pith. sign in

arxiv: 2605.24375 · v1 · pith:DG7R2NX6new · submitted 2026-05-23 · 💻 cs.AI

Distilling Game Code World Model Generation into Lightweight Large Language Models

Tyrone Serapio , Arjun Prakash , Haoyang Xu , Kevin Wang , Amy Greenwald This is my paper

Pith reviewed 2026-06-30 13:54 UTC · model grok-4.3

classification 💻 cs.AI
keywords Game Code World Modelscode generationsupervised fine-tuningreinforcement learning with verifiable rewardslarge language modelsautomatic environment generationgame rules
0
0 comments X

The pith

A post-training pipeline of supervised fine-tuning and reinforcement learning with verifiable rewards equips a 3B-parameter model to generate valid GameCWMs from natural language.

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

The paper shows that the ability to translate game rules into executable Python code implementing states, actions, transitions, observations, and rewards can be transferred to smaller LLMs. It creates a dataset covering 30 perfect- and imperfect-information games, builds a verifier that checks structural and semantic properties, and applies SFT to improve syntax followed by RLVR to improve rule adherence. If the approach holds, it would let modest-sized models replace frontier models and repeated inference loops for creating AI environments. Readers would care because it lowers the barrier to automatically building testable game simulators from descriptions alone.

Core claim

Through supervised fine-tuning on a curated set of 30 games followed by reinforcement learning with rewards from a verification framework that checks code against game properties, Qwen2.5-3B-Instruct produces more syntactically correct and semantically valid GameCWMs that correctly handle rules, legal actions, state transitions, observations, and rewards in both perfect and imperfect information settings.

What carries the argument

The verification framework that scores generated code on structural and semantic adherence to game rules, supplying the verifiable rewards for the RLVR stage after initial SFT.

If this is right

  • SFT alone raises the rate of syntactically correct GameCWM code.
  • RLVR adds measurable gains in execution-level rule adherence.
  • The gains appear in both perfect-information and imperfect-information games.
  • The pipeline reduces dependence on frontier models and inference-time refinement for environment creation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same SFT-plus-RLVR pattern could be tested on generating simulators for robotics or physics environments beyond games.
  • If the verifier generalizes, the method might distill other constrained code-generation tasks into small models.
  • Performance on games outside the original 30 would test whether the distilled capability transfers to novel rule sets.

Load-bearing premise

The verification framework catches all important rule violations and reward errors in the generated code.

What would settle it

Generated code that passes the verifier yet produces wrong game outcomes or permits illegal moves when executed inside an actual solver such as Monte Carlo Tree Search.

Figures

Figures reproduced from arXiv: 2605.24375 by Amy Greenwald, Arjun Prakash, Haoyang Xu, Kevin Wang, Tyrone Serapio.

Figure 1
Figure 1. Figure 1: Our Distillation Pipeline. Unlike prior approaches [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
read the original abstract

Large Language Models (LLMs) have shown great ability in generating executable code from natural language, opening the possibility of automatically constructing environments for AI agents. Recent work on Code World Models (CWMs) demonstrates that LLMs can translate game rules into Python implementations compatible with solvers like Monte Carlo Tree Search. We study this problem in game settings, where generated environments must implement rules, legal actions, state transitions, observations, and rewards. We refer to these game-specific executable models as Game Code World Models (GameCWMs). However, current approaches to generating code world models rely on frontier models and inference-time refinement loops, limiting accessibility and scalability. This work investigates whether GameCWM generation capabilities can be distilled into smaller models through post-training. We introduce: (1) a curated dataset of 30 games spanning perfect and imperfect information games, (2) a verification framework that evaluates generated code against structural and semantic game properties, and (3) a post-training pipeline combining Supervised Fine-Tuning (SFT) with Reinforcement Learning with Verifiable Rewards (RLVR). We experiment with Qwen2.5-3B-Instruct and find that SFT can increase syntactic correctness, while RLVR can improve execution-level adherence to game rules, thereby improving Qwen's ability to generate valid GameCWMs in both perfect and imperfect information games. Overall, our pipeline makes Qwen2.5-3B-Instruct more capable of generating valid GameCWMs, thereby offering a scalable path toward automatic environment generation from natural language.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper claims that post-training Qwen2.5-3B-Instruct via supervised fine-tuning (SFT) followed by reinforcement learning with verifiable rewards (RLVR) distills the ability to generate valid Game Code World Models (GameCWMs) from a curated dataset of 30 games (spanning perfect- and imperfect-information settings). The resulting model shows improved syntactic correctness after SFT and better execution-level adherence to game rules (state transitions, legal actions, observations, rewards) after RLVR, offering a scalable alternative to frontier-model inference-time refinement for automatic environment generation from natural language.

Significance. If the quantitative gains hold under rigorous verification, the work shows that a 3B-parameter model can acquire a non-trivial code-generation capability previously associated with much larger models, while the released dataset and verification framework would constitute reusable resources for the community. The explicit separation of syntactic and semantic/execution metrics is a methodological strength.

major comments (2)
  1. [Abstract and §3] Abstract and §3 (verification framework): the central RLVR stage assigns rewards based on 'structural and semantic game properties,' yet the manuscript supplies no concrete specification of the test harness, the exact semantic checks performed (e.g., hidden-state consistency, observation-function correctness, or reward calculation under imperfect information), or coverage statistics across the 30-game suite. Because any undetected rule violations would be reinforced by RLVR, this omission directly undermines confidence in the reported execution-level improvements.
  2. [§5] §5 (experiments): the claim that 'RLVR can improve execution-level adherence' is load-bearing for the distillation thesis, but the provided text contains no quantitative results, error bars, ablation tables, or per-game breakdowns that would allow assessment of robustness or of whether gains are driven by a subset of perfect-information games.
minor comments (2)
  1. [Abstract] The abstract states the pipeline works 'in both perfect and imperfect information games' but does not indicate the split or whether results are reported separately; adding this breakdown would improve clarity.
  2. [§2] Notation for GameCWM components (state transition, observation, reward functions) should be introduced once with consistent symbols rather than re-described in prose.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major comment below and will revise the manuscript to incorporate the requested details.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (verification framework): the central RLVR stage assigns rewards based on 'structural and semantic game properties,' yet the manuscript supplies no concrete specification of the test harness, the exact semantic checks performed (e.g., hidden-state consistency, observation-function correctness, or reward calculation under imperfect information), or coverage statistics across the 30-game suite. Because any undetected rule violations would be reinforced by RLVR, this omission directly undermines confidence in the reported execution-level improvements.

    Authors: We agree that additional concrete details on the verification framework are necessary. In the revised manuscript we will expand §3 (and the abstract where appropriate) to specify the test harness, the precise semantic checks performed (including hidden-state consistency, observation-function correctness, and reward calculation under imperfect information), and coverage statistics across the 30-game suite. This will make the reward assignment process fully transparent. revision: yes

  2. Referee: [§5] §5 (experiments): the claim that 'RLVR can improve execution-level adherence' is load-bearing for the distillation thesis, but the provided text contains no quantitative results, error bars, ablation tables, or per-game breakdowns that would allow assessment of robustness or of whether gains are driven by a subset of perfect-information games.

    Authors: We acknowledge that the current §5 lacks the quantitative evidence needed to substantiate the claims. The revised version will add quantitative results (with error bars), ablation tables, and per-game breakdowns to demonstrate robustness and to show that improvements hold across both perfect- and imperfect-information games. revision: yes

Circularity Check

0 steps flagged

No circularity: pipeline uses external verifier on curated dataset

full rationale

The paper presents an empirical pipeline (SFT followed by RLVR) that trains Qwen2.5-3B-Instruct on a new 30-game dataset and measures improvement via a newly introduced verification framework checking structural and semantic properties. No step reduces by construction to a fitted parameter, self-defined quantity, or load-bearing self-citation; the central claim rests on external execution-level checks rather than internal redefinition or renaming of prior results. The derivation chain is therefore self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that the hand-crafted verification framework is a reliable proxy for semantic correctness; no free parameters are explicitly named in the abstract, but typical RLVR training involves learning-rate and reward-scaling choices that are not reported.

axioms (1)
  • domain assumption The verification framework accurately measures adherence to game rules, legal actions, state transitions, observations, and rewards.
    Invoked when RLVR is said to improve execution-level adherence; if false, the training signal is invalid.

pith-pipeline@v0.9.1-grok · 5815 in / 1333 out tokens · 35095 ms · 2026-06-30T13:54:56.531787+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

40 extracted references · 33 canonical work pages · 9 internal anchors

  1. [1]

    Anton Bakhtin, Noam Brown, Emily Dinan, Gabriele Farina, Colin Flaherty, Daniel Fried, Andrew Goff, Jonathan Gray, Hengyuan Hu, Athul Jacob, Mojtaba Komeili, Karthik Konath, Minae Kwon, Adam Lerer, Mike Lewis, Alexander Miller, Sasha Mitts, Adithya Renduchintala, Stephen Roller, and Markus Zijlstra. 2022. Human-level play in the game of Diplomacy by combi...

    work page doi:10.1126/science.ade9097 2022

  2. [2]

    Bahar Bateni, Benjamin Pratt, and Jim Whitehead. 2025. LLM game rule under- standing through out-of-distribution fine-tuning. InProceedings of the Twenty- First AAAI Conference on Artificial Intelligence and Interactive Digital Entertain- ment(Edmonton, Alberta, Canada)(AIIDE ’25). AAAI Press, Article 1, 10 pages. https://doi.org/10.1609/aiide.v21i1.36804

    work page doi:10.1609/aiide.v21i1.36804 2025

  3. [3]

    Shiqi Chen, Tongyao Zhu, Zian Wang, Jinghan Zhang, Kangrui Wang, Siyang Gao, Teng Xiao, Yee Whye Teh, Junxian He, and Manling Li. 2025. Internalizing World Models via Self-Play Finetuning for Agentic RL. arXiv:2510.15047 [cs.LG] https://arxiv.org/abs/2510.15047

    work page arXiv 2025

  4. [4]

    Rémi Coulom. 2006. Efficient Selectivity and Backup Operators in Monte-Carlo Tree Search. InInternational Conference on Computers and Games. Springer, 72– 83

    2006

  5. [5]

    Peter I Cowling, Edward J Powley, and Daniel Whitehouse. 2012. Information Set Monte Carlo Tree Search.IEEE Transactions on Computational Intelligence and AI in Games4, 2 (2012), 120–143

    2012

  6. [6]

    Nicola Dainese, Matteo Merler, Minttu Alakuijala, and Pekka Marttinen. 2024. Generating Code World Models with Large Language Models Guided by Monte Carlo Tree Search. arXiv:2405.15383 [cs.AI] https://arxiv.org/abs/2405.15383

    work page arXiv 2024

  7. [7]

    Shilong Deng, Yongzhao Wang, and Rahul Savani. 2025. From Natural Language to Extensive-Form Game Representations. arXiv:2501.17282 [cs.AI] https: //arxiv.org/abs/2501.17282

    work page arXiv 2025

  8. [8]

    Shihan Dou, Yan Liu, Haoxiang Jia, Limao Xiong, Enyu Zhou, Wei Shen, Junjie Shan, Caishuang Huang, Xiao Wang, Xiaoran Fan, Zhiheng Xi, Yuhao Zhou, Tao Ji, Rui Zheng, Qi Zhang, Xuanjing Huang, and Tao Gui. 2024. StepCoder: Improve Code Generation with Reinforcement Learning from Compiler Feedback. arXiv:2402.01391 [cs.SE] https://arxiv.org/abs/2402.01391

    work page arXiv 2024

  9. [9]

    Xidong Feng, Yicheng Luo, Ziyan Wang, Hongrui Tang, Mengyue Yang, Kun Shao, David Mguni, Yali Du, and Jun Wang. 2023. ChessGPT: Bridging Policy Learning and Language Modeling. arXiv:2306.09200 [cs.LG] https://arxiv.org/ab s/2306.09200

    work page arXiv 2023

  10. [10]

    Ian Gemp, Roma Patel, Yoram Bachrach, Marc Lanctot, Vibhavari Dasagi, Luke Marris, Georgios Piliouras, Siqi Liu, and Karl Tuyls. 2024. Steering Language Models with Game-Theoretic Solvers.arXiv preprint arXiv:2402.01704(2024)

    work page arXiv 2024

  11. [11]

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Peiyi Wang, Qihao Zhu, Runxin Xu, Ruoyu Zhang, Shirong Ma, Xiao Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, Aixin Liu, Bing Xue, Bingxuan Wang, Bochao Wu, Bei Feng, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chong Ruan, Damai Dai, Deli Chen, Dongjie Ji, ...

    work page doi:10.1038/s41586-025-09422-z 2025

  12. [12]

    LoRA: Low-Rank Adaptation of Large Language Models

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. LoRA: Low-Rank Adaptation of Large Language Models. arXiv:2106.09685 [cs.CL] https://arxiv.org/abs/2106.09685

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  13. [13]

    Haitao Jiang, Wenbo Zhang, Jiarui Yao, Hengrui Cai, Sheng Wang, and Rui Song. 2026. Supervised Fine-Tuning versus Reinforcement Learning: A Study of Post-Training Methods for Large Language Models. arXiv:2603.13985 [cs.AI] https://arxiv.org/abs/2603.13985

    work page arXiv 2026

  14. [14]

    Xue Jiang, Yihong Dong, Mengyang Liu, Hongyi Deng, Tian Wang, Yongding Tao, Rongyu Cao, Binhua Li, Zhi Jin, Wenpin Jiao, Fei Huang, Yongbin Li, and Ge Li

  15. [15]

    CodeRL+: Improving Code Generation via Reinforcement with Execution Semantics Alignment

    CodeRL+: Improving Code Generation via Reinforcement with Execution Semantics Alignment. arXiv:2510.18471 [cs.SE] https://arxiv.org/abs/2510.18471

    work page internal anchor Pith review Pith/arXiv arXiv

  16. [16]

    Tulu 3: Pushing Frontiers in Open Language Model Post-Training

    Nathan Lambert, Jacob Morrison, Valentina Pyatkin, Shengyi Huang, Hamish Ivison, Faeze Brahman, Lester James V. Miranda, Alisa Liu, Nouha Dziri, Shane Lyu, Yuling Gu, Saumya Malik, Victoria Graf, Jena D. Hwang, Jiangjiang Yang, Ronan Le Bras, Oyvind Tafjord, Chris Wilhelm, Luca Soldaini, Noah A. Smith, Yizhong Wang, Pradeep Dasigi, and Hannaneh Hajishirzi...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  17. [17]

    Marc Lanctot, Edward Lockhart, Jean-Baptiste Lespiau, Vinicius Zambaldi, Satyaki Upadhyay, Julien Pérolat, Sriram Srinivasan, Finbarr Timbers, Karl Tuyls, Shayegan Omidshafiei, Daniel Hennes, Dustin Morrill, Paul Muller, Timo Ewalds, Ryan Faulkner, János Kramár, Bart De Vylder, Brennan Saeta, James Bradbury, David Ding, Sebastian Borgeaud, Matthew Lai, Ju...

    work page arXiv 2019

  18. [18]

    Hung Le, Yue Wang, Akhilesh Deepak Gotmare, Silvio Savarese, and Steven C. H. Hoi. 2022. CodeRL: Mastering Code Generation through Pretrained Models and Deep Reinforcement Learning. arXiv:2207.01780 [cs.LG] https://arxiv.org/abs/ 2207.01780

    work page arXiv 2022

  19. [19]

    Wolfgang Lehrach, Daniel Hennes, Miguel Lazaro-Gredilla, Xinghua Lou, Carter Wendelken, Zun Li, Antoine Dedieu, Jordi Grau-Moya, Marc Lanctot, Atil Is- cen, John Schultz, Marcus Chiam, Ian Gemp, Piotr Zielinski, Satinder Singh, and Kevin P. Murphy. 2025. Code World Models for General Game Playing. arXiv:2510.04542 [cs.AI] https://arxiv.org/abs/2510.04542

    work page arXiv 2025

  20. [20]

    Generative Agents: Interactive Simulacra of Human Behavior

    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. arXiv:2304.03442 [cs.HC] https://arxiv.org/abs/2304.03442

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  21. [21]

    Annu Rana and Gaurav Kumar. 2025. Model-First Reasoning LLM Agents: Reduc- ing Hallucinations through Explicit Problem Modeling. arXiv:2512.14474 [cs.AI] https://arxiv.org/abs/2512.14474

    work page arXiv 2025

  22. [22]

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

  23. [23]

    Proximal Policy Optimization Algorithms

    Proximal Policy Optimization Algorithms. arXiv:1707.06347 [cs.LG] https://arxiv.org/abs/1707.06347

    work page internal anchor Pith review Pith/arXiv arXiv

  24. [24]

    John Schultz, Jakub Adamek, Matej Jusup, Marc Lanctot, Michael Kaisers, Sarah Perrin, Daniel Hennes, Jeremy Shar, Cannada Lewis, Anian Ruoss, Tom Zahavy, Petar Veličković, Laurel Prince, Satinder Singh, Eric Malmi, and Nenad Tomašev

  25. [25]

    arXiv:2412.12119 [cs.AI] https://arxiv.org/abs/2412.12119

    Mastering Board Games by External and Internal Planning with Language Models. arXiv:2412.12119 [cs.AI] https://arxiv.org/abs/2412.12119

    work page arXiv

  26. [26]

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, et al. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300(2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  27. [27]

    2008.Multiagent Systems: Algorithmic, Game-Theoretic, and Logical Foundations

    Yoav Shoham and Kevin Leyton-Brown. 2008.Multiagent Systems: Algorithmic, Game-Theoretic, and Logical Foundations. Cambridge University Press, USA

    2008

  28. [28]

    Biegler, and Aaron M

    Ardalan Tajbakhsh, Lorenz T. Biegler, and Aaron M. Johnson. 2024. Conflict- Based Model Predictive Control for Scalable Multi-Robot Motion Planning. arXiv:2303.01619 [cs.RO] https://arxiv.org/abs/2303.01619

    work page arXiv 2024

  29. [29]

    Hao Tang, Keya Hu, Jin Peng Zhou, Sicheng Zhong, Wei-Long Zheng, Xujie Si, and Kevin Ellis. 2024. Code Repair with LLMs gives an Exploration-Exploitation Tradeoff. arXiv:2405.17503 [cs.SE] https://arxiv.org/abs/2405.17503

    work page arXiv 2024

  30. [30]

    Qwen Team. 2024. Qwen2.5: A Party of Foundation Models. https://qwenlm.git hub.io/blog/qwen2.5/

    2024

  31. [31]

    Emanuel Todorov, Tom Erez, and Yuval Tassa. 2012. MuJoCo: A physics engine for model-based control. In2012 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 5026–5033. https://doi.org/10.1109/IROS.2012.6386109

    work page doi:10.1109/iros.2012.6386109 2012

  32. [32]

    Mark Towers, Ariel Kwiatkowski, Jordan Terry, John U Balis, Gianluca De Cola, Tristan Deleu, Manuel Goulão, Andreas Kallinteris, Markus Krimmel, Arjun KG, et al. 2024. Gymnasium: A Standard Interface for Reinforcement Learning Environments.arXiv preprint arXiv:2407.17032(2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  33. [33]

    Vivek Verma, David Huang, William Chen, Dan Klein, and Nicholas Tomlin. 2025. Measuring General Intelligence with Generated Games. arXiv:2505.07215 [cs.AI] https://arxiv.org/abs/2505.07215

    work page arXiv 2025

  34. [34]

    Di Wang, Zhenhua Wu, Yu Liu, Kai Chang, and Shaohua Wu. 2026. A Generalizable Framework for Building Executable Domain-Specific LLMs under Data Scarcity: Demonstration on Semiconductor TCAD Simulation. arXiv:2601.10128 [cs.CE] https://arxiv.org/abs/2601.10128

    work page arXiv 2026

  35. [35]

    Yinjie Wang, Ling Yang, Ye Tian, Ke Shen, and Mengdi Wang. 2025. Co-Evolving LLM Coder and Unit Tester via Reinforcement Learning. arXiv:2506.03136 [cs.CL] https://arxiv.org/abs/2506.03136

    work page arXiv 2025

  36. [36]

    Lai Wei, Yuting Li, Kaipeng Zheng, Chen Wang, Yue Wang, Linghe Kong, Lichao Sun, and Weiran Huang. 2025. Advancing Multimodal Reasoning via Reinforcement Learning with Cold Start. arXiv:2505.22334 [cs.CL] https: //arxiv.org/abs/2505.22334

    work page arXiv 2025

  37. [37]

    Yuzhuang Xu, Shuo Wang, Peng Li, Fuwen Luo, Xiaolong Wang, Weidong Liu, and Yang Liu. 2024. Exploring Large Language Models for Communication Games: An Empirical Study on Werewolf. arXiv:2309.04658 [cs.CL] https: //arxiv.org/abs/2309.04658

    work page arXiv 2024

  38. [38]

    Yang Yue, Zhiqi Chen, Rui Lu, Andrew Zhao, Zhaokai Wang, Yang Yue, Shiji Song, and Gao Huang. 2025. Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?arXiv preprint arXiv:2504.13837(2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  39. [39]

    Chunting Zhou, Pengfei Liu, Puxin Xu, Srini Iyer, Jiao Sun, Yuning Mao, Xuezhe Ma, Avia Efrat, Ping Yu, Lili Yu, Susan Zhang, Gargi Ghosh, Mike Lewis, Luke Zettlemoyer, and Omer Levy. 2023. LIMA: Less Is More for Alignment. arXiv:2305.11206 [cs.CL] https://arxiv.org/abs/2305.11206

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  40. [40]

    arXiv preprint arXiv:2310.11667 , year=

    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. SOTOPIA: Interactive Evaluation for Social Intelligence in Language Agents. arXiv:2310.11667 [cs.AI] https://arxiv.org/abs/2310.11667 Appendix A PSEUDOCODE A.1 Perfect Information Game Generat...

    work page arXiv 2024