A chess AI that plays
like a human
Chess engines became superhuman decades ago. Noctie goes the other way: it's trained on over a billion human games to play like a person at your level.
Seventy years of chess AI, briefly
Hard-coded rules and simple checklists. They could play legal chess, not much more.
Brute force wins: special hardware allowing looking at millions of positions per second to defeat the world champion.
Neural networks learn chess by playing themselves. Intuitive, less mechanical play that taught humans several new tricks.
Neural nets trained on human play and algorithms emulating the human thinking process. The goal isn't the best move - it's the move a human at your level would play.
Why perfect play makes a bad sparring partner
Turn a normal engine down to "beginner" and it doesn't play like a beginner. It mixes flawless moves with random blunders no human would make. You end up practicing against something that doesn't exist.
A regular engine
- Strength from calculating millions of positions per second
- Always hunts for the objectively best move
- Weakened by sprinkling in random blunders
- Snaps out tactical moves instantly
Noctie
- Strength from intuition, learned from a billion human games
- Plays the moves people at your rating actually choose
- Makes the mistakes a human at your level would make
- Thinks in a human rhythm, sometimes fast sometimes slow
It knows how a 1200 human player thinks
Set Noctie to 1200 and it falls for the tricks a 1200 falls for, misses the tactics a 1200 misses, and struggles with the same endgames.
Beginner to GM
Neural networks and thinking processes covering the whole human range. At GM strength it plays with the judgment of a strong human, not the cold precision of an engine.
Plays openings you'll actually meet online
Pick an opening to practice and Noctie responds with the lines real people play online - traps included.
Built for training
Every game comes with instant feedback, hints when you want them, and puzzles made from your own mistakes.
