Research map / Large language models / Agents, code and math
CausalGame: Benchmarking Causal Thinking of LLM Agents in Games
Zhenhao Chen, Yongqiang Chen, Chenxi Liu, Junchi Yu, Xiangchen Song, Zijian Li, Jialin Li, Philip Torr, Bo Han, Kun Zhang
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 |
|---|---|---|---|
| remove selection bias via balanced sampling (ablation) | LLM agents, Agent mode · CausalGame | survival rate (%): 64.9 -> 74.4 | Improved |
| remove selection bias via balanced sampling (ablation) | LLM agents, Prompt mode · CausalGame | survival rate (%): 64.6 -> 72.4 | Improved |
| increase reasoning effort from High to XHigh (ablation) | GPT-5.5, Agent mode · CausalGame | survival rate (%): 68.4 -> 65.1 | Hurt |
| enable thinking mode (ablation) | DeepSeek-V3.2 · CausalGame | survival rate (%): 61.7 -> 59.9 | Hurt |
| use agentic ReAct framework instead of single-turn prompting | top-performing LLMs · CausalGame | survival rate (%): up to +5.7% (DeepSeek-V4-Pro) | Mixed |
| use OpenCode coding agent instead of ReAct | 5 LLMs · CausalGame | survival rate (%): average +6.9% | Improved |
| give a locally optimal design as history | LLM agents · Antenna Trap | survival rate (%): not given | Improved |
| move from Antenna Trap to harder Deployment Zone Trap | LLM agents · CausalGame | survival rate (%): not given | Hurt |
| simple non-LLM baselines versus full agents | Uniform High / Random / Default · bias-heavy scenarios | survival rate (%): 49.0%-57.5% average | Mixed |
Closest papers by meaning
- ACED-Bench: Evaluating How LLM Agents Acquire and Act on Causal Evidence (NeurIPS 2026 · similarity 0.90)
- Unveiling Causal Reasoning in Large Language Models: Reality or Mirage? (NeurIPS 2024 · similarity 0.88)
- PhysGym: Benchmarking LLMs in Interactive Physics Discovery with Controlled Priors (NeurIPS 2025 · similarity 0.88)
- BALROG: Benchmarking Agentic LLM and VLM Reasoning On Games (ICLR 2025 · similarity 0.87)
- InnovatorBench: Evaluating Agents’ Ability to Conduct Innovative AI Research (ICLR 2026 · similarity 0.87)
- Language Agents Meet Causality -- Bridging LLMs and Causal World Models (ICLR 2025 · similarity 0.86)
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.
