Brendan Long's chess app trains like an AI model

Brendan Long, inspired by imitation learning techniques used in large language models (LLMs), has created a chess training app that forces users to mimic Stockfish's move choices. The app presents real game positions with two move options—typically the human player's actual move versus Stockfish's optimal move—and asks users to select the better option. After each choice, users receive instant feedback: Stockfish's evaluation, the next few moves Stockfish would play, and an Elo-style score that tracks progress (calibrated to 80% difficulty for optimal learning).

The app is designed to accelerate foundational chess skills by leveraging the same 'dense pretraining' approach used in LLMs, where models learn by imitating teacher-generated examples before advancing to real-world tasks. The backend sources positions from Lichess games, filters for critical moments, and uses Stockfish to generate alternative moves and win probabilities. All code is open-source (MIT licensed), and the tool is accessible instantly without signup. While not intended as a replacement for full-game practice, it offers a novel way to internalize engine-like patterns and decision-making heuristics.