I know that sabaki detects shapes and automatically links the Senseis page.
Now i wonder if someone has tried to pattern match certain proverbs.
Some should be fairly easy like “hane on the head of two stones”.
“making a base” or “protect a cut” are maybe more challenging.
But combining a proverb detection with AI analysis could be a quite powerful learning tool. Look at the blue move and see if it matches any proverbs and then link the sensei pages to this exact proverb.
That’s what I do mentally, when reviewing my games.
Trouble is you can probably find one proverb that agrees with a move, and another that disagrees. If go was as simple as “play by proverb” the expert-system style go bots before Monte Carlo and neural network techniques would have been stronger.
It’s possible to create an LLM that only cares about recognizing situation that resemble a proverb, and doesn’t care about the correctness of the move and doesn’t care about the correctness of the proverb.
I’m currently working on something similar to this, it has a large set of proverb checkers that get all possible moves for each type of proverb and then uses the katago analysis to check winrates. It is early work in progress
Wow looks nice.
Is it possible to determine what is a weak group? Impressive.
I mostly see shaped detection, like a cap or a shoulder hit. But I fear you have to distinguish between the useful and the useless to point out. Besides shapes may vanish in the fury of the fights.
It may not be very hard to find a proverb that agrees with the AI move, but more difficult to explain why the AI move is better than the move you played.
But yeah I agree, often I find when trying to find out why the ai move is better, I have to test a few more ideas, a few sequences, a few tenukis and follow-ups etc