When I’m analyzing a finished game using the on-site AI, it often makes move suggestions that I think are bonkers. What I’d naturally want to do is let the AI play its move, then play the move that I think refutes it, then let the AI respond to my move, and so on, for however many moves it takes until I see why I was wrong (which I always am). However, the way the interface currently works is that as soon as I put one stone down where the AI wants it, I’m shown the next dozen or so follow-up moves that the AI wants next. Seeing these moves makes it harder to think for myself about where I would play, which makes it harder to investigate why the AI’s thoughts are valid, since I’m not butting heads with it the way I would against a real player.
I would like to suggest a feature where you can start at any position in a finished game and play against the AI starting at that position, without having the AI “kibitz” about what the best moves are. So in my ideal world, whenever AI review is enabled, there would be a button labeled “play AI from position” that would hide the evaluation numbers, place a stone where the top AI suggestion was a moment ago, and then prompt me to place a stone in response, then the AI plays, then I play, etc, until I exit the mode.
When analysing your own games, the AI should also run on manual variations you put in. But it will also suggest moves for you.
You could fork the game, which starts a new game from the end state, but I don’t believe you can fork from a midpoint.
Indeed it does but I think Evan knows that. You need to endlessly toggle Ai on/off to get the sort of experience they are describing. ie:
Toggle on, see Ai move,
Toggle off, play Ai move and reply to it without seeing suggestions,
Rinse and repeat.
I think I see what they’re getting at but I’m not sure of the best implementation site-wise.
You can fork the game from wherever you want. Just go to the point of the game you want to fork from and click fork
I’ve used it to either test the scoring algorithm from a certain point, or to replay some endgame against katago micro
If katago-micro accepts forked challenges then that seems like an ideal solution for the OP.
Yes, interactive exploring of variations like this is a great way to learn. There are other programs designed for this, I use Lizzie on PC or Baduk AI on android.
That’s good to know, thanks for correcting me
Worth mentioning that the functionality described in the OP is how fineart reviewing works on fox and I like that implementation as well (though it’s monetized). You can input any position you want and then query for the best ai move. No ghost previews or anything like that.