Stockfish: Our lawsuit against ChessBase

[AlexChess]

Hi dkappe, just a little OT question: which is stronger between Toga III and Toga 1.1 + Dark Rose NNUE? I get controversial results using Arena or BanksiaGui. Night Nurse CF and Dark Horse CF are much stronger, anyway. Thank you for these strong alternatives to SF test NNUEs!

Solved my mistery: Toga III 0.312 ssse is DIFFERENT from Toga III CF 64 Numa (much weaker) I replace the engine on BanksiaGui.

Regards, Alex

[AlexChess]

Stockfish criteria to test progresses are often criticized. Eg: in the first phase, 1 second to verify a potential improvement is ridiculous. Strategic understanding of some positions is still very weak for all top programs. Playing against the previous version gives only a partial view. A GM could judge much better an engine, playing more games @40/2h. The contribute of the largest panel of engines is very important, also if they are only a simple fine tuning of the same source code.

But I really don't like to interact with you... although I still love Ethereal and I really cannot ignore it

Kind regards, Alex

[AndrewGrant]

Stockfish criteria to test progresses are often criticized. Eg: in the first phase, 1 second to verify a potential improvement is ridiculous. Strategic understanding of some positions is still very weak for all top programs. Playing against the previous version gives only a partial view. A GM could judge much better an engine, playing more games @40/2h. The contribute of the largest panel of engines is very important, also if they are only a simple fine tuning of the same source code.

Stockfish testing is critized, but by people who don't have a realistic workable alternative. I'm sure SF team would happily run hour long games, if you convince Google to turn whole countries worth of servers over to them. Everyone recognizes, and remains surprised, that STC/LTC 10s/60s is sufficient. I see no one else doing anything other than SPRT. If there is a good alternative, someone should put it into practice.

[amanjpro]

Stockfish criteria to test progresses are often criticized. Eg: in the first phase, 1 second to verify a potential improvement is ridiculous. Strategic understanding of some positions is still very weak for all top programs. Playing against the previous version gives only a partial view. A GM could judge much better an engine, playing more games @40/2h. The contribute of the largest panel of engines is very important, also if they are only a simple fine tuning of the same source code.

Stockfish testing is critized, but by people who don't have a realistic workable alternative. I'm sure SF team would happily run hour long games, if you convince Google to turn whole countries worth of servers over to them. Everyone recognizes, and remains surprised, that STC/LTC 10s/60s is sufficient. I see no one else doing anything other than SPRT. If there is a good alternative, someone should put it into practice.

I used to do LTC, and apart from wasting too much time, power, I saw no benefit... yeah I agree STC is the way to go... if you ask me how? I really don't know, it is counter intuitive, but it works

I believe, apart from testing time management, STC makes the most sense

[amanjpro]

Stockfish criteria to test progresses are often criticized. Eg: in the first phase, 1 second to verify a potential improvement is ridiculous. Strategic understanding of some positions is still very weak for all top programs. Playing against the previous version gives only a partial view. A GM could judge much better an engine, playing more games @40/2h. The contribute of the largest panel of engines is very important, also if they are only a simple fine tuning of the same source code.

But I really don't like to interact with you... although I still love Ethereal and I really cannot ignore it

Kind regards, Alex

Problem of judging strategic understanding of a top engine, by our standards is a flaw, we are wayyyy too weak to be able to judge SF moves, even when they look terrible, they might be great.

I stopped analyzing Zahak's games by myself a long time ago, now I use SF only. I am too weak to understand Zahak's plans (even though it is 1000 elo points weaker than SF). I myself is only 900 elo points weaker than the top human player, so you see the difference here

[Uri Blass]

Stockfish criteria to test progresses are often criticized. Eg: in the first phase, 1 second to verify a potential improvement is ridiculous. Strategic understanding of some positions is still very weak for all top programs. Playing against the previous version gives only a partial view. A GM could judge much better an engine, playing more games @40/2h. The contribute of the largest panel of engines is very important, also if they are only a simple fine tuning of the same source code.

But I really don't like to interact with you... although I still love Ethereal and I really cannot ignore it

Kind regards, Alex

I find that stockfish has an impressive positional knowledge about chess endgames.

For example stockfish knows the following position is probably a draw and give white less than +1 even at low depths.
If I replace the rooks by knights stockfish understand that white is winning and show very fast more than +3
If I replace the rooks by bishops stockfish understand again that it is a draw and show a score that is very close to 0.00

I do not know of one program that does not use NNUE that can show this deep positional knowledge.

[d]5rk1/5p1p/5p2/8/8/4P3/4RPPP/4K3 w - - 0 1

[AlexChess]

Thanks Uri for your contribute,
Yes, I agree with you. What was the "Achilles heel" of the old dedicated chess computers (I'm talking about old vintage Kasparov like Renaissance computer ) dued also to storage limits, it's today completely solved by stunnings endgames algorithms and huge tablebases up to 7 man. I was talking about particular positions (that you have suggested months ago ) and very hard positions to understand also by a good player, but sure not by a GM.

[Cornfed]

Stockfish criteria to test progresses are often criticized. Eg: in the first phase, 1 second to verify a potential improvement is ridiculous. Strategic understanding of some positions is still very weak for all top programs. Playing against the previous version gives only a partial view. A GM could judge much better an engine, playing more games @40/2h. The contribute of the largest panel of engines is very important, also if they are only a simple fine tuning of the same source code.

But I really don't like to interact with you... although I still love Ethereal and I really cannot ignore it

Kind regards, Alex

I find that stockfish has an impressive positional knowledge about chess endgames.

For example stockfish knows the following position is probably a draw and give white less than +1 even at low depths.
If I replace the rooks by knights stockfish understand that white is winning and show very fast more than +3
If I replace the rooks by bishops stockfish understand again that it is a draw and show a score that is very close to 0.00

I do not know of one program that does not use NNUE that can show this deep positional knowledge.

[d]5rk1/5p1p/5p2/8/8/4P3/4RPPP/4K3 w - - 0 1

This is a pretty common endgame structure. I checked two engines: The current SF says White is 'slightly better' while Komodo 13.3 say White is 'better'.

While the doubled pawns are NOT normally a weakness in this type of ending...the stronger side (White, here) wins a surprising number of these - about 40% in my database which basically excludes untitled players.
To my mind, for practical play (and chess IS a 'game' after all...), Komodo 13.3's 'better' seems more correct even if "objectively" this leans more toward the 'draw' spectrum.

[wickedpotus]

I am utterly amazed how much OT some posters go here without reflecting much on the thread subject.

[Sopel]

Stockfish criteria to test progresses are often criticized. Eg: in the first phase, 1 second to verify a potential improvement is ridiculous. Strategic understanding of some positions is still very weak for all top programs. Playing against the previous version gives only a partial view. A GM could judge much better an engine, playing more games @40/2h. The contribute of the largest panel of engines is very important, also if they are only a simple fine tuning of the same source code.

But I really don't like to interact with you... although I still love Ethereal and I really cannot ignore it

Kind regards, Alex

People who critisize that Stockfish uses 1s games to verify potential improvements are irrelevant, because Stockfish uses 10s and 60s games to verify potential improvements.