Bad Bishop win

[Dann Corbit]

Of course the same thing is true of every chess heuristic.
A knight on the rim is dim. Except when it is the best move.
Knights before bishops. Except when it is better to move the bishop first.
Never develop rooks or queens until after the minor pieces. Except when it is the best thing to do.
Piece square tables are little gooey gobs of heuristic guesses.
And the new chess approach (LC0, AlphaChess0) is the most amazing heuristic mass of them all.
It looks at a big web of goo, then decides (nobody knows exactly how) that the weights gathered over experience make move x better, and it makes the move.

[chrisw]

As to 'there is no "correct"':
1. One definition of correct would be that it gains Elo in the desired environment of play.
2. Another definition of correct is that it causes more good decisions in analysis than bad decisions.
As to whether it is correct by either of these definitions or some other definition, I cannot say.

When we are talking about code, correct means two things to me:
1. It does not contain harmful bugs and performs the desired task as designed
2. The code causes the intended benefit, whatever it may be.

Now, correct and truth are not the same thing here.
I would agree that we cannot come to any conclusion about truth in these code snippets.

Correct means to make better/right/straight/improve.
Chess program evaluation chunk code is never “correct” because it can always be made “more correct” by accounting, somehow, for interdependencies. In chess positions, everything depends on everything else. “truth” being unattainable, as the 8x8 game has designed it to be. Correct is not a tick box to tick and move on. Dependencies rule.

[chrisw]

Of course the same thing is true of every chess heuristic.
A knight on the rim is dim. Except when it is the best move.
Knights before bishops. Except when it is better to move the bishop first.
Never develop rooks or queens until after the minor pieces. Except when it is the best thing to do.
Piece square tables are little gooey gobs of heuristic guesses.
And the new chess approach (LC0, AlphaChess0) is the most amazing heuristic mass of them all.
It looks at a big web of goo, then decides (nobody knows exactly how) that the weights gathered over experience make move x better, and it makes the move.

We do know how it decides. It does lots of multiplications and additions. We know what it adds to what and how much it multiplies everything. We know absolutely how it does it. What you mean is that you can’t relate what it is doing to what you do, or how you think. You want simple breakdowns of it and simple explanations in terms you understand. That’s because you think you can explain your own decisions in relatively simple breakdowns and explanations. I did XYZ because ABCD and E. That’s a delusion in itself. Really you did XYZ because you did lots of little neuron thingies and they all came out as XYZ action. Then you make up ABCD and E afterwards with some other neurons.

[BeyondCritics]

Interesting game, worth studying i think.
The superiority of black on the queenside is obvious, moreover white has the "bad bishop", which is not an active piece in any case. So why is white able to sacrifice d4? Stockfish itself will help us:
[d]2r3k1/3qbp1p/p3p1pP/P2pP3/2rP4/5NP1/2nBQPK1/2RR4 b - - 0 26
After 26...Nxd4 27.Nxd4 Rxd4 white has full compensation with the natural moves 28.Rb1 and 28.Rxc8 and can even sac a piece with 28.Qxa6!?.
In any case, white can target a6 and has an "outpost" on b6 for its rook, as already noted. But the primary reason is seemingly the eternal (!) weakness of blacks back rank, which in turn comes from white having established a "thorn pawn" on h6.
It has been noticed, that the fawn pawn strategy is seen much more often in mastergames with alpha zero and leela than previously, e.g. by FM Kingscrusher and others (https://chess.stackexchange.com/questio ... cades-past) and i agree with that.
I personally recall that back in time, when i studied chess strategy, i was often told that the "thorn pawn strategy" here or there would bad or even completely silly. Since you are investing valuable time and often in the short run you get nothing(!) in return but a liability and an unclear hope for the future. Having seen all this fawn pawn strategy games, i have really learned something.

[BeyondCritics]

In this position
[d] 2r2rk1/p2qbppp/4p3/1p1pP3/1n1P4/5N1P/PP1BQPP1/R2R2K1 w - - 1 17
Leela played 17.h4!?
This move never occurs to my stockfish@home, even 17.Kh1 would be preferred.

After having established the thorn pawn leela violently opens the queen side with 26.a4!?
[d] 2r3k1/3qbp1p/p1r1p1pP/1p1pP3/3P4/1P3NP1/P1nBQPK1/2RR4 w - - 2 24
This looks kind of insane to me. White concedes to weaknesses on the wing, where he is decidedly weaker, at least it appears so.
Again, my stockfish never considers that.

[MikeGL]

My hunch tells me that by removing that "bad bishop" penalty in eval function of SF10, it won't lower its strength since there is already mobility code which would detect that bad bishop. Hence, I think that is redundant. Ok, I haven't done my own compile and tests, but considering the brutal fast searchers with current hardware, that "bad bishop" code is irrelevant in my opinion and just wastes some clock cycles.

Maybe that was important during the pentium I or Pentium II days where nps was low and can't reach higher depths.

[Nay Lin Tun]

The answer is

560.jpg

Leela give Fawn pawn value at least half a piece!

Leela is playing with extra half piece bonus!!