Research map / Large language models / Agents, code and math
Code World Models for General Game Playing
Wolfgang Lehrach, Daniel Hennes, Miguel Lazaro-Gredilla, Xinghua Lou, Carter Wendelken, Zun Li, Antoine Dedieu, Marc Lanctot, Atil Iscen, John Schultz, Marcus Chiam, Ian Gemp and 3 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use tree-search refinement of CWM instead of conversation refinement (ablation) | LLM-synthesized CWM · perfect information games | CWM accuracy | No clear effect |
| synthesize a value function for MCTS (ablation) | CWM + ISMCTS · Bargaining (CWM agent as player 1) | payoff | Improved |
| apply synthesized value functions (ablation) | CWM + ISMCTS · other imperfect information games | playing performance | No clear effect |
| use hidden history inference instead of hidden state inference (ablation) | CWM + ISMCTS · imperfect information games | inference accuracy | Improved |
| go from open deck to closed deck CWM synthesis (ablation) | CWM agent · Bargaining | test inference accuracy: 1.0000 -> 0.67359 | Hurt |
| go from open deck to closed deck CWM synthesis (ablation) | CWM agent · Gin rummy | test inference accuracy: 0.5376 -> 0.09523 | Hurt |
| go from open deck to closed deck CWM synthesis (ablation) | CWM agent · game play vs Gemini 2.5 Pro | win/payoff | Mixed |
| use CWM + MCTS planning instead of LLM as direct policy | Gemini 2.5 Pro · Backgammon | win rate: 1.00 (100/100 Gemini forfeits) | Improved |
| use CWM + MCTS planning instead of LLM as direct policy | Gemini 2.5 Pro · Tic-tac-toe | win/draw rate: 0.05 win, 0.95 draw | No clear effect |
| use CWM + ISMCTS instead of Gemini policy | Gemini 2.5 Pro · Hand of war | game outcome | Hurt |
| learn Gin rummy rules from trajectories | CWM synthesis · Gin rummy | test transition accuracy: 0.7455 | Hurt |
Closest papers by meaning
- Mastering Board Games by External and Internal Planning with Language Models (ICML 2025 · similarity 0.93)
- GTBench: Uncovering the Strategic Reasoning Capabilities of LLMs via Game-Theoretic Evaluations (NeurIPS 2024 · similarity 0.92)
- Can Large Language Models Master Complex Card Games? (NeurIPS 2025 · similarity 0.91)
- GameTraversalBenchmark: Evaluating Planning Abilities Of Large Language Models Through Traversing 2D Game Maps (NeurIPS 2024 · similarity 0.90)
- Large Language Models Are Neurosymbolic Reasoners (AAAI 2024 · similarity 0.89)
- Evaluating LLMs in Open-Source Games (NeurIPS 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
