A browser-playable neural chess engine trained on a home PC
A roughly 16M-parameter chess network trained on one RTX 4090 with AI-assisted research, wrapped in a browser app for playing and analysis.
Credits
- Reddit poster Adam_Jesion Source ↗
Earliest dated source:
The full story
Autochess NN is a neural chess engine the maker built to understand AlphaZero-style systems end to end. The model combines a residual CNN with a transformer and learned thought tokens, at about 16 million parameters. It was trained on over 100 million positions on a single RTX 4090 using Lichess pretraining, Syzygy endgame fine-tuning and self-play with search distillation.
A browser app lets you play the AI, edit boards, replay PGNs, solve puzzles and see top-move probabilities. The maker calls it roughly 2700 Elo and says the efficiency claim is a hypothesis to be tested.
How it was done
Tools
Themes
Loading themes…
Sources
From a short post by the maker, so the details are as they tell them.
- Reddit: Vibecoded on a home PC: building a ~2700 Elo browser-playable neural chess engine with a Karpathy-inReddit poster: Adam_JesionPublished
The maker's post on r/ClaudeCode.
If it disappears: archived copy of Reddit: Vibecoded on a home PC: building a ~2700 Elo browser-playable neural chess engine with a Karpathy-in (opens in a new tab)