Pith. sign in

REVIEW 3 major objections 5 minor 45 references

MultiMind: Enhancing Werewolf Agents with Multimodal Reasoning and Theory of Mind

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

Pith's one-line read Multimodal perception and a learned theory-of-mind model let a One Night Ultimate Werewolf agent win 49.8 percent of mixed-agent games and attract the fewest human votes.

desk verdict A well-engineered ONUW agent with a genuinely new ToM+MCTS integration whose win-rate gains are real but whose belief model rests on a proxy that needs a direct calibration check. read the letter →

arxiv 2504.18039 v4 pith:6OTB45FX submitted 2025-04-25 cs.AI

classification cs.AI
keywords socialdeductiongamesOneNightUltimateWerewolftheoryofmindmultimodalemotionrecognitionMonteCarloTreeSearchlargelanguagemodelagentsbeliefmodelingsuspicionminimization
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

This paper argues that a social-deduction agent plays better when it can see and hear its opponents, not just read their words. Its MultiMind system attaches facial-expression and vocal-tone labels to each statement, learns a theory-of-mind model that outputs how much each player suspects every other player of being the werewolf, and uses Monte Carlo Tree Search to choose statements that lower suspicion directed at itself. In 400 mixed-agent games the agent wins 49.8 percent of the time, versus 42.3 percent for the best text-only baseline, and in a study with human players it receives the fewest votes. The paper positions this as a step toward LLM agents that engage in human-like social reasoning across multimodal domains.

What carries the argument

The load-bearing object is the learned theory-of-mind model in the Reasoner: a 25.4-million-parameter causal Transformer that turns the game history—action triplets plus facial and vocal emotion labels—into a belief matrix $B_t[i,j]$ at each time step. This matrix converts 'what does the other player think?' from a prompt-level intuition into a callable reward signal. The Planner wraps it in Monte Carlo Tree Search: for each candidate statement the model predicts a new belief matrix, and the reward $-\sum_{j \neq i} B_{t+1}[j,i]$ propagates back through the tree, so the agent selects utterances that reduce the predicted suspicion pointed at it.

What would settle it

Set MultiMind against a held-out set of opponent agents, record each opponent's true suspicion after every statement, and compare it with the theory-of-mind model's predicted belief matrices; if the predictions are no better than chance, or if the search-chosen utterances do not reduce the opponents' measured suspicion below randomly chosen utterances, the paper's central explanation collapses.

Watch

Extended reading notes

Core claim

MultiMind's central claim is that in One Night Ultimate Werewolf, suspicion can be treated as a measurable, learnable quantity. Every player statement is parsed into action triplets (speaker, action, target) and each utterance is tagged with one of eight facial-emotion and one of eight vocal-tone labels; a small causal Transformer consumes this sequence and outputs a belief matrix $B_t[i,j]$, the probability that player $i$ believes player $j$ is the werewolf. The planning loop then searches over candidate next utterances with Monte Carlo Tree Search, scoring each by the negative sum of predicted suspicion directed at the agent after that utterance. Trained first on LLM self-play and fine-tuned on human game recordings, the pipeline wins 70.9 percent of games as the werewolf and 44.4 percent as a villager across 400 mixed-agent games (49.8 percent overall), and against human players it draws the fewest votes (0.87 average per game, 0.19 from humans). The paper takes this as evidence that multimodal emotion and second-order belief modeling improve active gameplay, not just retrospective analysis.

Load-bearing premise

The whole result leans on the learned suspicion model being accurate for opponents it was not trained on; if its predicted beliefs do not match what those opponents actually think, the search is planning against a false model and the win-rate gains would not transfer.

Editorial extensions

If this is right

  • Multimodal emotion labels earn their place in the model: removing either facial or vocal emotion raises the theory-of-mind validation loss, and removing both lowers win rates against every baseline.
  • The learned belief model is a viable substitute for asking an LLM to infer beliefs: at equal MCTS iterations it matches the LLM-based Reasoner while being far faster, freeing compute for more search.
  • The search structure itself contributes to the gain: MCTS beats random sampling, depth-first search, and breadth-first search over the same number of belief-model evaluations.
  • The performance carries over to an LLM backend that was not used in training-data generation, suggesting the suspicion model generalizes beyond the models that produced its training games.

Reading between the lines

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

  • A natural next test is whether the same perceiver-plus-belief-matrix-plus-search recipe transfers to other social games or negotiations, where the target is not a single werewolf probability but a richer attitude structure; the paper's equations only assume a suspicion distribution over players.
  • Because emotions are collapsed to eight coarse categories, the reported gains may understate what continuous facial and prosodic features could contribute to belief prediction, or the emotion classifiers' errors may be the current bottleneck.
  • The suspicion-minimization objective could reward bland or self-effacing statements that reduce votes at the cost of information useful to teammates; a testable extension is to add an information-gain or team-utility term and measure whether village win rate rises further.
  • The theory-of-mind model is used as a fixed oracle inside the search; an iterative planner that updates its beliefs about opponents as the game unfolds would test whether the fixed-oracle assumption is the main limit.
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 / 5 minor

Summary. MultiMind augments an LLM agent for One Night Ultimate Werewolf (ONUW) with multimodal emotion perception (facial-expression and vocal-tone labels converted to text), a trained Transformer-based Theory of Mind (ToM) model that predicts a belief matrix B_t of each player's suspicion toward each other player, and an MCTS planner that selects the next action/emotion sequence to minimize predicted suspicion against the agent. The ToM model is trained on LLM self-play data and then fine-tuned on human gameplay data. Experiments report higher win rates and lower vote counts for MultiMind than four baselines in mixed-agent games (Table 1), ablations comparing MCTS with random/DFS/BFS planning (Table 3) and the compact ToM reasoner with an LLM reasoner (Table 2), plus an 8-participant human study (Table 4). The central claim is that adding multimodal input, second-order ToM modeling, and MCTS planning makes the agent win more often and receive fewer votes.

Significance. If the results hold, the paper provides a practical demonstration of integrating multimodal perception, second-order ToM reasoning, and planning in an LLM social-deduction agent, with an open-source implementation and a sensible two-stage training scheme. The agent-agent results show a consistent direction across multiple tables, and the comparison of the lightweight ToM reasoner against an LLM reasoner under a matched planning budget is useful evidence for the efficiency of the proposed architecture. The main source of uncertainty is whether the mechanism attributed to ToM is actually what is being measured: the MCTS objective is defined in terms of the ToM model's predictions, but those predictions are trained on LLM self-reported suspicions that are not validated against private beliefs or actual votes. The multimodal claims are also only partially supported, because agent-agent games use self-generated text emotion labels rather than real perception, and the human study confounds emotion-label communication with the overall agent design.

major comments (3)
  1. [Section 3.6.1, Eq. (15) and Section 3.4, Eq. (11)] The ground-truth labels for the ToM model are the LLM agents' self-reported suspicions after each utterance. These self-reports are not shown to correspond to the agents' private beliefs or to their eventual votes, and in a social-deduction setting an agent may answer strategically. Because the MCTS objective in Eq. (11) minimizes the ToM model's predicted suspicion B_{t+1}[j,i], the planner could be optimizing against a model of stated suspicion that is mis-specified with respect to the actual voting behavior of held-out opponents. The paper reports no calibration or correlation check between B_t and the votes cast by Gemini-2.0-Flash or the baseline agents, and the human-data fine-tuning additionally imputes missing intention annotations with GPT-4o (Section 3.6.2), another LLM-generated label source. Please add a direct alignment test, for example predicting vote distributions from B_t and comparing them with observed votes on held-out games, and report the correlation for opponents not used in training.
  2. [Section 4.1.2 and Section 3.5] In the agent-agent experiments, e_face and e_tone are text labels written by the LLM actors themselves rather than outputs of the audio/video perception pipeline, so Tables 1-3 do not test multimodal perception, only the effect of adding an explicit emotion-label communication channel. In the human study, only MultiMind's Actor emits such labels, so the lower vote count for MultiMind in Table 4 is confounded with the presence of this extra channel. Please either restrict the claims to 'text emotion labels' for agent experiments, add a control baseline that emits random or fixed emotion labels, or run an ablation in which MultiMind does not append emotion labels.
  3. [Section 4.2, Table 1 and Section 4.3, Tables 2-3] The paper reports no confidence intervals, standard errors, or significance tests for any win-rate or vote-count comparison. For example, the 61.0% versus 66.0% overall win rates in Table 3 are based on 50 games per condition, which gives a standard error of roughly 7 percentage points for each rate; differences of this size are compatible with sampling noise. Please report bootstrap confidence intervals and appropriate tests for all headline comparisons, including the mixed-agent results in Table 1 and the human study in Table 4.
minor comments (5)
  1. [Section 1] The sentence claiming to be 'the first to develop a framework that integrates multimodal information into SDG agents' should be qualified, since Section 2.2 only surveys retrospective multimodal analyses; if no active multimodal SDG agents exist, a sentence stating that explicitly would help.
  2. [Section 3.4] The maximum action-sequence depth |A_{t+1}|_max = 3 is asserted without justification; please report sensitivity to this hyperparameter or at least discuss why three statements is sufficient.
  3. [Section 4.3, Figure 5] Panel (b) appears to compare win rates against each baseline, but the caption does not define the grouping or the number of games per bar; please add a descriptive caption and error bars.
  4. [Section 3.6.2] The use of GPT-4o to fill missing intention annotations should be validated on a small subset or acknowledged as a limitation, since it introduces an additional LLM-labeled target into the human fine-tuning set.
  5. [Table 1] The table formatting is corrupted in the submitted PDF (e.g., '2900.8337.2' and '861.05 70.93380.83'); please ensure all numeric columns are clearly separated.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the ToM model is a supervised proxy and the headline results are external win rates and votes.

full rationale

The paper's derivation chain is not circular. The ToM model is trained (Eqs. 15–16) on externally elicited suspicion labels from LLM self-play and on human intention annotations from Lai et al. [13]; the MCTS reward (Eqs. 8–11) then optimizes against this learned proxy. Crucially, the paper's headline evidence—win rates in 400 mixed-agent games (Table 1), the team win-rate matrix (Figure 4), and human-study votes (Table 4)—is measured externally, not computed from the ToM model. The planner's objective is definitionally 'minimize predicted belief-matrix entries,' but the paper does not define gameplay success as that objective; it reports independent win and vote outcomes. No equation equates the target result with its own input, and no fitted parameter is renamed as a prediction: the ToM model is trained on self-play and human data, then applied to held-out opponents (Gemini-2.0-Flash and baseline agents) without fitting to the test games. The self-citations in the reference list ([14], [40]) are related-work context and are not load-bearing for the central claim. The skeptic's concern—that stated suspicion labels may not track actual votes—is a proxy-validity or generalization risk, not a definitional circularity, and does not change the finding under the paper's external evaluation.

Assumptions & free parameters 5 free parameters · 6 assumptions · 2 invented entities

The framework rests on five load-bearing assumptions: the Transformer's final hidden state summarizes the full history; LLM self-reports are treated as true beliefs; synthetic text emotion labels stand in for real multimodal signals; self-play training transfers to unseen backends; and the off-the-shelf emotion recognizers are accurate here. The MCTS objective also assumes that minimizing predicted suspicion is the right proxy for winning. These are mostly domain assumptions rather than standard-math axioms, and none is empirically validated in the paper.

free parameters (5)
  • Emotion category set = 8 categories per modality
    Face and tone emotions are discretized into 8 labels from OSUM and Emotion-LLaMA; this fixed label space shapes all training and planning and is chosen by hand.
  • MCTS exploration constant C = 1.414
    Conventional UCT constant set in Section 3.4; not fitted to data.
  • Maximum action sequence length = 3
    The planner caps the action sequence at depth 3 in Section 3.4, bounding the strategy space by design.
  • MCTS iteration count = 500
    Default budget in main experiments; the ablation shows 500 beats 200 and matches 1000, so the value is tuned on agent win rates.
  • Ground-truth belief proxy = uniform over suspected set
    Equation (15) assigns equal probability to each suspected player and uniform probability across all players when none are named; this is a hand-defined proxy for real beliefs.
assumptions (6)
  • domain assumption The final hidden state of the causal Transformer is sufficient to predict all players' belief matrices.
    Equations (5)-(6) compute B_t from h_Nt only; no pooling over intermediate hidden states is used.
  • ad hoc to paper LLM self-reported suspicions are valid ground truth for private beliefs.
    Equation (15) builds B_GT from what agents say they suspect after each utterance; in a deception game stated suspicion is strategic, not necessarily private.
  • ad hoc to paper Text emotion labels generated by LLMs in self-play are representative of real facial and vocal expressions.
    Section 4.1.2 states agent players use only text labels; the ToM model is trained mostly on these synthetic labels, so learning may not transfer to real signals.
  • domain assumption The ToM model trained on GPT-4o, Qwen, and Llama self-play generalizes to Gemini-2.0-Flash and to the baseline agents.
    Section 4.1.1 trains on three LLM backends then evaluates with Gemini; no belief-prediction accuracy on the evaluation opponents is reported.
  • domain assumption OSUM and Emotion-LLaMA classify audio and face emotions accurately in this setting.
    Section 3.2 relies on these off-the-shelf models for human games; no accuracy or validation on ONUW recordings is provided.
  • domain assumption Minimizing predicted suspicion from Equation (11) is a valid proxy for winning the game.
    The MCTS reward is negative predicted suspicion, but victory is determined by votes; the link is assumed and only indirectly supported by win rates.
invented entities (2)
  • Belief matrix B_t
    purpose: Represents each player's suspicion probability toward every other player and serves as the optimization target for MCTS.
    The matrix is a latent model output trained on self-reported suspicions and intention annotations; no external benchmark validates it.
  • Text emotion-label channel
    purpose: Allows the Actor to append chosen emotion labels to statements so human players and agents can read them; used in both agent and human games.
    Section 3.5 says emotion labels are presented directly as text tags. This is an invented communication affordance not available to baseline agents, and no study isolates its effect.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MultiMind: Enhancing Werewolf Agents with Multimodal Reasoning and Theory of Mind." pith.science (2026). https://pith.science/paper/6OTB45FX

@misc{pith2026250418039,
  author       = {Pith},
  title        = {Pith review of: MultiMind: Enhancing Werewolf Agents with Multimodal Reasoning and Theory of Mind},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6OTB45FX}},
  note         = {Machine review of arXiv:2504.18039}
}
read the original abstract

Large Language Model (LLM) agents have demonstrated impressive capabilities in social deduction games (SDGs) like Werewolf, where strategic reasoning and social deception are essential. However, current approaches remain limited to textual information, ignoring crucial multimodal cues such as facial expressions and tone of voice that humans naturally use to communicate. Moreover, existing SDG agents primarily focus on inferring other players' identities without modeling how others perceive themselves or fellow players. To address these limitations, we use One Night Ultimate Werewolf (ONUW) as a testbed and present MultiMind, the first framework integrating multimodal information into SDG agents. MultiMind processes facial expressions and vocal tones alongside verbal content, while employing a Theory of Mind (ToM) model to represent each player's suspicion levels toward others. By combining this ToM model with Monte Carlo Tree Search (MCTS), our agent identifies communication strategies that minimize suspicion directed at itself. Through comprehensive evaluation in both agent-versus-agent simulations and studies with human players, we demonstrate MultiMind's superior performance in gameplay. Our work presents a significant advancement toward LLM agents capable of human-like social reasoning across multimodal domains.

Figures

Figures reproduced from arXiv: 2504.18039 by the authors.

Figure 1
Figure 1. An example of Theory of Mind (ToM). First-order [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The overall framework of MultiMind. The Perceiver extracts structured information from multimodal inputs. The [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The tree structure of the MCTS algorithm in the Planner. The first level of the tree consists of nodes representing [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Team Village’s win rates. We conduct 30 games for [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Ablation study on multimodal inputs. (a) Compar [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 18 canonical work pages

  1. [1]

    Baker, Rebecca Saxe, and Joshua B

    Chris L. Baker, Rebecca Saxe, and Joshua B. Tenenbaum. 2009. Action under- standing as inverse planning.Cognition113, 3 (2009), 329–349. doi:10.1016/j. cognition.2009.07.005 Reinforcement learning and higher cognition

  2. [2]

    Cristian-Paul Bara, Sky CH-Wang, and Joyce Chai. 2021. MindCraft: Theory of Mind Modeling for Situated Dialogue in Collaborative Tasks. InProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, Marie- Francine Moens, Xuanjing Huang, Lucia Specia, and Scott Wen-tau Yih (Eds.). Association for Computational Linguistics, Onlin...

  3. [3]

    1987.Intention, Plans, and Practical Reason

    Michael Bratman. 1987.Intention, Plans, and Practical Reason. Cambridge, MA: Harvard University Press, Cambridge

  4. [4]

    Buehler and Thomas H

    Moritz C. Buehler and Thomas H. Weisswange. 2020. Theory of Mind based Com- munication for Human Agent Cooperation. In2020 IEEE International Conference on Human-Machine Systems (ICHMS). 1–6. doi:10.1109/ICHMS49158.2020.9209472

  5. [5]

    Zebang Cheng, Zhi-Qi Cheng, Jun-Yan He, Kai Wang, Yuxiang Lin, Zheng Lian, Xiaojiang Peng, and Alexander Hauptmann. 2024. Emotion-LLaMA: Multimodal Emotion Recognition and Reasoning with Instruction Tuning. InAdvances in Neural Information Processing Systems, A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang (Eds.), Vol. 37...

  6. [6]

    Hauptmann

    Zebang Cheng, Shuyuan Tu, Dawei Huang, Minghan Li, Xiaojiang Peng, Zhi-Qi Cheng, and Alexander G. Hauptmann. 2024. SZTU-CMU at MER2024: Improving Emotion-LLaMA with Conv-Attention for Multimodal Emotion Recognition. In Proceedings of the 2nd International Workshop on Multimodal and Responsible Affective Computing(Melbourne VIC, Australia)(MRAC ’24). Assoc...

  7. [7]

    Xuelong Geng, Kun Wei, Qijie Shao, Shuiyun Liu, Zhennan Lin, Zhixian Zhao, Guojian Li, Wenjie Tian, Peikun Chen, Yangze Li, Pengcheng Guo, Mingchen Shao, Shuiyuan Wang, Yuang Cao, Chengyou Wang, Tianyi Xu, Yuhang Dai, Xinfa Zhu, Yue Li, Li Zhang, and Lei Xie. 2025. OSUM: Advancing Open Speech Understanding Models with Limited Resources in Academia.arXiv p...

  8. [8]

    Griffiths

    Erin Grant, Aida Nematzadeh, and Thomas L. Griffiths. 2017. How Can Memory- Augmented Neural Networks Pass a False-Belief Task?Cognitive Science(2017). https://api.semanticscholar.org/CorpusID:7340345

Show all 45 references
  1. [9]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Art...

  2. [10]

    Kristen Grauman, Andrew Westbury, Eugene Byrne, Zachary Chavis, Antonino Furnari, Rohit Girdhar, Jackson Hamburger, Hao Jiang, Miao Liu, Xingyu Liu, Miguel Martin, Tushar Nagarajan, Ilija Radosavovic, Santhosh Kumar Ramakrish- nan, Fiona Ryan, Jayant Sharma, Michael Wray, Meng...

  3. [11]

    Yuya Hirata, Michimasa Inaba, Kenichi Takahashi, Fujio Toriumi, Hirotaka Osawa, Daisuke Katagami, and Kousuke Shinoda. 2016. Werewolf Game Modeling Using Action Probabilities Based on Play Log Analysis. InComputers and Games. https://api.semanticscholar.org/CorpusID:37838481

  4. [12]

    Xuanfa Jin, Ziyan Wang, Yali Du, Meng Fang, Haifeng Zhang, and Jun Wang. 2024. Learning to Discuss Strategically: A Case Study on One Night Ultimate Werewolf. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems. https://openreview.net/forum?id=1f82rnwCbl

  5. [13]

    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. InFindings of the Association for Computation...

  6. [14]

    Yihuai Lan, Zhiqiang Hu, Lei Wang, Yang Wang, Deheng Ye, Peilin Zhao, Ee-Peng Lim, Hui Xiong, and Hao Wang. 2024. LLM-Based Agent Society Investigation: Collaboration and Confrontation in Avalon Gameplay. InProceedings of the 2024 Conference on Empirical Methods in Natural Lan...

  7. [15]

    Jin Joo Lee, Fei Sha, and Cynthia Breazeal. 2019. A Bayesian Theory of Mind Approach to Nonverbal Communication. In2019 14th ACM/IEEE International Conference on Human-Robot Interaction (HRI). 487–496. doi:10.1109/HRI.2019. 8673023

  8. [16]

    Sangmin Lee, Bolin Lai, Fiona Ryan, Bikram Boote, and James M. Rehg. 2024. Modeling Multimodal Social Interactions: New Challenges and Baselines with Densely Aligned Representations. In2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 14585–14595. doi...

  9. [17]

    Zaijing Li, Yuquan Xie, Rui Shao, Gongwei Chen, Dongmei Jiang, and Liqiang Nie

  10. [18]

    Toups Dugas, Gillian Smith, and Rose Bohrer

    Shano Liang, Max Chen, Phoebe O. Toups Dugas, Gillian Smith, and Rose Bohrer

  11. [19]

    Jonathan Light, Min Cai, Sheng Shen, and Ziniu Hu. 2023. From Text to Tactic: Evaluating LLMs Playing the Game of Avalon. InNeurIPS 2023 Foundation Models for Decision Making Workshop. https://openreview.net/forum?id=ltUrSryS0K

  12. [20]

    Weiyu Ma, Qirui Mi, Yongcheng Zeng, Xue Yan, Runji Lin, Yuqiao Wu, Jun Wang, and Haifeng Zhang. 2024. Large Language Models Play StarCraft II:Benchmarks and A Chain of Summarization Approach. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems. https...

  13. [21]

    Noritsugu Nakamura, Michimasa Inaba, Kenichi Takahashi, Fujio Toriumi, Hi- rotaka Osawa, Daisuke Katagami, and Kousuke Shinoda. 2016. Constructing a Human-like agent for the Werewolf Game using a psychological model based multiple perspectives.2016 IEEE Symposium Series on Com...

  14. [22]

    OpenAI, :, Aaron Hurst, Adam Lerer, Adam P. Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, Aleksander Mądry, Alex Baker-Whitcomb, Alex Beutel, Alex Borzunov, Alex Car- ney, Alex Chow, Alex Kirillov, Alex Nichol, Alex P...

  15. [23]

    Yiran Qin, Enshen Zhou, Qichang Liu, Zhenfei Yin, Lu Sheng, Ruimao Zhang, Yu Qiao, and Jing Shao. 2024. MP5: A Multi-modal Open-ended Embodied System in Minecraft via Active Perception. In2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 16307–16316. ...

  16. [24]

    Liang Qiu, Yizhou Zhao, Yuan Liang, Pan Lu, Weiyan Shi, Zhou Yu, and Song- Chun Zhu. 2022. Towards Socially Intelligent Agents with Mental State Transition and Human Value. InProceedings of the 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue, Oliver...

  17. [25]

    Qwen, :, An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, Keming Lu, Keqin Bao, Kexin Yang, ...

  18. [26]

    Neil Rabinowitz, Frank Perbet, Francis Song, Chiyuan Zhang, S. M. Ali Eslami, and Matthew Botvinick. 2018. Machine Theory of Mind. InProceedings of the 35th International Conference on Machine Learning (Proceedings of Machine Learning Research, Vol. 80), Jennifer Dy and Andrea...

  19. [27]

    Parkes, and Joshua B

    Jack Serrino, Max Kleiman-Weiner, David C. Parkes, and Joshua B. Tenenbaum. 2019.Finding friend and foe in multi-agent games. Curran Associates Inc., Red Hook, NY, USA

  20. [28]

    Xiao Shao, Weifu Jiang, Fei Zuo, and Mengqing Liu. 2024. SwarmBrain: Em- bodied agent for real-time strategy game StarCraft II via large language models. arXiv:2401.17749 [cs.AI] https://arxiv.org/abs/2401.17749 MM ’25, October 27–31, 2025, Dublin, Ireland. Zheng Zhang, Nuoqia...

  21. [29]

    Zijing Shi, Meng Fang, Shunfeng Zheng, Shilong Deng, Ling Chen, and Yali Du

  22. [30]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Ł ukasz Kaiser, and Illia Polosukhin. 2017. Attention is All you Need. InAdvances in Neural Information Processing Systems, I. Guyon, U. Von Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vi...

  23. [31]

    Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. 2024. Voyager: An Open-Ended Embodied Agent with Large Language Models.Transactions on Machine Learning Research (2024). https://openreview.net/forum?id=ehfRiF0R3a

  24. [32]

    Qiaosi Wang, Koustuv Saha, Eric Gregori, David Joyner, and Ashok Goel. 2021. Towards Mutual Theory of Mind in Human-AI Interaction: How Language Re- flects What Students Perceive About a Virtual Teaching Assistant. InProceedings of the 2021 CHI Conference on Human Factors in C...

  25. [33]

    Shenzhi Wang, Chang Liu, Zilong Zheng, Siyuan Qi, Shuo Chen, Qisen Yang, Andrew Zhao, Chaofei Wang, Shiji Song, and Gao Huang. 2023. Avalon’s Game of Thoughts: Battle Against Deception through Recursive Contempla- tion. arXiv:2310.01320 [cs.AI] https://arxiv.org/abs/2310.01320

  26. [34]

    Tianhe Wang and Tomoyuki Kaneko. 2018. Application of Deep Reinforcement Learning in Werewolf Game Agents.2018 Conference on Technologies and Appli- cations of Artificial Intelligence (TAAI)(2018), 28–33. https://api.semanticscholar. org/CorpusID:57191228

  27. [35]

    Dekun Wu, Haochen Shi, Zhiyuan Sun, and Bang Liu. 2024. Deciphering Digital Detectives: Understanding LLM Behaviors and Capabilities in Multi-Agent Mys- tery Games. InFindings of the Association for Computational Linguistics: ACL 2024, Lun-Wei Ku, Andre Martins, and Vivek Srik...

  28. [36]

    Shuang Wu, Liwen Zhu, Tao Yang, Shiwei Xu, Qiang Fu, Yang Wei, and Haobo Fu. 2024. Enhance Reasoning for Large Language Models in the Game Werewolf. arXiv:2402.02330 [cs.AI] https://arxiv.org/abs/2402.02330

  29. [37]

    Yuzhuang Xu, Shuo Wang, Peng Li, Fuwen Luo, Xiaolong Wang, Weidong Liu, and Yang Liu. 2023. Exploring large language models for communication games: An empirical study on werewolf.arXiv preprint arXiv:2309.04658(2023)

  30. [38]

    Zelai Xu, Chao Yu, Fei Fang, Yu Wang, and Yi Wu. 2024. Language agents with reinforcement learning for strategic play in the Werewolf game. InProceedings of the 41st International Conference on Machine Learning(Vienna, Austria)(ICML’24). JMLR.org, Article 2285, 31 pages

  31. [39]

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. ReAct: Synergizing Reasoning and Acting in Language Models. InInternational Conference on Learning Representations (ICLR)

  32. [40]

    Zheng Zhang, Yihuai Lan, Yangsen Chen, Lei Wang, Xiang Wang, and Hao Wang

  33. [41]

    Pei Zhou, Andrew Zhu, Jennifer Hu, Jay Pujara, Xiang Ren, Chris Callison- Burch, Yejin Choi, and Prithviraj Ammanabrolu. 2023. I Cast Detect Thoughts: Learning to Converse and Guide with Intents and Theory-of-Mind in Dungeons and Dragons. InProceedings of the 61st Annual Meeti...

  34. [44]

    In ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

    DVM: Towards Controllable LLM Agents in Social Deduction Games. In ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). 1–5. doi:10.1109/ICASSP49660.2025.10888525

  35. [2023]

    arXiv:2312.17515 [cs.CL] https://arxiv.org/abs/2312.17515

    Cooperation on the Fly: Exploring Language Agents for Ad Hoc Teamwork in the Avalon Game. arXiv:2312.17515 [cs.CL] https://arxiv.org/abs/2312.17515

  36. [2024]

    InThe Thirty-eighth Annual Conference on Neural Information Processing Systems

    Optimus-1: Hybrid Multimodal Memory Empowered Agents Excel in Long- Horizon Tasks. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems. https://openreview.net/forum?id=XXOMCwZ6by

  37. [2025]

    doi:10.1145/3721121

    The Collaborative Sensemaking Play of Jubensha Games: A Deconstruction, Taxonomy, and Analysis.ACM Games3, 1, Article 6 (March 2025), 34 pages. doi:10.1145/3721121

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.