A Chess Training App for Imitation Learning

A developer shared on LessWrong about a new chess training app built on imitation-learning principles. The app shows you board positions from real Lichess games and asks you to pick the stronger move between what the player actually played and what Stockfish recommends. After you guess, it reveals Stockfish's evaluation and plays out the position so you can understand the reasoning. Positions are filtered for moments where the choice was meaningful, and the app tracks your improvement with an Elo-style rating. It's open-source, MIT-licensed, and playable instantly with no sign-up. The developer's theory borrows from how language models learn: humans might get proficient quickly by imitating dense expert examples—though they predict eventually hitting a wall where studying full games becomes more efficient than continued position drilling.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton