Pith. sign in

REVIEW 3 major objections 4 minor 115 references

Relational Priors as Convergence Pressure in LLM-Based Multi-Agent Systems

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Explicit relational priors in LLM multi-agent systems act mainly as convergence pressure: they make agents coordinate and agree more readily, but they do not reliably improve correctness.

desk verdict A careful, honestly-scoped empirical study of signed relational priors as convergence pressure; the central claim holds for the tested tasks, and the paper deserves a serious referee despite a thin manipulation check and no code release. read the letter →

arxiv 2608.03239 v1 pith:3CYGI27L submitted 2026-08-04 cs.CL

Ming Shen , Chao Shang , Sadat Shahriar , Devang Kulshreshtha , Yi Zhang , Sandesh Swamy , Yanjun Qi This is my paper
classification cs.CL
keywords LLMmulti-agentsystemsrelationalpriorssignednetworksconvergencepressuredebatecommonsgovernanceconsensuscorrectnesspromptinterventions
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that when you explicitly tell LLM agents to relate to one another in positive or negative terms—through a signed relation network injected into system prompts—the main effect is convergence pressure: more positive relations make agents coordinate or agree more readily. This can help when the task rewards behavioral alignment, such as sustainable commons governance or subjective consensus. It does not reliably improve accuracy: in objective QA debates, higher positivity can increase agreement while correctness-conditioned agreement stays flat or declines. The paper therefore argues that relational priors should not be a default add-on, and that explicit neutrality is itself an intervention that differs from omitting relational framing.

What carries the argument

The central object is a signed relation network: an undirected graph over agents with each edge labeled +1 for a maximally positive prior or -1 for a maximally negative prior, with positivity $\lambda$ defined as the fraction of positive edges. This network is rendered as a natural-language relational block inserted into each agent's system prompt, with relation type varying among Attitude (affective orientation), Trust (epistemic credibility), and Influence (mutual responsiveness). The mechanism does the argument's work because it lets sign, relation type, and topology be varied independently of the communication protocol, isolating relational semantics as the cause of observed convergence

What would settle it

Run the same positivity sweep with a word-count-matched relational block that is semantically meaningless—for example, statements about weather or random names with arbitrary numbers instead of agent relations. If consensus and coordination still track positivity, the convergence-pressure effect is a prompt-format artifact rather than relational semantics; a second check is whether a positive-prior condition with no actual peer-answer exchange still produces the same consensus gain, which would indicate individual susceptibility rather than group dynamics.

Watch

Extended reading notes

Core claim

The paper's central claim is that relational priors—signed statements about how agents should feel about, trust, or respond to one another, rendered in natural language and inserted into system prompts—act primarily as convergence pressure in LLM-based multi-agent systems. Holding the task protocol, message routing, and aggregation fixed, increasing the fraction of positive ties in the relation network makes agents coordinate or agree more readily: in a commons-governance simulation this makes sustainable restraint easier, and in debate it raises consensus rates. The same pressure does not translate into better truth-finding: on objective QA, higher positivity can increase consensus while co

Load-bearing premise

The load-bearing premise is that the natural-language relational statements added to each agent's system prompt are actually read by the models as the intended signed social relation, so the measured behavioral changes come from relational semantics rather than from prompt length, wording, or format.

Editorial extensions

If this is right

  • If relational priors are convergence pressure, then adding positive relational framing to coordination-heavy tasks can help agents align on mutually beneficial restraint, but only in configurations where behavioral alignment is what the task rewards.
  • On objective tasks, observed increases in consensus cannot be read as correctness gains; evaluations should report agreement and correctness-conditioned metrics such as Consensus Correctness Rate together.
  • Explicitly neutral relational statements are an active intervention rather than a harmless annotation, so comparisons against a no-prior baseline are necessary before drawing conclusions about relational effects.
  • Because effects vary by model backbone, relation type, and topology, results from one configuration do not transfer reliably; relational priors should be validated per task and model rather than deployed by default.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural next step the paper leaves implicit is testing asymmetric, private, or evolving relations: the paper's common-knowledge, fixed, symmetric setup may understate or overstate convergence pressure in settings where agents have different information about their relationships.
  • The agreement–correctness divergence in objective QA suggests relational positivity could serve as a controllable stress test for a team's tendency to collapse into consensus, complementing research on sycophancy and debate failure modes.
  • A testable extension would sweep positivity in free-form conversation rather than fixed debate rounds; the prediction is that convergence pressure persists but its magnitude depends on whether relation statements are reiterated in dialogue.
  • The paper's diagnostic framing implies that relational priors might be used deliberately to slow convergence when opinion diversity is valuable, though the paper does not test this directly.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper studies whether injecting explicit signed relational priors (Attitude, Trust, Influence) into LLM agent system prompts changes multi-agent system behavior while holding the task protocol fixed. In GovSim, increasing relational positivity generally raises a sustainability-oriented aggregate score; in multi-agent debate, positivity tends to increase consensus, but on objective QA it does not reliably improve majority-vote accuracy or consensus correctness, and may increase wrong consensus. The paper also finds that explicit neutrality is not equivalent to omitting relational framing. The authors conclude that relational priors are convergence-pressure interventions whose usefulness is task-specific, and recommend no-prior prompting as the default for accuracy-centric tasks.

Significance. If the results hold, the paper provides a useful, conservative design guideline for LLM-MAS: relational priors should be validated against a no-prior baseline and judged with correctness-conditioned metrics, not defaulted to. The study is carefully structured in several respects: outcomes (GovSim sustainability, debate consensus/accuracy) are external to the prompt manipulation; intermediate positivity levels are swept by enumerating all sign assignments; endpoint contrasts use paired question-level or run-level bootstrap CIs; and the paper consistently separates within-prior trends from NO-PRIOR comparisons. I found no circularity or fitted-parameter concern. The main weakness is the limited manipulation check for whether the natural-language relational block actually conveys signed relational semantics, which is load-bearing for the central interpretation.

major comments (3)
  1. [§3.3, Fig. 9, App. D] The central claim that signed relational priors act as convergence pressure rests on the assumption that the prompt's relational block is read by LLMs as the intended signed relational semantics (e.g., 'A trusts B' vs 'A distrusts B'), rather than as a generic positive/negative valence or a format/length cue. The only robustness test, App. D, covers one relation type (Influence), one dataset (MMLU-Pro), one model team (open-weight), and only the two homogeneous endpoints λ=0 and λ=1. The paper itself states this check is 'limited to one relation type, one dataset, one model team, the two homogeneous-sign endpoints, and two controlled prompt changes.' This is a load-bearing gap: if the models do not actually represent the signed relation network, the monotonic trends in Figs. 2, 5, and 11 could be driven by affective valence of the sign labels or by prompt-format artifacts, and the conver
  2. [§5.1, App. A.3, Table 1] The GovSim endpoint bootstrap uses only five complete runs per condition, and the paper enumerates all 5^5 resamples per endpoint. This yields very coarse confidence intervals, and the claim that '10 of the 12 contrasts have CIs entirely above zero' is consequently based on intervals whose endpoints are constrained by a small discrete sample. While the enumeration is transparent, the statistical strength of the GovSim endpoint results is limited. This is not a fatal issue, but the paper should either report sensitivity to the choice of resampling unit (e.g., agent-month vs run) or present the results as exploratory for the endpoint comparisons, with the within-prior sweep as the primary evidence.
  3. [§4.2, App. B.3, §5.2] The CCR comparison in objective QA conditions on the set of questions that reach final consensus, and this set changes with λ. The paper correctly recomputes denominators in the paired bootstrap, but the comparison of CCR between λ=0 and λ=1 is still a comparison of different subpopulations. The observed CCR decline under higher positivity could partly reflect a compositional shift in which questions reach consensus. The paper addresses this indirectly via the transition analysis in App. B.4, but the interpretation that 'higher positivity increases agreement without improving correctness' would be strengthened by reporting CCR on a fixed set of questions that reach consensus at both endpoints, or by reporting sensitivity excluding questions that are marginal.
minor comments (4)
  1. [Fig. 1 caption] Typo: 'an d' should be 'and' in 'the antagonism an d likely competition'.
  2. [§5.2] The phrase '10 of the 12 model-group–dataset–relation comparisons' could be more precise by listing the conditions or pointing to a table/figure, since the reader must otherwise parse Figs. 5–7 to verify the count.
  3. [App. A.2, Eq. (5)] The weights for the aggregate score S are stated but not justified. Since S is the primary GovSim outcome, a brief sentence explaining the choice (e.g., why survival-related weights are larger) would aid interpretation.
  4. [Related Work] The self-citation (Yao et al., 2025) is relevant but not load-bearing; consider citing it alongside other debate-sycophancy work without emphasizing it in the main text.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical prompt-intervention study whose outcome metrics are external to the manipulation; the single self-citation is not load-bearing.

full rationale

Walking the claimed derivation chain, the paper makes no derivation that reduces to its inputs. The relational prior is an input (a signed network rendered into system prompts), and the reported quantities—GovSim sustainability scores, debate consensus rate, majority-vote accuracy, and Consensus Correctness Rate—are external behavioral outcomes, not functions of the input by construction. No parameter is fitted to a subset of the data and then renamed a prediction; the positivity sweep over λ is a controlled experimental manipulation, and the endpoint and intermediate-λ comparisons are direct measurements. The central phrase "convergence pressure" is introduced as a descriptive label ('how strongly a system tends toward coordination or agreement') and then applied to summarize the observed trend; it is not a derived quantity equivalent to the input, so this is at most a mild descriptive framing, not circularity. The comparisons against NO-PRIOR and EXPLICIT-NEUTRAL provide independent baselines, and the correctness-conditioned CCR metric can diverge from consensus, which the paper explicitly demonstrates. The only self-citation, Yao et al. 2025, appears in a related-work discussion of sycophancy and is not load-bearing for any of the paper's claims; no uniqueness theorem or unverified prior result is invoked to force the conclusion. Appendix D's limited manipulation check is a validity limitation about prompt wording, not a circular step. Under the quote-and-reduction standard, no circular step can be exhibited, so the appropriate score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

There are no fitted free parameters; the lambda sweep is a controlled manipulation rather than a fit. The only hand-chosen constants are the S-score aggregation weights, which are inherited from the GovSim benchmark and do not affect the central convergence-pressure claim. The operationalization of relational semantics in natural language is the core domain assumption. No new entities beyond signed relation networks are introduced.

assumptions (3)
  • domain assumption LLM agents respond consistently to natural-language relation statements embedded in system prompts.
    The entire intervention is a prompt change; if agents do not interpret the signed relation statements as intended, the behavioral differences would be a wording artifact. Partially checked in App. D for one endpoint and one relation type.
  • domain assumption Relations are fixed, symmetric, and common knowledge throughout each interaction.
    Stated in Sections 3.2 and Limitations. This scopes the conclusions and means findings do not extend to asymmetric, private, learned, or evolving relations.
  • domain assumption The GovSim and debate protocols replicate the cited benchmark behavior.
    Environment dynamics are taken from Piatti et al. (2024) and the debate protocol from Wu et al. (2024); the paper does not independently verify implementation fidelity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Relational Priors as Convergence Pressure in LLM-Based Multi-Agent Systems." pith.science (2026). https://pith.science/paper/3CYGI27L

@misc{pith2026260803239,
  author       = {Pith},
  title        = {Pith review of: Relational Priors as Convergence Pressure in LLM-Based Multi-Agent Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3CYGI27L}},
  note         = {Machine review of arXiv:2608.03239}
}
read the original abstract

Large language model-based multi-agent systems (LLM-MAS) are designed through roles, debate protocols, and aggregation rules. These choices create implicit social expectations: agents may be expected to trust, challenge, defer to, or collaborate with peers. We study the effects of making inter-agent relation semantics explicit. We use a minimal signed-network formulation of relational priors and inject natural-language renderings into agent system prompts while holding the task protocol fixed. Across a commons-governance simulation and multi-agent debate, relational priors primarily act as convergence pressure: increasing relational positivity tends to make agents coordinate or agree more readily. This pressure can help when utility rewards behavioral alignment, as in sustainable resource governance and subjective consensus. It does not, however, reliably improve accuracy. In objective QA debates, higher positivity can increase agreement even when correctness-conditioned agreement does not improve and may decline in some settings. Effects vary by model backbone, relation type, and topology; explicit neutrality is not equivalent to omitting relational framing. We argue that relational priors should not be a default add-on for LLM-MAS. Their safer use is diagnostic and task-specific: compare against a no-prior baseline, monitor correctness-conditioned metrics when truth matters, and omit the relational layer when validation does not justify it.

Figures

Figures reproduced from arXiv: 2608.03239 by the authors.

Figure 1
Figure 1. Identical task and protocol, different relational priors. In these illustrative first-month GovSim runs, mixed positive/negative Attitude priors are accompanied by extraction choices that push the group toward collapse (left), whereas fully positive priors are accompanied by an equal, sustainable split (right). This example motivates the systematic comparisons below; it is not used to identify an internal reasoning … view at source ↗
Figure 2
Figure 2. Three-agent GovSim sustainability-oriented overall score S (higher is better) across model backbones, relation types, and values of λ. only in GovSim, where larger teams allow meaning￾ful sparse relation structures; debate uses complete connectivity to isolate how relational priors affect deliberative convergence. For every intermediate λ, we enumerate all sign assignments consistent with that positivity, so results… view at source ↗
Figure 3
Figure 3. Change in GovSim overall score S under fully positive priors (λ = 1) relative to NO-PRIOR, by model backbone, relation type, topology, and team size. NO-PRIOR scores are shown in parentheses. majority dynamics. Each revision round exposes every agent to the other agents’ previous-round re￾sponses, so the interaction structure is complete. We also use a complete relation network. We evaluate one GPT group (GPT-4.1-na… view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Three-agent GovSim overall score S under EXPLICIT-NEUTRAL, by model backbone and relation type. Dashed lines show the corresponding NO-PRIOR base￾lines. Model labels shorten GPT-4.1-mini, Qwen-3-32B, and LLaMA-3.3-70B-Instruct to “GPT-4.1-m”, “Qwen”, and “LLaMA”, respe…
Figure 5
Figure 5. Figure 5: Debate consensus rate (CR) differences in percentage points relative to NO-PRIOR. Dashed lines mark zero difference; annotations show NO-PRIOR CR. −25 −20 −15 −10 −5 0 5 74.4 MMLU-Pro GPT Group 73.1 Open-Weight Group 0% 33% 67% 100% −15 −10 −5 0 5 68.2 λ GPQA-Diamond 0…
Figure 6
Figure 6. Figure 6: Debate accuracy (ACC) differences in percentage points relative to NO-PRIOR. Dashed lines mark zero differ￾ence; annotations show NO-PRIOR ACC. the relation statements. All three paired 95% CIs excluded zero. Both changes preserve the direc￾tion of the contrast, althou…
Figure 8
Figure 8. Figure 8: Debate consensus rate (CR), accuracy (ACC), and Consensus Correctness Rate (CCR) under EXPLICIT￾NEUTRAL (sij = 0), by dataset, model group, and rela￾tion type. Values are percentages. Dashed lines show the corresponding NO-PRIOR baselines. Panel labels shorten OpinionQ…
Figure 10
Figure 10. Figure 10: Five-agent relation topologies used in GovSim. Each topology contains four relational edges and changes only which agent pairs receive explicit relational context; it does not change message routing. where ri,t is the quantity harvested by agent i at month t, and the …
Figure 9
Figure 9. Figure 9: Example GovSim system prompt for agent John under the Attitude relation type. It combines the task instruc￾tions with the relation definition and verbalized signed edges of the relation network. A GovSim Details A.1 Simulation Dynamics Resource dynamics. At the start o…
Figure 11
Figure 11. Figure 11: Five-agent GovSim overall score S across model backbones, relation types, topologies, and values of λ. Attitude Trust Influence 0 25 50 75 100 Overall (S) GPT-4.1-mini Attitude Trust Influence GPT-4.1 Attitude Trust Influence Qwen-3-32B Attitude Trust Influence LLaMA-…
Figure 12
Figure 12. Figure 12: Five-agent GovSim overall score S under EXPLICIT-NEUTRAL, by model backbone, relation type, and topology. Dashed lines show the corresponding NO-PRIOR baselines. P|I| i=1 P|I| j=1 |Ri−Rj | 2|I| P|I| i=1 Ri . • Over-usage (o) — Amount of unsustainable behavior across a…
Figure 13
Figure 13. Figure 13: Percentage changes in debate consensus rate (CR) and accuracy (ACC) at λ = 0 relative to NO-PRIOR, by model group, dataset, and relation type. Solid lines mark zero change. group reached that outcome. We therefore exam￾ine initial-to-final transitions for the open-wei…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

115 extracted references · 48 canonical work pages

  1. [1]

    Cooperate or Collapse: Emergence of Sustainable Cooperation in a Society of

    Giorgio Piatti and Zhijing Jin and Max Kleiman-Weiner and Bernhard Sch. Cooperate or Collapse: Emergence of Sustainable Cooperation in a Society of. The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=

  2. [2]

    AutoGen: Enabling Next-Gen

    Qingyun Wu and Gagan Bansal and Jieyu Zhang and Yiran Wu and Beibin Li and Erkang Zhu and Li Jiang and Xiaoyun Zhang and Shaokun Zhang and Jiale Liu and Ahmed Hassan Awadallah and Ryen W White and Doug Burger and Chi Wang , booktitle=. AutoGen: Enabling Next-Gen. 2024 , url=

    2024

  3. [4]

    ArXiv , year=

    BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agents , author=. ArXiv , year=

  4. [5]

    ArXiv , year=

    Researchy Questions: A Dataset of Multi-Perspective, Decompositional Questions for LLM Web Agents , author=. ArXiv , year=

  5. [6]

    2024 , url=

    Carlos E Jimenez and John Yang and Alexander Wettig and Shunyu Yao and Kexin Pei and Ofir Press and Karthik R Narasimhan , booktitle=. 2024 , url=

    2024

  6. [8]

    2023 , url=

    Guohao Li and Hasan Abed Al Kader Hammoud and Hani Itani and Dmitrii Khizbullin and Bernard Ghanem , booktitle=. 2023 , url=

    2023

  7. [9]

    Forty-first International Conference on Machine Learning , year=

    Improving Factuality and Reasoning in Language Models through Multiagent Debate , author=. Forty-first International Conference on Machine Learning , year=

  8. [11]

    Sirui Hong and Mingchen Zhuge and Jonathan Chen and Xiawu Zheng and Yuheng Cheng and Jinlin Wang and Ceyao Zhang and Zili Wang and Steven Ka Shing Yau and Zijuan Lin and Liyang Zhou and Chenyu Ran and Lingfeng Xiao and Chenglin Wu and J. Meta. The Twelfth International Conference on Learning Representations , year=

Show all 115 references
  1. [12]

    Yongliang Shen and Kaitao Song and Xu Tan and Dongsheng Li and Weiming Lu and Yueting Zhuang , booktitle=. Hugging. 2023 , url=

    2023

  2. [13]

    Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology , year=

    Generative Agents: Interactive Simulacra of Human Behavior , author=. Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology , year=

  3. [14]

    ArXiv , year=

    Collaborative Memory: Multi-User Memory Sharing in LLM Agents with Dynamic Access Control , author=. ArXiv , year=

  4. [15]

    The Knowledge Engineering Review , year=

    Trust in multi-agent systems , author=. The Knowledge Engineering Review , year=

  5. [16]

    , author=

    Task versus relationship conflict, team performance, and team member satisfaction: a meta-analysis. , author=. The Journal of applied psychology , year=

  6. [17]

    , author=

    Trust and team performance: A meta-analysis of main effects, moderators, and covariates. , author=. The Journal of applied psychology , year=

  7. [18]

    , author=

    Attitudes and cognitive organization. , author=. The Journal of psychology , year=

  8. [19]

    , author=

    Generalized expectancies for interpersonal trust. , author=. American Psychologist , year=

  9. [20]

    , author=

    A formal theory of social power. , author=. Psychological review , year=

  10. [21]

    Proceedings of the 40th International Conference on Machine Learning , pages =

    Whose Opinions Do Language Models Reflect? , author =. Proceedings of the 40th International Conference on Machine Learning , pages =. 2023 , editor =

    2023

  11. [22]

    2024 , url=

    Yubo Wang and Xueguang Ma and Ge Zhang and Yuansheng Ni and Abhranil Chandra and Shiguang Guo and Weiming Ren and Aaran Arulraj and Xuan He and Ziyan Jiang and Tianle Li and Max Ku and Kai Wang and Alex Zhuang and Rongqi Fan and Xiang Yue and Wenhu Chen , booktitle=. 2024 , url=

    2024

  12. [23]

    Bowman , booktitle=

    David Rein and Betty Li Hou and Asa Cooper Stickland and Jackson Petty and Richard Yuanzhe Pang and Julien Dirani and Julian Michael and Samuel R. Bowman , booktitle=. 2024 , url=

    2024

  13. [24]

    ArXiv , year=

    AgentOrchestra: Orchestrating Hierarchical Multi-Agent Intelligence with the Tool-Environment-Agent(TEA) Protocol , author=. ArXiv , year=

  14. [25]

    The Twelfth International Conference on Learning Representations , year=

    AgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors , author=. The Twelfth International Conference on Learning Representations , year=

  15. [27]

    The Eleventh International Conference on Learning Representations , year=

    Self-Consistency Improves Chain of Thought Reasoning in Language Models , author=. The Eleventh International Conference on Learning Representations , year=

  16. [28]

    Thirty-seventh Conference on Neural Information Processing Systems , year=

    Self-Refine: Iterative Refinement with Self-Feedback , author=. Thirty-seventh Conference on Neural Information Processing Systems , year=

  17. [29]

    Thirty-seventh Conference on Neural Information Processing Systems , year=

    Reflexion: language agents with verbal reinforcement learning , author=. Thirty-seventh Conference on Neural Information Processing Systems , year=

  18. [30]

    A-Mem: Agentic Memory for

    Wujiang Xu and Zujie Liang and Kai Mei and Hang Gao and Juntao Tan and Yongfeng Zhang , booktitle=. A-Mem: Agentic Memory for. 2025 , url=

    2025

  19. [31]

    The Eleventh International Conference on Learning Representations , year=

    ReAct: Synergizing Reasoning and Acting in Language Models , author=. The Eleventh International Conference on Learning Representations , year=

  20. [32]

    Thirty-seventh Conference on Neural Information Processing Systems , year=

    Toolformer: Language Models Can Teach Themselves to Use Tools , author=. Thirty-seventh Conference on Neural Information Processing Systems , year=

  21. [33]

    The Knowledge Engineering Review , year=

    Intelligent agents: theory and practice , author=. The Knowledge Engineering Review , year=

  22. [34]

    Decentralized AI , volume=

    Social power: A point missed in multi-agent, DAI and HCI , author=. Decentralized AI , volume=. 1990 , publisher=

    1990

  23. [35]

    Artificial Intelligence Review , year=

    Review on Computational Trust and Reputation Models , author=. Artificial Intelligence Review , year=

  24. [36]

    On Social Laws for Artificial Agent Societies: Off-Line Design , author=. Artif. Intell. , year=

  25. [37]

    AAAI Conference on Artificial Intelligence , year=

    On the Synthesis of Useful Social Laws for Artificial Agent Societies (Preliminary Report) , author=. AAAI Conference on Artificial Intelligence , year=

  26. [38]

    Negotiation and Cooperation in Multi-Agent Environments , author=. Artif. Intell. , year=

  27. [39]

    Group Decision and Negotiation , volume =

    Automated Negotiation: Prospects, Methods and Challenges , author =. Group Decision and Negotiation , volume =. 2001 , doi =

    2001

  28. [40]

    European Conference on Artificial Intelligence , year=

    A Social Reasoning Mechanism Based On Dependence Networks , author=. European Conference on Artificial Intelligence , year=

  29. [42]

    Vicinagearth , year=

    A survey on LLM-based multi-agent systems: workflow, infrastructure, and challenges , author=. Vicinagearth , year=

  30. [44]

    Cut the Crap: An Economical Communication Pipeline for LLM-based Multi-Agent Systems , url =

    Zhang, Guibin and Yue, Yanwei and Li, Zhixun and Yun, Sukwon and Wan, Guancheng and Wang, Kun and Cheng, Dawei and Yu, Jeffrey and Chen, Tianlong , booktitle =. Cut the Crap: An Economical Communication Pipeline for LLM-based Multi-Agent Systems , url =

  31. [45]

    The Thirteenth International Conference on Learning Representations , year=

    Mixture-of-Agents Enhances Large Language Model Capabilities , author=. The Thirteenth International Conference on Learning Representations , year=

  32. [46]

    ArXiv , year=

    S3: Social-network Simulation System with Large Language Model-Empowered Agents , author=. ArXiv , year=

  33. [47]

    ArXiv , year=

    War and Peace (WarAgent): Large Language Model-based Multi-Agent Simulation of World Wars , author=. ArXiv , year=

  34. [49]

    2024 , url=

    Xuhui Zhou and Hao Zhu and Leena Mathur and Ruohong Zhang and Haofei Yu and Zhengyang Qi and Louis-Philippe Morency and Yonatan Bisk and Daniel Fried and Graham Neubig and Maarten Sap , booktitle=. 2024 , url=

    2024

  35. [50]

    Humanities and Social Sciences Communications , year=

    Large language models empowered agent-based modeling and simulation: a survey and perspectives , author=. Humanities and Social Sciences Communications , year=

  36. [51]

    ICLR 2024 Workshop: How Far Are We From AGI , year=

    Can Large Language Model Agents Simulate Human Trust Behaviors? , author=. ICLR 2024 Workshop: How Far Are We From AGI , year=

    2024

  37. [52]

    Scientific Reports , year=

    Value-based large language model agent simulation for mutual evaluation of trust and interpersonal closeness , author=. Scientific Reports , year=

  38. [53]

    Nature Human Behaviour , year=

    Playing repeated games with large language models , author=. Nature Human Behaviour , year=

  39. [55]

    Proceedings of the ACM on Human-Computer Interaction , year=

    Multi-Agents are Social Groups: Investigating Social Influence of Multiple Agents in Human-Agent Interactions , author=. Proceedings of the ACM on Human-Computer Interaction , year=

  40. [59]

    Pedro M. P. Curvo , booktitle=. \. 2025 , url=

    2025

  41. [60]

    Schneider and LIN TIAN and Marian-Andrei Rizoiu , booktitle=

    Philipp J. Schneider and LIN TIAN and Marian-Andrei Rizoiu , booktitle=. Learning to Make Friends: Coaching. 2025 , url=

    2025

  42. [61]

    , author=

    Structural balance: a generalization of Heider's theory. , author=. Psychological review , year=

  43. [62]

    The Twelfth International Conference on Learning Representations , year=

    Towards Understanding Sycophancy in Language Models , author=. The Twelfth International Conference on Learning Representations , year=

  44. [63]

    ArXiv , year=

    Peacemaker or Troublemaker: How Sycophancy Shapes Multi-Agent Debate , author=. ArXiv , year=

  45. [64]

    ArXiv , year=

    Talk Isn't Always Cheap: Understanding Failure Modes in Multi-Agent Debate , author=. ArXiv , year=

  46. [65]

    Elif Akata, Lion Schulz, Julian Coda-Forno, Seong Joon Oh, Matthias Bethge, and Eric Schulz. 2023. https://api.semanticscholar.org/CorpusID:258947115 Playing repeated games with large language models . Nature Human Behaviour, 9:1380 -- 1390

    2023

  47. [66]

    Ariel Flint Ashery, Luca Maria Aiello, and Andrea Baronchelli. 2025. https://doi.org/10.1126/sciadv.adu9368 Emergent social conventions and collective bias in llm populations . Science Advances, 11(20):eadu9368

    2025 doi

  48. [67]

    Dorwin Cartwright and Frank Harary. 1956. https://api.semanticscholar.org/CorpusID:262457963 Structural balance: a generalization of heider's theory. Psychological review, 63 5:277--93

    1956

  49. [68]

    Cristiano Castelfranchi. 1990. Social power: A point missed in multi-agent, dai and hci. Decentralized AI, 1:33--46

    1990

  50. [69]

    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, and Jie Zhou. 2024. https://openreview.net/forum?id=EHg5GDnyq1 Agentverse: Facilitating multi-a...

    2024

  51. [70]

    Pedro M. P. Curvo. 2025. https://openreview.net/forum?id=1YjrWLAXJr \ textit\ The Traitors\ \ : Deception and trust in multi-agent language model simulations . In First Workshop on Multi-Turn Interactions in Large Language Models

    2025

  52. [71]

    Dirks, and Nicole M

    Bart de Jong, Kurt T. Dirks, and Nicole M. Gillespie. 2016. https://api.semanticscholar.org/CorpusID:207550427 Trust and team performance: A meta-analysis of main effects, moderators, and covariates. The Journal of applied psychology, 101 8:1134--50

    2016

  53. [72]

    C. D. De Dreu and Laurie R. Weingart. 2003. https://api.semanticscholar.org/CorpusID:16345021 Task versus relationship conflict, team performance, and team member satisfaction: a meta-analysis. The Journal of applied psychology, 88 4:741--9

    2003

  54. [73]

    Tenenbaum, and Igor Mordatch

    Yilun Du, Shuang Li, Antonio Torralba, Joshua B. Tenenbaum, and Igor Mordatch. 2024. https://openreview.net/forum?id=zj7YuTE4t8 Improving factuality and reasoning in language models through multiagent debate . In Forty-first International Conference on Machine Learning

    2024

  55. [74]

    Nicolò Fontana, Francesco Pierri, and Luca Maria Aiello. 2025. https://doi.org/10.1609/icwsm.v19i1.35829 Nicer than humans: How do large language models behave in the prisoner’s dilemma? Proceedings of the International AAAI Conference on Web and Social Media, 19(1):522--535

    2025 doi

  56. [75]

    John French. 1956. https://api.semanticscholar.org/CorpusID:12939333 A formal theory of social power. Psychological review, 63 3:181--94

    1956

  57. [76]

    Chen Gao, Xiaochong Lan, Zhi jie Lu, Jinzhu Mao, Jing Piao, Huandong Wang, Depeng Jin, and Yong Li. 2023 a . https://api.semanticscholar.org/CorpusID:260202947 S3: Social-network simulation system with large language model-empowered agents . ArXiv, abs/2307.14984

    2023 arXiv

  58. [77]

    Chen Gao, Xiaochong Lan, Nian Li, Yuan Yuan, Jingtao Ding, Zhilun Zhou, Fengli Xu, and Yong Li. 2023 b . https://api.semanticscholar.org/CorpusID:266362356 Large language models empowered agent-based modeling and simulation: a survey and perspectives . Humanities and Social Sc...

    2023

  59. [78]

    Chawla, Olaf Wiest, and Xiangliang Zhang

    Taicheng Guo, Xiuying Chen, Yaqi Wang, Ruidi Chang, Shichao Pei, Nitesh V. Chawla, Olaf Wiest, and Xiangliang Zhang. 2024. https://doi.org/10.24963/ijcai.2024/890 Large language model based multi-agents: A survey of progress and challenges . In Proceedings of the Thirty-Third ...

    2024 doi

  60. [79]

    Fritz Heider. 1946. https://api.semanticscholar.org/CorpusID:43393067 Attitudes and cognitive organization. The Journal of psychology, 21:107--12

    1946

  61. [80]

    Sirui Hong, Mingchen Zhuge, Jonathan Chen, Xiawu Zheng, Yuheng Cheng, Jinlin Wang, Ceyao Zhang, Zili Wang, Steven Ka Shing Yau, Zijuan Lin, Liyang Zhou, Chenyu Ran, Lingfeng Xiao, Chenglin Wu, and J \"u rgen Schmidhuber. 2024. https://openreview.net/forum?id=VtmBAGCN7o Meta GP...

    2024

  62. [81]

    Wenyue Hua, Lizhou Fan, Lingyao Li, Kai Mei, Jianchao Ji, Yingqiang Ge, Libby Hemphill, and Yongfeng Zhang. 2023. https://api.semanticscholar.org/CorpusID:265498466 War and peace (waragent): Large language model-based multi-agent simulation of world wars . ArXiv, abs/2311.17227

    2023 arXiv

  63. [82]

    N. R. Jennings, P. Faratin, A. R. Lomuscio, S. Parsons, M. J. Wooldridge, and C. Sierra. 2001. https://doi.org/10.1023/A:1008746126376 Automated negotiation: Prospects, methods and challenges . Group Decision and Negotiation, 10(2):199--215

    2001 doi

  64. [83]

    Carlos E Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik R Narasimhan. 2024. https://openreview.net/forum?id=VTF8yNQM66 SWE -bench: Can language models resolve real-world github issues? In The Twelfth International Conference on Learning Re...

    2024

  65. [84]

    Sarit Kraus. 1997. https://api.semanticscholar.org/CorpusID:2316322 Negotiation and cooperation in multi-agent environments . Artif. Intell., 94:79--97

    1997

  66. [85]

    Yihuai Lan, Zhiqiang Hu, Lei Wang, Yang Wang, Deheng Ye, Peilin Zhao, Ee-Peng Lim, Hui Xiong, and Hao Wang. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.7 LLM -based agent society investigation: Collaboration and confrontation in avalon gameplay . In Proceedings of the 20...

    2024 doi

  67. [86]

    Guohao Li, Hasan Abed Al Kader Hammoud, Hani Itani, Dmitrii Khizbullin, and Bernard Ghanem. 2023. https://openreview.net/forum?id=3IyL2XWDkG CAMEL : Communicative agents for ''mind'' exploration of large language model society . In Thirty-seventh Conference on Neural Informati...

    2023

  68. [87]

    Xinyi Li, Sai Wang, Siqi Zeng, Yu Wu, and Yi Yang. 2024 a . https://api.semanticscholar.org/CorpusID:273218743 A survey on llm-based multi-agent systems: workflow, infrastructure, and challenges . Vicinagearth, 1

    2024

  69. [88]

    Yunxuan Li, Yibing Du, Jiageng Zhang, Le Hou, Peter Grabowski, Yeqing Li, and Eugene Ie. 2024 b . https://doi.org/10.18653/v1/2024.findings-emnlp.427 Improving multi-agent debate with sparse communication topology . In Findings of the Association for Computational Linguistics:...

    2024 doi

  70. [89]

    Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Shuming Shi, and Zhaopeng Tu. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.992 Encouraging divergent thinking in large language models through multi-agent debate . In Proceedings of the 2024 ...

    2024 doi

  71. [90]

    Le, Salman Rahman, Elisa Kreiss, Marzyeh Ghassemi, and Saadia Gabriel

    Genglin Liu, Vivian T. Le, Salman Rahman, Elisa Kreiss, Marzyeh Ghassemi, and Saadia Gabriel. 2025. https://doi.org/10.18653/v1/2025.emnlp-main.325 MOSAIC : Modeling social AI for content dissemination and regulation in multi-agent simulations . In Proceedings of the 2025 Conf...

    2025 doi

  72. [91]

    Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, Shashank Gupta, Bodhisattwa Prasad Majumder, Katherine Hermann, Sean Welleck, Amir Yazdanbakhsh, and Peter Clark. 2023. https://openrev...

    2023

  73. [92]

    O’Brien, Carrie J

    Joon Sung Park, Joseph C. O’Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. 2023. https://api.semanticscholar.org/CorpusID:258040990 Generative agents: Interactive simulacra of human behavior . Proceedings of the 36th Annual ACM Symposium o...

    2023

  74. [93]

    Giorgio Piatti, Zhijing Jin, Max Kleiman-Weiner, Bernhard Sch \"o lkopf, Mrinmaya Sachan, and Rada Mihalcea. 2024. https://openreview.net/forum?id=0zWzJj6lO3 Cooperate or collapse: Emergence of sustainable cooperation in a society of LLM agents . In The Thirty-eighth Annual Co...

    2024

  75. [94]

    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, and Maosong Sun. 2024. https://doi.org/10.18653/v1/2024.acl-long.810 C hat D ev: Communicative agents for software development . ...

    2024 doi

  76. [95]

    Ramchurn, Trung Dong Huynh, and Nicholas R

    Sarvapali D. Ramchurn, Trung Dong Huynh, and Nicholas R. Jennings. 2004. https://api.semanticscholar.org/CorpusID:3078938 Trust in multi-agent systems . The Knowledge Engineering Review, 19:1 -- 25

    2004

  77. [96]

    David Rein, Betty Li Hou, Asa Cooper Stickland, Jackson Petty, Richard Yuanzhe Pang, Julien Dirani, Julian Michael, and Samuel R. Bowman. 2024. https://openreview.net/forum?id=Ti67584b98 GPQA : A graduate-level google-proof q&a benchmark . In First Conference on Language Modeling

    2024

  78. [97]

    Siyue Ren, Zhiyao Cui, Ruiqi Song, Zhen Wang, and Shuyue Hu. 2024. https://doi.org/10.24963/ijcai.2024/874 Emergence of social norms in generative agent societies: Principles and architecture . In Proceedings of the Thirty-Third International Joint Conference on Artificial Int...

    2024 doi

  79. [98]

    Alireza Rezazadeh, Zichao Li, Ange Lou, Yuying Zhao, Wei Wei, and Yujia Bao. 2025. https://api.semanticscholar.org/CorpusID:278904585 Collaborative memory: Multi-user memory sharing in llm agents with dynamic access control . ArXiv, abs/2505.18279

    2025 arXiv

  80. [99]

    Chau, Zhuo Feng, Ahmed Awadallah, Jennifer Neville, and Nikhil Rao

    Corby Rosset, Ho-Lam Chung, Guanghui Qin, Ethan C. Chau, Zhuo Feng, Ahmed Awadallah, Jennifer Neville, and Nikhil Rao. 2024. https://api.semanticscholar.org/CorpusID:268041467 Researchy questions: A dataset of multi-perspective, decompositional questions for llm web agents . A...

    2024 arXiv

  81. [100]

    Julian B. Rotter. 1971. https://api.semanticscholar.org/CorpusID:144694593 Generalized expectancies for interpersonal trust. American Psychologist, 26:443--452

    1971

  82. [101]

    Jordi Sabater-Mir and Carles Sierra. 2005. https://api.semanticscholar.org/CorpusID:2296304 Review on computational trust and reputation models . Artificial Intelligence Review, 24:33--60

    2005

  83. [102]

    Yuki Sakamoto, Takahisa Uchida, and Hiroshi Ishiguro. 2025. https://api.semanticscholar.org/CorpusID:280290349 Value-based large language model agent simulation for mutual evaluation of trust and interpersonal closeness . Scientific Reports, 15

    2025

  84. [103]

    Shibani Santurkar, Esin Durmus, Faisal Ladhak, Cinoo Lee, Percy Liang, and Tatsunori Hashimoto. 2023. https://proceedings.mlr.press/v202/santurkar23a.html Whose opinions do language models reflect? In Proceedings of the 40th International Conference on Machine Learning, volume...

    2023

  85. [104]

    Timo Schick, Jane Dwivedi-Yu, Roberto Dessi, Roberta Raileanu, Maria Lomeli, Eric Hambro, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. 2023. https://openreview.net/forum?id=Yacmpz84TH Toolformer: Language models can teach themselves to use tools . In Thirty-seventh C...

    2023

  86. [105]

    Schneider, LIN TIAN, and Marian-Andrei Rizoiu

    Philipp J. Schneider, LIN TIAN, and Marian-Andrei Rizoiu. 2025. https://openreview.net/forum?id=dylEDdHOyY Learning to make friends: Coaching LLM agents toward emergent social ties . In Workshop on Scaling Environments for Agents

    2025

  87. [106]

    Mrinank Sharma, Meg Tong, Tomasz Korbak, David Duvenaud, Amanda Askell, Samuel R. Bowman, Esin DURMUS, Zac Hatfield-Dodds, Scott R Johnston, Shauna M Kravec, Timothy Maxwell, Sam McCandlish, Kamal Ndousse, Oliver Rausch, Nicholas Schiefer, Da Yan, Miranda Zhang, and Ethan Pere...

    2024

  88. [107]

    Yongliang Shen, Kaitao Song, Xu Tan, Dongsheng Li, Weiming Lu, and Yueting Zhuang. 2023. https://openreview.net/forum?id=yHdTscY6Ci Hugging GPT : Solving AI tasks with chat GPT and its friends in hugging face . In Thirty-seventh Conference on Neural Information Processing Systems

    2023

  89. [108]

    Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik R Narasimhan, and Shunyu Yao. 2023. https://openreview.net/forum?id=vAElhFcKW6 Reflexion: language agents with verbal reinforcement learning . In Thirty-seventh Conference on Neural Information Processing Systems

    2023

  90. [109]

    Yoav Shoham and Moshe Tennenholtz. 1995. https://api.semanticscholar.org/CorpusID:1211335 On social laws for artificial agent societies: Off-line design . Artif. Intell., 73:231--252

    1995

  91. [110]

    Jaime Sim \ a o Sichman, Rosaria Conte, Cristiano Castelfranchi, and Yves Demazeau. 1997. https://api.semanticscholar.org/CorpusID:10644468 A social reasoning mechanism based on dependence networks . In European Conference on Artificial Intelligence

    1997

  92. [111]

    Tianqi Song, Yugin Tan, Zicheng Zhu, Yibin Feng, and Yi-Chieh Lee. 2024. https://api.semanticscholar.org/CorpusID:273877359 Multi-agents are social groups: Investigating social influence of multiple agents in human-agent interactions . Proceedings of the ACM on Human-Computer ...

    2024

  93. [112]

    Junlin Wang, Jue WANG, Ben Athiwaratkun, Ce Zhang, and James Zou. 2025. https://openreview.net/forum?id=h0ZfDIrj7T Mixture-of-agents enhances large language model capabilities . In The Thirteenth International Conference on Learning Representations

    2025

  94. [113]

    Yubo Wang, Xueguang Ma, Ge Zhang, Yuansheng Ni, Abhranil Chandra, Shiguang Guo, Weiming Ren, Aaran Arulraj, Xuan He, Ziyan Jiang, Tianle Li, Max Ku, Kai Wang, Alex Zhuang, Rongqi Fan, Xiang Yue, and Wenhu Chen. 2024. https://openreview.net/forum?id=y10DM6R2r3 MMLU -pro: A more...

    2024

  95. [114]

    Jason Wei, Zhiqing Sun, Spencer Papay, Scott McKinney, Jeffrey Han, Isa Fulford, Hyung Won Chung, Alexandre Passos, William Fedus, and Amelia Glaese. 2025. https://api.semanticscholar.org/CorpusID:277857238 Browsecomp: A simple yet challenging benchmark for browsing agents . A...

    2025 arXiv

  96. [115]

    Jennings

    Michael Wooldridge and Nicholas R. Jennings. 1995. https://api.semanticscholar.org/CorpusID:221342993 Intelligent agents: theory and practice . The Knowledge Engineering Review, 10:115 -- 152

    1995

  97. [116]

    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, and Chi Wang. 2024. https://openreview.net/forum?id=BAakY1hNKS Autogen: Enabling next-gen LLM applicati...

    2024

  98. [117]

    Andrea Wynn, Harsh Satija, and Gillian K Hadfield. 2025. https://api.semanticscholar.org/CorpusID:281203300 Talk isn't always cheap: Understanding failure modes in multi-agent debate . ArXiv, abs/2509.05396

    2025

  99. [118]

    Chengxing Xie, Canyu Chen, Feiran Jia, Ziyu Ye, Kai Shu, Adel Bibi, Ziniu Hu, Philip Torr, Bernard Ghanem, and Guohao Li. 2024. https://openreview.net/forum?id=jxCaWgbFp4 Can large language model agents simulate human trust behaviors? In ICLR 2024 Workshop: How Far Are We From AGI

    2024

  100. [119]

    Wujiang Xu, Zujie Liang, Kai Mei, Hang Gao, Juntao Tan, and Yongfeng Zhang. 2025. https://openreview.net/forum?id=FiM0M8gcct A-mem: Agentic memory for LLM agents . In The Thirty-ninth Annual Conference on Neural Information Processing Systems

    2025

  101. [120]

    Binwei Yao, Chao Shang, Wanyu Du, Jianfeng He, Ruixue Lian, Yi Zhang, Hang Su, Sandesh Swamy, and Yanjun Qi. 2025. https://api.semanticscholar.org/CorpusID:281675961 Peacemaker or troublemaker: How sycophancy shapes multi-agent debate . ArXiv, abs/2509.23055

    2025

  102. [121]

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik R Narasimhan, and Yuan Cao. 2023. https://openreview.net/forum?id=WE_vluYUL-X React: Synergizing reasoning and acting in language models . In The Eleventh International Conference on Learning Representations

    2023

  103. [122]

    Yanwei Yue, Guibin Zhang, Boyang Liu, Guancheng Wan, Kun Wang, Dawei Cheng, and Yiyan Qi. 2025. https://doi.org/10.18653/v1/2025.acl-long.757 M as R outer: Learning to route LLM s for multi-agent systems . In Proceedings of the 63rd Annual Meeting of the Association for Comput...

    2025 doi

  104. [123]

    Guibin Zhang, Yanwei Yue, Zhixun Li, Sukwon Yun, Guancheng Wan, Kun Wang, Dawei Cheng, Jeffrey Yu, and Tianlong Chen. 2025 a . https://proceedings.iclr.cc/paper_files/paper/2025/file/bbc461518c59a2a8d64e70e2c38c4a0e-Paper-Conference.pdf Cut the crap: An economical communicatio...

    2025

  105. [124]

    Wentao Zhang, Liang Zeng, Yuzhen Xiao, Yongcong Li, Ce Cui, Yilei Zhao, Rui Hu, Yang Liu, Yahui Zhou, and Bo An. 2025 b . https://api.semanticscholar.org/CorpusID:281658740 Agentorchestra: Orchestrating hierarchical multi-agent intelligence with the tool-environment-agent(tea)...

    2025 arXiv

  106. [125]

    Xuhui Zhou, Hao Zhu, Leena Mathur, Ruohong Zhang, Haofei Yu, Zhengyang Qi, Louis-Philippe Morency, Yonatan Bisk, Daniel Fried, Graham Neubig, and Maarten Sap. 2024. https://openreview.net/forum?id=mM7VurbA4r SOTOPIA : Interactive evaluation for social intelligence in language ...

    2024

  107. [126]

    Kunlun Zhu, Hongyi Du, Zhaochen Hong, Xiaocheng Yang, Shuyi Guo, Zhe Wang, Zhenhailong Wang, Cheng Qian, Robert Tang, Heng Ji, and Jiaxuan You. 2025. https://doi.org/10.18653/v1/2025.acl-long.421 M ulti A gent B ench : Evaluating the collaboration and competition of LLM agents...

    2025 doi

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.