</a>
</p>
</div>
<script>
$(document).ready(function () {
var keyword = "Communication";
Papa.parse("papers.csv", {
download: true,
header: true,
complete: function (results) {
var data = results.data;
var filteredData = data
.filter(function (row) {
return row.AwesomeListCategory && row.AwesomeListCategory === keyword;
})
.map(function (row, index) {
row.originalIndex = index;
return row;
});
$('#paperTable').DataTable({
data: filteredData,
order: [],
columns: [
{ "data": "Title" },
{ "data": "Authors" },
{ "data": "Affiliation" },
{ "data": "Url", "render": function (data) { return `<a href="${data}">Link</a>`; } },
{ "data": "Date" },
],
rowCallback: function (row, data, index) {
$(row).attr('data-original-index', data.originalIndex);
}
});
}
});
});
</script>
</body>
</html>

159
MultiAgentEbook/evolution.html Normal file
View File

@ -0,0 +1,159 @@
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<meta http-equiv="Cache-Control" content="no-store">
<meta http-equiv="Pragma" content="no-cache">
<meta http-equiv="Expires" content="0">
<link rel="icon" type="image/png" sizes="32x32" href="./images/logo.png" />
<title>§3: Evolution</title>
<link rel="preconnect" href="https://fonts.gstatic.com" />
<link href="https://fonts.googleapis.com/css2?family=Rubik:wght@400;500&display=swap" rel="stylesheet" />
<link rel="stylesheet" href="style.css" />
<script src="https://cdnjs.cloudflare.com/ajax/libs/PapaParse/5.3.0/papaparse.min.js"></script>
<link rel="stylesheet" href="https://cdn.datatables.net/1.11.5/css/jquery.dataTables.min.css" />
<script src="https://code.jquery.com/jquery-3.6.0.min.js"></script>
<script src="https://cdn.datatables.net/1.11.5/js/jquery.dataTables.min.js"></script>
<style>
.table-container {
overflow-x: auto;
}
.paper-table {
width: 100%;
border-collapse: collapse;
margin: 20px 0;
}
.paper-table th,
.paper-table td {
padding: 12px;
border: 1px solid #ddd;
text-align: left;
}
.paper-table th {
cursor: pointer;
background-color: #f4f4f4;
}
.paper-table th:hover {
background-color: #e0e0e0;
}
.sorting:after {
content: ' \25B2';
}
.sorting_asc:after {
content: ' \25B2';
}
.sorting_desc:after {
content: ' \25BC';
}
#interactiveBookContainer {
width: 1250px;
height: 850px;
margin: 0 auto;
}
iframe {
width: 100%;
height: 100%;
border: none;
}
</style>
</head>
<body>
<header>
<div class="container">
</div>
</header>
<section class="paper-list">
<div class="container">
<div class="container" style="display: flex; align-items: center; position: relative;">
<a href="index.html" class="btn clr3" style="position: absolute; left: 0; top: 50%; transform: translateY(-50%);">← Back Homepage</a>
<div style="flex: 1; text-align: center;">
<h2 class="section-heading" style="display: inline-block; margin: 0;">§3: Evolution</h2>
</div>
</div>
<p class="section-description text-center">
The evolution of multi-agent systems focuses on <b>cross-task experience accumulation</b>, enabling agents to enhance their capabilities and adapt to increasingly complex challenges. Click on the ebook below to read.
</p>
<div id="interactiveBookContainer">
<iframe src="./book_evolution_index.html"></iframe>
</div>
<br>
<br>
<br>
<br>
<div class="table-container">
<table id="paperTable" class="display paper-table">
<thead>
<tr>
<th>Title</th>
<th>Authors</th>
<th>Affiliations</th>
<th>Link</th>
<th>Date</th>
</tr>
</thead>
<tbody>
</tbody>
</table>
</div>
</div>
</section>
<!-- ATTRIBUTION -->
<div class="attribution">
<p>
Initiated by the <a href="https://github.com/OpenBMB/ChatDev" target="_blank">ChatDev</a> Group, Tsinghua
University
<br>Contact us via <a href="mailto:qianc62@gmail.com">qianc62@gmail.com</a>
</p>
</div>
<script>
$(document).ready(function () {
var keyword = "Evolution";
Papa.parse("papers.csv", {
download: true,
header: true,
complete: function (results) {
var data = results.data;
var filteredData = data
.filter(function (row) {
return row.AwesomeListCategory && row.AwesomeListCategory === keyword;
})
.map(function (row, index) {
row.originalIndex = index;
return row;
});
$('#paperTable').DataTable({
data: filteredData,
order: [],
columns: [
{ "data": "Title" },
{ "data": "Authors" },
{ "data": "Affiliation" },
{ "data": "Url", "render": function (data) { return `<a href="${data}">Link</a>`; } },
{ "data": "Date" },
],
rowCallback: function (row, data, index) {
$(row).attr('data-original-index', data.originalIndex);
}
});
}
});
});
</script>
</body>
</html>

BIN
MultiAgentEbook/images/(perhaps)_beyond_human_translation_20240520.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.7 MiB

BIN
MultiAgentEbook/images/1.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 6.1 KiB

BIN
MultiAgentEbook/images/1a.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 857 B

BIN
MultiAgentEbook/images/1b.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 51 KiB

BIN
MultiAgentEbook/images/1d.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 51 KiB

BIN
MultiAgentEbook/images/1e.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 154 KiB

BIN
MultiAgentEbook/images/2.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 11 KiB

BIN
MultiAgentEbook/images/2a.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 954 B

BIN
MultiAgentEbook/images/2b.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 78 KiB

BIN
MultiAgentEbook/images/2d.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 81 KiB

BIN
MultiAgentEbook/images/2e.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 237 KiB

BIN
MultiAgentEbook/images/3.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 6.1 KiB

BIN
MultiAgentEbook/images/360°rea_towards_a_reusable_20240408.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 609 KiB

BIN
MultiAgentEbook/images/3a.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 954 B

BIN
MultiAgentEbook/images/3b.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 45 KiB

BIN
MultiAgentEbook/images/3d.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 42 KiB

BIN
MultiAgentEbook/images/3e.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 135 KiB

BIN
MultiAgentEbook/images/4.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 5.8 KiB

BIN
MultiAgentEbook/images/4a.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 857 B

BIN
MultiAgentEbook/images/4b.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 125 KiB

BIN
MultiAgentEbook/images/4d.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 58 KiB

BIN
MultiAgentEbook/images/4e.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 196 KiB

BIN
MultiAgentEbook/images/affordable_generative_agents_20240203.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 682 KiB

BIN
MultiAgentEbook/images/agent_hospital_a_simulacrum_20240505.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 2.0 MiB

BIN
MultiAgentEbook/images/agentcf_collaborative_learning_with_20231013.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 883 KiB

BIN
MultiAgentEbook/images/agentverse_cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 334 KiB

BIN
MultiAgentEbook/images/agentverse_facilitating_multi-agent_collaboration_20230821.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.4 MiB

BIN
MultiAgentEbook/images/ai_hospital_interactive_evaluation_20240215.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 938 KiB

BIN
MultiAgentEbook/images/apollo's_oracle_retrieval-augmented_reasoning_20231208.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1022 KiB

BIN
MultiAgentEbook/images/application_cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 41 KiB

BIN
MultiAgentEbook/images/are_you_in_a_20230719.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 516 KiB

BIN
MultiAgentEbook/images/atm_adversarial_tuning_multi-agent_20240528.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 648 KiB

BIN
MultiAgentEbook/images/auto_arena_of_llms_20240530.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 604 KiB

BIN
MultiAgentEbook/images/autoform_cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 201 KiB

BIN
MultiAgentEbook/images/autogen_enabling_next-gen_llm_20230816.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.1 MiB

BIN
MultiAgentEbook/images/autonomous_agents_for_collaborative_20240621.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.2 MiB

BIN
MultiAgentEbook/images/avalon's_game_of_thoughts_20231002.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.4 MiB

BIN
MultiAgentEbook/images/back_page_edge_shading.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 22 KiB

BIN
MultiAgentEbook/images/battleagent_multi-modal_dynamic_emulation_20240423.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 3.4 MiB

BIN
MultiAgentEbook/images/beyond_natural_language_llms_20240228.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 631 KiB

3
MultiAgentEbook/images/bg-pattern.svg Executable file
View File

@ -0,0 +1,3 @@
<svg width="1391" height="456" viewBox="0 0 1391 456" fill="none" xmlns="http://www.w3.org/2000/svg">
<path d="M0 228C0 102.079 102.079 0 228 0H1391V456H228C102.079 456 0 353.921 0 228V228Z" fill="#660874"/>
</svg>
Side by Side

After

Width:  |  Height:  |  Size: 215 B

BIN
MultiAgentEbook/images/building_cooperative_embodied_agents_20230705.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.2 MiB

BIN
MultiAgentEbook/images/camel_communicative_agents_for_20230331.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 612 KiB

BIN
MultiAgentEbook/images/can_large_language_model_20240207.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.1 MiB

BIN
MultiAgentEbook/images/chain_of_agents_large_20240604.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 699 KiB

BIN
MultiAgentEbook/images/chatcoder_chat-based_refine_requirement_20231101.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 740 KiB

BIN
MultiAgentEbook/images/chatdev.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 816 KiB

BIN
MultiAgentEbook/images/chatdev_communicative_agents_for_20230716.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 970 KiB

BIN
MultiAgentEbook/images/chatdev_cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.7 MiB

BIN
MultiAgentEbook/images/chateval_cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 120 KiB

BIN
MultiAgentEbook/images/chateval_towards_better_llm-based_20230814.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1000 KiB

BIN
MultiAgentEbook/images/colearning_cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 57 KiB

BIN
MultiAgentEbook/images/comm_collaborative_multi-agent,_multi-reasoning-path_20240426.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 312 KiB

BIN
MultiAgentEbook/images/communication_cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.1 MiB

BIN
MultiAgentEbook/images/competeai_understanding_the_competition_20231026.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 592 KiB

BIN
MultiAgentEbook/images/cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 310 KiB

BIN
MultiAgentEbook/images/ctc_cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 191 KiB

BIN
MultiAgentEbook/images/dataset_cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 84 KiB

BIN
MultiAgentEbook/images/describe,_explain,_plan_and_20230203.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.2 MiB

BIN
MultiAgentEbook/images/dynamic_llm-agent_network_an_20231003.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 784 KiB

BIN
MultiAgentEbook/images/econagent_large_language_model-empowered_20231016.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 374 KiB

BIN
MultiAgentEbook/images/ei_cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 148 KiB

BIN
MultiAgentEbook/images/encouraging_divergent_thinking_in_20230530.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 889 KiB

BIN
MultiAgentEbook/images/epidemic_modeling_with_generative_20230711.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 572 KiB

BIN
MultiAgentEbook/images/evolution_cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 974 KiB

BIN
MultiAgentEbook/images/examining_inter-consistency_of_large_20230519.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1014 KiB

BIN
MultiAgentEbook/images/experiential_co-learning_of_software-developing_20231228.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 835 KiB

BIN
MultiAgentEbook/images/exploring_collaboration_mechanisms_for_20231003.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.3 MiB

BIN
MultiAgentEbook/images/exploring_large_language_models_20230909.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.3 MiB

BIN
MultiAgentEbook/images/facilitating_multi-role_and_multi-behavior_20240528.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 912 KiB

BIN
MultiAgentEbook/images/favicon.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 31 KiB

BIN
MultiAgentEbook/images/flip_book_edge_shading.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 4.5 KiB

BIN
MultiAgentEbook/images/framework_cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 143 KiB

BIN
MultiAgentEbook/images/gamegpt_multi-agent_collaborative_framework_20231012.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.9 MiB

BIN
MultiAgentEbook/images/generative_agents_interactive_simulacra_20230407.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 2.3 MiB

BIN
MultiAgentEbook/images/github.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 2.2 KiB

BIN
MultiAgentEbook/images/github_normal.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 2.1 KiB

BIN
MultiAgentEbook/images/humanoid_agents_platform_for_20231009.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 808 KiB

BIN
MultiAgentEbook/images/iagents_cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 841 KiB

1
MultiAgentEbook/images/icon-close.svg Executable file
View File

@ -0,0 +1 @@
<svg xmlns="http://www.w3.org/2000/svg" width="16" height="15"><path fill="#FFF" fill-rule="evenodd" d="M8 5.379L13.303.075l2.122 2.122L10.12 7.5l5.304 5.303-2.122 2.122L8 9.62l-5.303 5.304-2.122-2.122L5.88 7.5.575 2.197 2.697.075 8 5.38z"/></svg>
Side by Side

After

Width:  |  Height:  |  Size: 247 B

1
MultiAgentEbook/images/icon-hamburger.svg Executable file
View File

@ -0,0 +1 @@
<svg xmlns="http://www.w3.org/2000/svg" width="18" height="15"><path fill="#242A45" fill-rule="evenodd" d="M0 0h18v3H0V0zm0 6h18v3H0V6zm0 6h18v3H0v-3z"/></svg>
Side by Side

After

Width:  |  Height:  |  Size: 159 B

BIN
MultiAgentEbook/images/improving_factuality_and_reasoning_20230523.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.5 MiB

BIN
MultiAgentEbook/images/improving_language_model_negotiation_20230517.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.4 MiB

BIN
MultiAgentEbook/images/improving_multi-agent_debate_with_20240617.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 592 KiB

Some files were not shown because too many files have changed in this diff Show More