https://www.theoriachess.org/download/
We made a release with some minor bug fixes and improvements to Theoria. I used Claude Fable to analyze the code. The newer versions of Theoria use ply-to-ply stability to judge when to stop searching. Stability termination is the main code difference from Stockfish 16, but it also correctly evaluates odds games now as well.
We're developing a web app GUI to go with it, using natural language commentary from a language model, but it isn't yet public and is still being playtested.
theoria 0.3.1
Moderator: Ras
-
thomasahle
- Posts: 95
- Joined: Thu Feb 27, 2014 8:19 pm
Re: theoria 0.3.1
Who are "we" in this context? You and Claude?
> Instead of learning from iteration on Stockfish's evaluation, Theoria's network learns positional understanding from Lc0's self-play data.
This is also how Stockfish does it, in case you didn't know.
> Instead of learning from iteration on Stockfish's evaluation, Theoria's network learns positional understanding from Lc0's self-play data.
This is also how Stockfish does it, in case you didn't know.
-
Ciekce
- Posts: 222
- Joined: Sun Oct 30, 2022 5:26 pm
- Full name: Conor Anstey
Re: theoria 0.3.1
he's been told many times, he just doesn't care about such petty things as realitythomasahle wrote: ↑Tue Aug 11, 2026 11:01 pm This is also how Stockfish does it, in case you didn't know.