Amortized Planning with Large-Scale Transformers: A Case Study on Chess

2024/02/07 by Anian Ruoss, Ruoss, Anian, Grégoire Delétang +17 · 10 voices · 11 citations
Computer Science · Economics, Econometrics and Finance · Psychology · #Artificial Intelligence in Games #Gambling Behavior and Treatments #Sports Analytics and Performance #cs.AI #cs.LG #stat.ML

paper · pdf · doi:10.48550/arxiv.2402.04494

openalex publication_date 2024/02/07 · openalex created_date 2024/02/09 · openalex updated_date 2026/07/28

Abstract

This dataset was prepared for the experiments in the Chess Reasoner project. It is derived from ChessBench, the large-scale chess benchmark introduced by Ruoss et al. in Amortized Planning with Large-Scale Transformers: A Case Study on Chess. The original ChessBench dataset contains chess positions from Lichess games annotated with legal moves and Stockfish 16 value estimates. This release extends the selected ChessBench-derived positions with additional Stockfish MultiPV line annotations. These annotations are used to train and evaluate transformer-based chess models that generate candidate continuations before selecting a final move. The archive contains the processed dataset used in our experiments. The corresponding code repository contains the preprocessing scripts, configuration files, random seeds, and a small sample dataset.

Cited by

• Three-Body Alignment: Aligning Chess Agent with Human Reasoning through Reranked Rationale
• Understanding Reasoning from Pretraining to Post-Training
• Chessformer: A Unified Architecture for Chess Modeling
• LLM CHESS: Benchmarking Reasoning and Instruction-Following in LLMs through Chess
• Genomic Next-Token Predictors are In-Context Learners
• Causal Masking on Spatial Data: An Information-Theoretic Case for Learning Spatial Datasets with Unimodal Language Models
• Exploring Human-AI Conceptual Alignment through the Prism of Chess
• Out-of-distribution Tests Reveal Compositionality in Chess Transformers
• Programmatic Representation Learning with Language Models
• ChessArena: A Chess Testbed for Evaluating Strategic Reasoning Capabilities of Large Language Models
• Transformer (deep learning) [wikipedia]

Discussions

• Grandmaster-Level Chess Without Search [hn, 198 points, 130 comments]
• A chess algorithm's search strategy is entirely replaced with a transformer capable of only looking one move ahead, and it still beats humans almost every time arxiv.org/abs/2402.04494 [bsky, 2 points, 0 comments]
• Early computer chess programs got their strength from searching far deeper than a human could. DeepMind now tells us they don't have to search at all. arxiv.org/abs/2402.0... [bsky, 1 points, 0 comments]
• Did you use the Chessbench dataset from arxiv.org/abs/2402.04494 (github.com/google-deepm...)? [bsky, 1 points, 1 comments]
• Grandmaster-Level Chess Without Search (arxiv.org) Main Link | Discussion [bsky, 0 points, 0 comments]
• "I see only one move ahead, but it is always the correct one." - Capablanca "Deep Mind has done it: their new engine doesn't calculate a single move yet it reached a 2895 Lichess blitz rating. Imagin [bsky, 0 points, 0 comments]
• Generalizing arxiv.org/abs/2402.044... further would be so cool... or maybe applying it to robotics... [bsky, 0 points, 0 comments]
• Grandmaster-Level Chess Without Search (arxiv.org) Main Link | Discussion [bsky, 0 points, 0 comments]
• A no-history and no-search approach to chess, using a transformer to predict values only for the next state that "plays" in simulation at a grandmaster level. arxiv.org/pdf/2402.04494 [bsky, 0 points, 0 comments]
• alphagoからmuzeroまで一貫してMCTSと深層学習で責めてたけどデータがたくさんあるなら巨大transformerで殴ったほうが強いっちゅうことかね… arxiv.org/abs/2402.04494 [bsky, 0 points, 0 comments]

Related

• Out-of-distribution Tests Reveal Compositionality in Chess Transformers
• Mastering Chess with a Transformer Model
• Large-scale Analysis of Chess Games with Chess Engines: A Preliminary Report
• Learning to Plan via Supervised Contrastive Learning and Strategic Interpolation: A Chess Case Study
• Predicting Chess Puzzle Difficulty with Transformers