lichess.org
Donate

AI behavior gives away part of the puzzle solution

In puzzle 0G6WJ (lichess.org/training/0G6WJ), the AI response to my third move is to underpromote to rook. It's pretty clear that this type of underpromotion is due to Stockfish's equivalent evaluation between all possible pieces. And that sort of thing usually happens when the next move of the puzzle (or game) is for the promoted piece to be captured. Hence, my fourth move is obvious, not from the puzzle itself, but from Stockfish's behavior. If the promotion were to a queen, it would be a better puzzle. Is there any way to steer Stockfish's choice among equivalent outcomes?
@mcgoves said in #1:
> In puzzle 0G6WJ (lichess.org/training/0G6WJ), the AI response to my third move is to underpromote to rook. It's pretty clear that this type of underpromotion is due to Stockfish's equivalent evaluation between all possible pieces. And that sort of thing usually happens when the next move of the puzzle (or game) is for the promoted piece to be captured.

This is all correct.

> Hence, my fourth move is obvious, not from the puzzle itself, but from Stockfish's behavior. If the promotion were to a queen, it would be a better puzzle. Is there any way to steer Stockfish's choice among equivalent outcomes?

This is what I don't understand. White can promote to any piece and black's move doesn't change. So why promoting to a queen would be better?
@Deadban said in #2:
> This is what I don't understand. White can promote to any piece and black's move doesn't change. So why promoting to a queen would be better?

If it promotes to rook, it is obvious that you should capture it rather than considering a different move. If it were queen, ambiguity could in theory remain.
You make a good point, that this is almost always true.

But it's not important that you're able to guess the answer to puzzles. Puzzle ratings don't mean anything. The problem is that you're making a move, seeing a response, and then figuring out what to do. They call that Hope Chess. Make a move and hope you're able to handle the response. It's literally practicing a bad habit.

If instead you'd calculate the entire solution before making a move then you'd get better at chess, rather than merely raising your puzzle rating.
Yes, that's the proper approach. Sometimes, however, it doesn't always work. Some puzzles extend a few moves beyond what I've analyzed. I mis-analyze some puzzles; get a move right, but get a surprising response. And occasionally theres a puzle whose first move is obvious no matter the response. Sometimes I'm lazy and just play the first move without finding the full thread.

In 0G6WJ, I expected the puzzle to be over once the passed H pawn was established.

This topic has been archived and can no longer be replied to.