♟️Chessbook-1: Human-Like Chess Bot After 10 Months
Chessbook-1 predicts human moves better than Stockfish
TL;DR
Chessbook-1, a new chess bot trained over 10 months, outperforms older models in predicting human moves and think time. It's particularly strong for ratings 1400-1800, with a 54.3% top-1 accuracy and 25.2% average probability assigned to the human's actual move.
Chessbook-1, a chess bot trained over 10 months, outperforms previous models in predicting human moves and think time. For ratings 1400-1800, it has a 54.3% top-1 accuracy and assigns an average probability of 25.2% to the human's actual move. This makes it particularly useful for understanding human chess strategies and improving player performance. Chessbook-1 also outperforms Stockfish and other models when limited to the player's rating, making it a valuable tool for chess analysis and training.
Key Points
Chessbook-1 trained on games through July 2026, 10 months in total.
Top-1 accuracy predicting human moves: 54.3% for ratings 1400-1800.
Average probability assigned to human's actual move: 25.2% for ratings 1400-1800.
Outperforms Maia-3 and Allie 2.0 in predicting human moves and think time.
Outperforms Stockfish when limited to the player's rating for ratings 600-1800.
Why It Matters
Chessbook-1's ability to predict human moves and think time with high accuracy makes it a valuable tool for chess players and analysts. For ratings 1400-1800, it has a 54.3% top-1 accuracy and assigns an average probability of 25.2% to the human's actual move. This could help players improve their strategies and understand their opponents better. Chess coaches and analysts should consider integrating Chessbook-1 into their training and analysis workflows.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,569 builders reading daily.