For my stacker chess playing brethren.
I have (with the extensive help of Codex) developed an Android app that bundles the Maia 3 chess engine.
Maia-3 is the latest Maia model from the University of Toronto’s CSSLab, published this year as part of the Chessformer paper Its largest model matches the move played by a human about 57% of the time.
You can set Maia to your elo range and spar against it. You can also just use the app as a free game review with the added bonus of seeing the most likely human move (for the set elo) along side the best two computer moves. It's great for opening prep--to find lines that a computer could defend but that a human would likely blunder in.
Credit to the FOSS projects Lichess, from which the design is taken, En Croissant, from which the game review comes from, and of course the Maia team at the University of Toronto. This is not in any way affiliated with them, but simply building on their work.
- Source code and documentation
- [Latest Android release](https://github.com/Dash1971/maia-chess-android/releases/latest)
- [Install and receive updates with Obtainium](https://github.com/Dash1971/maia-chess-android#install-and-update-with-obtainium)
This is Android only and not available through Google Play. You need to do Obtanium or side load.
Screenshots below.
@denlillaapan 👀
saw it, great work; don't understand the point
Playing around over at https://www.maiachess.com/ official website of the Maia project, may help to grok it. Even if you don't like playing AI, run one of your game reviews through their analysis board and note where the Maia engine diverts from Stockfish. Pay special attention when that is a blunder because it means humans at whatever level you have set Maia at are most likely to blunder in that position.
My app just takes their engine and turns it into an offline mobile app. Once you become a fan of Maia you want it wherever you are so you can pull up an analysis board when you have an idea and compare the human likely responses with the engine lines.
so teaching me to play an objectively worse move because, at my rating level, it's statistically likely to work? Hm, interesting. Stockfish/chess.com review suggestions are sometimes useless precisely for this reason... unfindable moves at the 1700 level
Editing this as I misunderstood your comment. Yes I think you get it. I mean, it sounds a bit hacky when you put it like that but that is one way to use it. Another way is just to become more mindful of common mistakes so you can better root them put of your own play or anticipate and exploit them in others'
Also useful is to set Maia to a higher level than you play at so you can see what stronger human players play in a given position. As you say, often the computer line does not make practical sense and the stronger human move may give you a more playable position. (The app allows you to configure the play Maia to a different level to the analysis Maia)