Research map / Large language models / Agents, code and math
Evaluation beyond Task Performance: Analyzing Concepts in AlphaZero in Hex
Charles Lovering, Jessica Zosa Forde, George Konidaris, Ellie Pavlick, Michael Littman
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 MCTS to select moves before the policy network has learned the concepts | AlphaZero in Hex · Hex positive-concept behavioral tests | behavioral test pass rate | Improved |
| Train the policy network to match MCTS logits | AlphaZero in Hex · Hex concept representation | concept probing accuracy | Improved |
| Use upper network layers for short-term concepts rather than long-term concepts | AlphaZero in Hex · Hex short-term concepts | probe accuracy | Improved |
| Use middle network layers for long-term concepts rather than short-term concepts | AlphaZero in Hex · Hex long-term concepts | probe accuracy | Improved |
| Train AlphaZero to recognize and use positive concepts | AlphaZero in Hex · Hex positive-concept behavioral tests | pass rate | Improved |
| Train AlphaZero to recognize negative concepts such as dead and captured cells | AlphaZero in Hex · Hex negative-concept behavioral tests | pass rate | Mixed |
| Allow AlphaZero to choose moves without a preference for faster wins | AlphaZero in Hex · Hex end-game play | moves that do not impact the outcome: 25% | Hurt |
| Connect the defender in behavioral-test boards to force the conceptually correct move | AlphaZero in Hex · Hex positive-concept behavioral tests | pass rate | Improved |
Closest papers by meaning
- Mastering Board Games by External and Internal Planning with Language Models (ICML 2025 · similarity 0.89)
- AlphaZero Neural Scaling and Zipf's Law: a Tale of Board Games and Power Laws (NeurIPS 2025 · similarity 0.88)
- Implicit Search via Discrete Diffusion: A Study on Chess (ICLR 2025 · similarity 0.87)
- Prompted Policy Search: Reinforcement Learning through Linguistic and Numerical Reasoning in LLMs (NeurIPS 2025 · similarity 0.87)
- On the Modeling Capabilities of Large Language Models for Sequential Decision Making (ICLR 2025 · similarity 0.86)
- BALROG: Benchmarking Agentic LLM and VLM Reasoning On Games (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.
