Seong-lae ChoStockfish
- One of the strongest chess engines in the world (without deep learning)
- Used complex heuristics and prunings
- Utilize all known human chess strategies
AI Chess Models
Micro chess
– Each player has one of chess piece excepting for Queen
– The movement of knight is changed and all other pieces have same movements with the original chess (Due to neutralize Stockfish)
– Castling is available
– Pawn can move 2 step for the first time
– Tie rule exist
– Pawn promotion is possible
– Max turn: 30
– Winning condition: check-mate the opponent King
– If no check-mate until max turn, then calculate the score of remaining pieces
Tie rule: checkmate unavailable due to lack of pieces, No movement of any pawn or no capture of any piece for 50 turns, repeating same movement for 5 times
distillation
Chess Engines Do Weird Stuff
Since AlphaZero, lc0-style chess engines have been trained with RL. Specifically, you have the engine (search + model) play itself a bunch of times, and train the model to predict the outcome of the game.
https://girl.surgery/chess
LLM is bad at chess
Something weird is happening with LLMs and chess
are they good or bad?
https://dynomight.net/chess/
