Pith. sign in

REVIEW 3 major objections 6 minor 3 cited by

Training Language Models for Social Deduction with Multi-Agent Reinforcement Learning

T0 review · 3 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read This paper claims that natural-language communication among agents can be learned without human demonstrations by training language models to listen and speak through dense belief-based rewards, roughly doubling crewmate win rates in a…

desk verdict A solid training recipe whose headline win rate is carried by the supervision signal, not the speaking reward; the speaking reward has a real specification bug worth checking in the code. read the letter →

arxiv 2502.06060 v1 pith:SRWEO5F5 submitted 2025-02-09 cs.AI cs.CLcs.LGcs.MA

Bidipta Sarkar , Warren Xia , C. Karen Liu , Dorsa Sadigh This is my paper · ORCID
classification cs.AIcs.CLcs.LGcs.MA
keywords multi-agentreinforcementlearninglanguagemodelssocialdeductiongamesemergentcommunicationlisteninglossspeakingrewardAmongUsself-play
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 claims that natural-language communication among agents can be learned from scratch in a partially observable social deduction game, without any human demonstrations. The authors decompose communication into listening—training each agent to predict the true imposter from the discussion so far—and speaking—using multi-agent reinforcement learning to reward messages that raise the other crewmates' summed probability assigned to the true imposter. In a simplified Among Us environment, the combined training roughly doubles crewmate win rates relative to standard reinforcement learning alone and produces qualitative behaviors familiar from human play, such as directly accusing suspects and citing observations as evidence. The point is that dense, task-relevant belief signals can replace human dialogue data as the learning signal for multi-agent language communication.

What carries the argument

The load-bearing objects are two auxiliary training signals added to the sparse win/loss game reward. The listening loss $\mathcal{L}_L(\pi,\tau_t^i) = -\log\pi(q|\tau_t^i)$ trains each crewmate to assign high probability to the true imposter $q$ given its action-observation history at belief-query points. The speaking reward $r_s^t = B_t - B_{t'}$ with $B_t = \sum_{k\in C_t}\pi_k(q|\tau_t^k)$ rewards a message by the increase in the other living crewmates' summed belief in the true imposter after the message. These are combined with a sparse RL loss, a KL penalty that keeps generations in natural language, and a world-modeling loss that preserves the model's language abilities; one crewmate is frozen to the listening-only policy to prevent degenerate conventions. The mechanism turns the open-ended problem of saying something useful into a dense signal tied to the game's actual deduction question.

What would settle it

The claim would be falsified by a controlled experiment in which the speaking reward is replaced by a reward for truthfulness—say, matching the message's content to the speaker's actual observations—and win rates do not drop; or by showing that a speaker trained only to maximize $r_s^t$ systematically makes false accusations and still wins, which would demonstrate the reward exploits uncalibrated listener beliefs rather than genuine information transfer.

Watch

Extended reading notes

Core claim

The central discovery is that the goal of predicting a hidden world state (who is the imposter) supplies a dense reward that makes language-based communication learnable without demonstrations. Listening is trained as supervised prediction of the true imposter at belief-query timesteps; speaking is trained by the reward $r_s^t = B_t - B_{t'}$, the change in the sum of living crewmates' beliefs in the true imposter after a message. Trained crewmates reach win rates of roughly $0.51$--$0.56$ against adversarially trained imposters on the base environment, about twice the RL-only baseline, while a much larger untrained base model stays below $0.2$. The paper argues that this shows self-improving multi-agent discussion does not require task-specific human data.

Load-bearing premise

The whole scheme rests on the assumption that a message is good exactly when it increases the listeners' stated probability of the true imposter, so if those belief updates can be manipulated by confident lies, the reward will select persuasive rather than truthful communication.

Editorial extensions

If this is right

  • If correct, training an LLM crewmate with the listening loss and speaking reward yields win rates around twice those of RL-only training and well above much larger untrained models.
  • The learned discussion strategies carry across environment shapes, task counts, and player counts, including settings outside the training distribution.
  • Crewmate strategies keep their effectiveness against imposters that are adversarially trained in self-play, with win rates converging to between roughly 0.51 and 0.56.
  • The approach produces interpretable, human-like discussion behavior—accusations and evidence-giving—without any human dialogue data.
  • The same listen-and-speak decomposition could apply to other social deduction games and partially observable tasks that have a well-defined hidden question.

Reading between the lines

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

  • The speaking reward only measures belief shifts, not truthfulness, so the observed fabrication of evidence may be a direct consequence of the reward design; a natural test is to add a truthfulness penalty and check whether win rates and message honesty both improve.
  • The reported gain from adding the speaking reward over listening-plus-RL is small, which suggests a cheaper recipe for other domains might omit the RL speaking term and rely on pretrained language priors for generation.
  • The reward is computed from the same policy being trained, so it is a self-referential signal; generalizing to humans would require tracking human belief updates, which the paper does not test.
  • If the method scales to larger models, it offers a path to training language agents for coordination tasks where human demonstrations are scarce or impossible, an application the paper mentions but does not pursue.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes a method for training language-model agents in the social deduction game Among Us without human demonstrations. The method decomposes communication into listening (supervised prediction of the true imposter after discussion messages) and speaking (a reward based on the change in teammates' beliefs in the true imposter after each message), combined with PPO and a world-modeling loss. The authors report that the full method roughly doubles crewmate win rates relative to RL-only, and that trained agents exhibit human-like behaviors such as direct accusations and evidence sharing. The paper includes code and model releases.

Significance. If the method works as described, it would be a useful contribution to training LLM agents to coordinate through natural language in partially observable settings, without needing task-specific human demonstrations. The idea of grounding communication in a dense prediction reward is appealing, and the Among Us-like environment is a good testbed. The paper ships code and models, which is a strength. However, the speaking reward as defined appears to credit the speaker with changes in its own belief, and the reported error bars are min-max over three seeds, which weakens the empirical support for the 'doubling' claim and for the additive benefit of the speaking term.

major comments (3)
  1. [Section 5.2, Eq. (5)-(6)] The speaking reward r_s^t = B_t - B_t' uses B_t = sum_{k in C_t} pi_k(q|tau_k^t) over all living crewmates C_t, which includes the speaker i. Since the speaker's own post-message belief is part of B_t, the reward can be increased by messages that shift the speaker's own posterior toward the true imposter, without requiring any effect on other agents. This contradicts the stated goal in the abstract and Section 5.2 of rewarding 'influence on other agents.' The paper should either exclude the speaker from the sum (if that is the intended semantics) or justify why self-influence is a legitimate component of the reward. If the released code already excludes the speaker, then Eq. (5) is mis-specified and should be corrected; if not, the current experiments do not isolate the effect of communication on other agents.
  2. [Section 6.1, Fig. 3] The paper states that the win-rate difference between RL+L and RL+L+S is 'relatively small' (Section 6.1), but the central claim of 'doubling the win rates' refers to RL+L+S vs RL. The reported error bars are only the minimum and maximum over three seeds, with no standard errors, confidence intervals, or number of evaluation episodes. This makes it impossible to assess whether the speaking reward provides a statistically significant improvement over RL+L, or whether the differences among the curves in Fig. 4 are meaningful. Please report the evaluation protocol (number of episodes per seed, standard errors) and, if possible, provide a significance test or at least non-overlapping confidence intervals for the key comparisons.
  3. [Section 5.2 and Appendix D] The speaking reward is an internal measure based on the trained agents' own belief updates, and the paper acknowledges (Appendix D, Section 7) that agents sometimes make up evidence and state whatever is most convincing. This raises the concern that the reward may incentivize persuasive falsehoods rather than truthful, informative communication, especially since the belief probabilities are produced by the same models being trained. The paper should validate the speaking reward against an external, task-level signal (e.g., correlation with eventual voting accuracy or with the factual correctness of statements) or at least discuss this risk explicitly as a limitation of the reward definition itself. The current discussion treats deception as an emergent human-like behavior, but it may also indicate a misspecified reward.
minor comments (6)
  1. [Section 5.2] Typo: 'animposter' should be 'an imposter'.
  2. [Fig. 3 caption] The caption says 'maximum and minimum expected win rates,' but the values are observed min/max across three runs, not expected values; please rephrase.
  3. [Appendix E, Table 2] The hyperparameter lambda_BC is listed but never defined in the main text or appendix; please either define it or remove it.
  4. [Section 6.1] Please state the number of evaluation episodes used per seed for the win-rate estimates in Figs. 3-5; without this information the reader cannot gauge the precision of the reported differences.
  5. [Appendix C] The introductory sentence 'All samples below are from the perspective of the imposter' is confusing because some samples appear to be from the perspective of a crewmate; please clarify the perspective in each sample.
  6. [Section 5.3] The design choice of freezing one crewmate to the listening-only policy pi_L is central to the training and evaluation, but the paper does not analyze how sensitive the results are to this choice (e.g., varying the number of frozen agents or the frozen policy type); a brief ablation or discussion would strengthen the robustness claims.

Circularity Check

1 steps flagged · score 4.0 of 10

Eq. (6) credits the speaker's own belief change as part of the speaking reward, so the claimed other-agent influence signal is partly self-referential; the central win-rate result remains externally evaluated.

  1. self definitional [Section 5.2, Eqs. (5)-(6)]
    "let B_t be the sum of all living crewmates' beliefs, B_t = Σ_{k∈C_t} π_k(q|τ_k^t), where the q represents voting out the correct imposter, and C_t is the set of all living crewmates at time t. If t′ is the previous belief-querying timestep, then the reward for crewmate i, who just finished speaking, is r_s^t = B_t − B_t′."

    Because C_t includes every living crewmate, the speaker i is inside both sums, so r_s^t contains π_i(q|τ_i^t) − π_i(q|τ_i^{t′}). The only event between the two belief queries is i's own message, so the reward credits the speaker's change in its own belief. The policy being trained generates both the message and the post-message belief, so r_s^t can be increased by self-persuasion (e.g., making an accusation that raises the model's own posterior on q) without any other agent updating. This contradicts the paper's stated objective of 'rewarding messages based on their influence on other agents.' The reported behavior that agents 'simply make up evidence and state whatever is most convincing' (Section 6.1) is consistent with optimizing this self-referential reward.

full rationale

The central claim—that RL plus the listening loss approximately doubles crewmate win rates relative to RL-only—is supported by external win-rate evaluations against base models and RL baselines (Figs. 3-4), so it is not circular. The listening loss (Eq. 2) uses ground-truth imposter identity q as a supervised signal, which is independent of the policy's own outputs. However, the speaking reward is partially self-definitional: Eq. (5) sums beliefs over all living crewmates, including the speaker, so Eq. (6) rewards the speaker's own belief change between successive surveys. This is a concrete specification issue in the paper's own equations, not a speculation about intent. The paper itself notes the marginal gain from the speaking term is small, so this issue does not undermine the primary win-rate result, but it does weaken the claim that speaking was improved by rewarding influence on other agents. No load-bearing self-citation chain or imported uniqueness theorem was found; self-citations (Hu & Sadigh, Kwon et al., Hawkins et al.) appear only as related work. Score of 4 reflects one real self-referential reward term while the central result retains independent external content.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The method relies on several tuned hyperparameters and the assumption that belief change is a good proxy for communication success. No new structural or physical entities are introduced.

free parameters (4)
  • lambda_L (listening loss weight) = 3.0 for pi_RL+L+S, 0.1 for pi_RL+L (Table 2)
    Tuned manually to stabilize training; the paper notes different values for different policies.
  • lambda_S (speaking reward weight) = 1.0
    Tuned manually; scales the speaking reward.
  • lambda_WM (world modeling loss weight) = 1.0
    Tuned manually; added to preserve language modeling abilities and stabilize training.
  • lambda_NL (KL constraint weight) = 0.05
    Tuned manually; the paper says careful tuning is required to keep generations in natural language.
assumptions (4)
  • standard math The social deduction game is modeled as a partially observable Markov game with the additional answer structure (Q, q).
    The POMG framing in Section 3 is a standard formalization.
  • domain assumption The change in crewmates' beliefs, B_t - B_t', is a valid causal measure of a message's influence on communication.
    This is the speaking reward in Eq. (6). It assumes no confounding factors change beliefs between the two query timesteps, and that belief changes reflect useful information.
  • domain assumption The listening loss with ground-truth imposter identity produces well-calibrated beliefs that can serve as a reward signal for speaking.
    The speaking reward uses pi_k(q|tau_k^t), which is trained by the listening loss. If beliefs are miscalibrated, the reward may be misleading.
  • ad hoc to paper Freezing one crewmate to the listening-only policy pi_L is sufficient to prevent degenerate conventions without biasing the evaluation.
    Section 6.3 says training all crewmates leads to degenerate solutions, and freezing one crewmate fixes it. This is a design choice not derived from first principles.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Training Language Models for Social Deduction with Multi-Agent Reinforcement Learning." pith.science (2026). https://pith.science/paper/SRWEO5F5

@misc{pith2026250206060,
  author       = {Pith},
  title        = {Pith review of: Training Language Models for Social Deduction with Multi-Agent Reinforcement Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SRWEO5F5}},
  note         = {Machine review of arXiv:2502.06060}
}
read the original abstract

Communicating in natural language is a powerful tool in multi-agent settings, as it enables independent agents to share information in partially observable settings and allows zero-shot coordination with humans. However, most prior works are limited as they either rely on training with large amounts of human demonstrations or lack the ability to generate natural and useful communication strategies. In this work, we train language models to have productive discussions about their environment in natural language without any human demonstrations. We decompose the communication problem into listening and speaking. Our key idea is to leverage the agent's goal to predict useful information about the world as a dense reward signal that guides communication. Specifically, we improve a model's listening skills by training them to predict information about the environment based on discussions, and we simultaneously improve a model's speaking skills with multi-agent reinforcement learning by rewarding messages based on their influence on other agents. To investigate the role and necessity of communication in complex social settings, we study an embodied social deduction game based on Among Us, where the key question to answer is the identity of an adversarial imposter. We analyze emergent behaviors due to our technique, such as accusing suspects and providing evidence, and find that it enables strong discussions, doubling the win rates compared to standard RL. We release our code and models at https://socialdeductionllm.github.io/

Figures

Figures reproduced from arXiv: 2502.06060 by the authors.

Figure 1
Figure 1. Examples of the gameplay and discussion phases of [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Diagram of the embodied gameplay loop. The envi [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Win rates for crewmates trained with different [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Win rates for crewmates trained with different algorithms over different configurations of the environment, modifying [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Exploitability curves for policies over self-play it [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Chasing Moving Targets with Online Self-Play Reinforcement Learning for Safer Language Models

    cs.LG 2025-06 conditional novelty 7.0 of 10

    Online self-play between attacker and defender roles of a single LLM improves safety robustness and attack diversity across Llama and Qwen models.

  2. Learning Decentralized LLM Collaboration with Multi-Agent Actor Critic

    cs.AI 2026-01 unverdicted novelty 6.0 of 10

    Multi-agent actor-critic methods with a centralized critic improve decentralized LLM collaboration over Monte Carlo baselines in long-horizon and sparse-reward settings.

  3. AI Agent Behavioral Science

    q-bio.NC 2025-06 conditional novelty 4.0 of 10

    AI agents should be studied as behavioral entities shaped by context and interaction, not only as trained models.

Reference graph

Works this paper leans on

43 extracted references · 28 canonical work pages · cited by 3 Pith papers

  1. [1]

    Michael Ahn, Anthony Brohan, Noah Brown, Yevgen Chebotar, Omar Cortes, Byron David, Chelsea Finn, Chuyuan Fu, Keerthana Gopalakrishnan, Karol Haus- man, Alex Herzog, Daniel Ho, Jasmine Hsu, Julian Ibarz, Brian Ichter, Alex Irpan, Eric Jang, Rosario Jauregui Ruano, Kyle Jeffrey, Sally Jesmonth, Nikhil Joshi, Ryan Julian, Dmitry Kalashnikov, Yuheng Kuang, K...

    arXiv 2022

  2. [2]

    Mark Braverman, Omid Etesami, and Elchanan Mossel. 2008. Mafia: A Theoretical Study of Players and Coalitions in a Partial Information Environment.The Annals of Applied Probability 18, 3 (2008), 825–846. http://www.jstor.org/stable/25442651

    work page arXiv 2008

  3. [3]

    Sébastien Bubeck, Varun Chandrasekaran, Ronen Eldan, Johannes Gehrke, Eric Horvitz, Ece Kamar, Peter Lee, Yin Tat Lee, Yuanzhi Li, Scott Lundberg, Harsha Nori, Hamid Palangi, Marco Tulio Ribeiro, and Yi Zhang. 2023. Sparks of Artificial General Intelligence: Early experiments with GPT-4. arXiv:2303.12712 [cs.CL]

    arXiv 2023

  4. [4]

    Luca Carminati, Brian Hu Zhang, Gabriele Farina, Nicola Gatti, and Tuomas Sandholm. 2023. Hidden-Role Games: Equilibrium Concepts and Computation. arXiv:2308.16017 [cs.GT]

    work page arXiv 2023

  5. [5]

    Ho, Thomas L

    Micah Carroll, Rohin Shah, Mark K. Ho, Thomas L. Griffiths, Sanjit A. Seshia, Pieter Abbeel, and Anca Dragan. 2019. On the utility of learning about humans for human-AI coordination. Curran Associates Inc., Red Hook, NY, USA

    work page 2019

  6. [6]

    Aaron Defazio, Xingyu Yang, Harsh Mehta, Konstantin Mishchenko, Ahmed Khaled, and Ashok Cutkosky. 2024. The Road Less Scheduled. In Thirty-eighth Conference on Neural Information Processing Systems

    work page 2024

  7. [7]

    Yujian Dong, Tianyu Wu, and Chaoyang Song. 2024. Optimizing Robotic Ma- nipulation with Decision-RWKV: A Recurrent Sequence Modeling Approach for Lifelong Learning. arXiv:2407.16306 [cs.RO] https://arxiv.org/abs/2407.16306

    work page Pith review arXiv 2024

  8. [8]

    Miller, Sasha Mitts, Adithya Renduchintala, Stephen Roller, Dirk Rowe, Weiyan Shi, Joe Spisak, Alexander Wei, David Wu, Hugh Zhang, and Markus Zijlstra

    FAIR, Anton Bakhtin, Noam Brown, Emily Dinan, Gabriele Farina, Colin Fla- herty, Daniel Fried, Andrew Goff, Jonathan Gray, Hengyuan Hu, Athul Paul Jacob, Mojtaba Komeili, Karthik Konath, Minae Kwon, Adam Lerer, Mike Lewis, Alexander H. Miller, Sasha Mitts, Adithya Renduchintala, Stephen Roller, Dirk Rowe, Weiyan Shi, Joe Spisak, Alexander Wei, David Wu, H...

    2022

Show all 43 references
  1. [9]

    Frank and Noah D

    Michael C. Frank and Noah D. Goodman. 2014. Inferring word meanings by assuming that speakers are informative. Cognitive Psychology 75 (2014), 80–96. https://doi.org/10.1016/j.cogpsych.2014.08.002

    2014 doi

  2. [10]

    Ran Gong, Qiuyuan Huang, Xiaojian Ma, Yusuke Noda, Zane Durante, Zilong Zheng, Demetri Terzopoulos, Li Fei-Fei, Jianfeng Gao, and Hoi Vo. 2024. MindA- gent: Emergent Gaming Interaction. 3154–3183. https://doi.org/10.18653/v1/ 2024.findings-naacl.200

    2024 doi

  3. [11]

    Serhii Havrylov and Ivan Titov. 2017. Emergence of language with multi-agent games: learning to communicate with sequences of symbols. In Proceedings of the 31st International Conference on Neural Information Processing Systems (Long Beach, California, USA) (NIPS’17). Curran A...

    2017

  4. [12]

    Robert Hawkins, Minae Kwon, Dorsa Sadigh, and Noah Goodman. 2020. Contin- ual Adaptation for Efficient Machine Communication. In Proceedings of the 24th Conference on Computational Natural Language Learning , Raquel Fernández and Tal Linzen (Eds.). Association for Computationa...

    2020 doi

  5. [13]

    Other- Play

    Hengyuan Hu, Adam Lerer, Alex Peysakhovich, and Jakob Foerster. 2020. "Other- Play " for zero-shot coordination. In Proceedings of the 37th International Confer- ence on Machine Learning (ICML’20) . JMLR.org, Article 409, 12 pages

    2020

  6. [14]

    Hengyuan Hu and Dorsa Sadigh. 2023. Language Instructed Reinforcement Learning for Human-AI Coordination. In40th International Conference on Machine Learning (ICML)

    2023

  7. [15]

    Jerry Huang. 2024. How Well Can a Long Sequence Model Model Long Se- quences? Comparing Architechtural Inductive Biases on Long-Context Abilities. arXiv:2407.08112 [cs.LG] https://arxiv.org/abs/2407.08112

    2024 arXiv

  8. [16]

    Shengyi Huang, Rousslan Fernand Julien Dossa, Chang Ye, Jeff Braga, Dipam Chakraborty, Kinal Mehta, and João G.M. Araújo. 2022. CleanRL: High-quality Single-file Implementations of Deep Reinforcement Learning Algorithms.Journal of Machine Learning Research 23, 274 (2022), 1–18...

    2022

  9. [17]

    Wenlong Huang, Pieter Abbeel, Deepak Pathak, and Igor Mordatch. 2022. Lan- guage Models as Zero-Shot Planners: Extracting Actionable Knowledge for Em- bodied Agents. In Proceedings of the 39th International Conference on Machine Learning (Proceedings of Machine Learning Resear...

    2022

  10. [18]

    Innersloth. 2024. Among Us. https://www.innersloth.com/games/among-us/. [Online; accessed 25-February-2024]

    2024

  11. [19]

    Duéñez-Guzmán, Jayd Matyas, Alexander Sasha Vezhnevets, John P

    Kavya Kopparapu, Edgar A. Duéñez-Guzmán, Jayd Matyas, Alexander Sasha Vezhnevets, John P. Agapiou, Kevin R. McKee, Richard Everett, Janusz Marecki, Joel Z. Leibo, and Thore Graepel. 2022. Hidden Agenda: a Social Deduction Game with Diverse Learned Equilibria. arXiv:2201.01816 [cs.AI]

    2022 arXiv

  12. [20]

    Minae Kwon, Hengyuan Hu, Vivek Myers, Siddharth Karamcheti, Anca Dragan, and Dorsa Sadigh. 2024. Toward Grounded Social Reasoning. In International Conference on Robotics and Automation (ICRA)

    2024

  13. [21]

    Minae Kwon, Sang Michael Xie, Kalesha Bullard, and Dorsa Sadigh. 2023. Re- ward Design with Language Models. In International Conference on Learning Representations (ICLR)

    2023

  14. [22]

    Bolin Lai, Hongxin Zhang, Miao Liu, Aryan Pariani, Fiona Ryan, Wenqi Jia, Shirley Anugrah Hayati, James Rehg, and Diyi Yang. 2023. Werewolf Among Us: Multimodal Resources for Modeling Persuasion Behaviors in Social Deduction Games. In Findings of the Association for Computatio...

    2023

  15. [23]

    Angeliki Lazaridou, Alexander Peysakhovich, and Marco Baroni. 2017. Multi- Agent Cooperation and the Emergence of (Natural) Language. In International Conference on Learning Representations . https://openreview.net/forum?id= Hk8N3Sclg

    2017

  16. [24]

    Kevin Lin, Christopher Agia, Toki Migimatsu, Marco Pavone, and Jeannette Bohg. 2023. Text2Motion: from natural language instructions to feasible plans. Autonomous Robots (14 Nov 2023). https://doi.org/10.1007/s10514-023-10131-7

    2023 doi

  17. [25]

    Qinghua Liu, Csaba Szepesvari, and Chi Jin. 2022. Sample-Efficient Reinforce- ment Learning of Partially Observable Markov Games. In Advances in Neural Information Processing Systems, Alice H. Oh, Alekh Agarwal, Danielle Belgrave, and Kyunghyun Cho (Eds.). https://openreview.n...

    2022

  18. [26]

    McCarthy, Robert D

    William P. McCarthy, Robert D. Hawkins, Haoliang Wang, Cameron Holdaway, and Judith E. Fan. 2021. Learning to communicate about shared procedural abstractions. arXiv:2107.00077 [cs.CL]

    2021 arXiv

  19. [27]

    Suvir Mirchandani, Fei Xia, Pete Florence, Brian Ichter, Danny Driess, Montser- rat Gonzalez Arenas, Kanishka Rao, Dorsa Sadigh, and Andy Zeng. 2023. Large Language Models as General Pattern Machines. In Proceedings of the 7th Confer- ence on Robot Learning (CoRL)

    2023

  20. [28]

    Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schul- man, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Pe- ter Welinder, Paul Christiano, Jan Leike,...

    2022 arXiv

  21. [29]

    O’Brien, Carrie J

    Joon Sung Park, Joseph C. O’Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. 2023. Generative Agents: Interactive Simulacra of Human Behavior. InIn the 36th Annual ACM Symposium on User Interface Software and Technology (UIST ’23) (San Franc...

    2023

  22. [30]

    Bo Peng, Eric Alcaide, Quentin Anthony, Alon Albalak, Samuel Arcadinho, Stella Biderman, Huanqi Cao, Xin Cheng, Michael Chung, Leon Derczynski, Xingjian Du, Matteo Grella, Kranthi Gv, Xuzheng He, Haowen Hou, Przemyslaw Kazienko, Jan Kocon, Jiaming Kong, Bartłomiej Koptyra, Hay...

    2023

  23. [31]

    Bidipta Sarkar, Andy Shih, and Dorsa Sadigh. 2024. Diverse conventions for human-AI collaboration. In Proceedings of the 37th International Conference on Neural Information Processing Systems (New Orleans, LA, USA) (NIPS ’23). Curran Associates Inc., Red Hook, NY, USA, Article...

    2024

  24. [32]

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov

  25. [33]

    Natalie Shapira, Mosh Levy, Seyed Hossein Alavi, Xuhui Zhou, Yejin Choi, Yoav Goldberg, Maarten Sap, and Vered Shwartz. 2023. Clever Hans or Neural Theory of Mind? Stress Testing Social Reasoning in Large Language Models. arXiv:2305.14763 [cs.CL]

    2023 arXiv

  26. [34]

    Kaya Stechly, Matthew Marquez, and Subbarao Kambhampati. 2023. GPT-4 Doesn’t Know It’s Wrong: An Analysis of Iterative Prompting for Reasoning Problems. arXiv:2310.12397 [cs.AI]

    2023 arXiv

  27. [35]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2023. Attention Is All You Need. arXiv:1706.03762 [cs.CL] https://arxiv.org/abs/1706.03762

    2023 arXiv

  28. [36]

    Oriol Vinyals, Igor Babuschkin, Wojciech M. Czarnecki, Michaël Mathieu, Andrew Dudzik, Junyoung Chung, David Choi, Richard Powell, Timo Ewalds, Petko Georgiev, Junhyuk Oh, Dan Horgan, Manuel Kroiss, Ivo Danihelka, Aja Huang, L. Sifre, Trevor Cai, John P. Agapiou, Max Jaderberg...

    2019

  29. [37]

    Caroline Wang, Arrasy Rahman, Ishan Durugkar, Elad Liebman, and Peter Stone

  30. [38]

    Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. 2023. Voyager: An Open-Ended Embodied Agent with Large Language Models.arXiv preprint arXiv: Arxiv-2305.16291(2023)

    2023 arXiv

  31. [39]

    elimination cooldown

    Weizhe Yuan, Richard Yuanzhe Pang, Kyunghyun Cho, Xian Li, Sainbayar Sukhbaatar, Jing Xu, and Jason Weston. 2024. Self-Rewarding Language Models. arXiv:2401.10020 [cs.CL] A ENVIRONMENT DESIGN Gameplay Phase. The main gameplay loop consists of players navigating a 2D environmen...

    2024 arXiv

  32. [43]

    wait” in a room until something changes in the environment, or “go

    You: wait All agents are allowed to just “wait” in a room until something changes in the environment, or “go” to an adjacent room, taking time to travel. If there is a corpse near an agent, they can “report body” and initiate the discussion phase. Crewmates can “do task” to do...

  33. [56]

    World: You can perform any of the following actions: go north; wait; do task; go south; wait

  34. [2017]

    arXiv:1707.06347 [cs.LG]

    Proximal Policy Optimization Algorithms. arXiv:1707.06347 [cs.LG]

    arXiv

  35. [2024]

    In Advances in Neural Information Processing Systems (NeurIPS)

    N-Agent Ad Hoc Teamwork. In Advances in Neural Information Processing Systems (NeurIPS)

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.