Reproduce “ChessGPT: Bridging Policy Learning and Language Modeling” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Xidong Feng, Yicheng Luo, Ziyan Wang, Hongrui Tang, Mengyue Yang, Kun Shao, David Mguni, Yali Du and 1 more

ChessGPT trains chess language models on game records, chess commentary, and instruction data, treating moves and explanations as text. Its evaluation spans board-state tracking, position judgments, and move selection. ChessGPT-Base reaches 99.5 ± 0.2% on the Real Short legal-destination task, versus 23.2 ± 1.3% for RedPajama-Base; this measures benchmark correctness, not playing strength, and does not isolate which training data produced the result.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.