Opening Theory according to Computers

[Peter Berger]

For quite some time we have been getting this line by Stockfish and lc0 , and they really like it for white for quite some time – what is it that they don’t understand? Or maybe it is us humans who don’t understand?

For human eyes even of very mediocre players the position after 9. d4 already looks trivial to understand from black’s point of view: rooks get traded on the e-file, knight goes back to e8, d5 gets played , Bishop goes to f5, knight back to d6 – and this should be dead equal and very hard to lose.
Once the knight lands on d6, Stockfish kind of agrees and the eval becomes equal.

A sample line Stockfish-me:
1. e4 e5 2. Nf3 Nc6 3. Bb5 Nf6 4. O-O Nxe4 5. Re1 Nd6 6. Nxe5 Nxe5 7. Rxe5+ Be7 8. Bf1 O-O 9. d4 Bf6 10. Re1 Re8 11. c3 Rxe1 12. Qxe1 Ne8 13. Bf4 d5 14. Nd2 Bf5 15. a4 a5 16. Qe2 Nd6 17. Nf3 h6

[carldaman]

After 1.e4, White's game, as Gyula Breyer brilliantly put it some 100 years ago, is in its last throes!

All joking aside, not sure if you've noticed, but these NNUE engines have one serious flaw, as they've adopted a safety-first approach, playing as if not to lose, rather than taking any risks in trying to go for a win. Their wins come as a result of other engines beating themselves with some bad moves, and also some very strong transition play into winning endgames.

You'll have to take any analysis they provide with that grain of salt in mind.

[jtwright]

After 1.e4, White's game, as Gyula Breyer brilliantly put it some 100 years ago, is in its last throes!

All joking aside, not sure if you've noticed, but these NNUE engines have one serious flaw, as they've adopted a safety-first approach, playing as if not to lose, rather than taking any risks in trying to go for a win. Their wins come as a result of other engines beating themselves with some bad moves, and also some very strong transition play into winning endgames.

You'll have to take any analysis they provide with that grain of salt in mind.

Is that a feature of NNUE or of very deep A/B search? If an engine just keeps refuting any attacks it can dream up it makes sense it would "play not to lose." Or even just the natural result of incredibly potent chess knowledge (and expectations that your opponent shares such knowledge). After all, lc0 is not an NNUE nor an A/B engine and according to OP, it likes this line too.

[Sopel]

After 1.e4, White's game, as Gyula Breyer brilliantly put it some 100 years ago, is in its last throes!

All joking aside, not sure if you've noticed, but these NNUE engines have one serious flaw, as they've adopted a safety-first approach, playing as if not to lose, rather than taking any risks in trying to go for a win. Their wins come as a result of other engines beating themselves with some bad moves, and also some very strong transition play into winning endgames.

You'll have to take any analysis they provide with that grain of salt in mind.

Is that a feature of NNUE or of very deep A/B search? If an engine just keeps refuting any attacks it can dream up it makes sense it would "play not to lose." Or even just the natural result of incredibly potent chess knowledge (and expectations that your opponent shares such knowledge). After all, lc0 is not an NNUE nor an A/B engine and according to OP, it likes this line too.

No, he just realized that winning in chess at highest level means that the opponent made a mistake. It's more apparent now that engines have gotten stronger, through NNUE.

[Peter Berger]

No, he just realized that winning in chess at highest level means that the opponent made a mistake. It's more apparent now that engines have gotten stronger, through NNUE.

No, that's not my point. The engines give a too high eval for white, and only realize that the position is actually 0.00 if you continue for a few more moves. And we are not talking many moves here, and none of the black moves have any deep point.

And they show this behaviour even at a very long time control.

[amanjpro]

No, he just realized that winning in chess at highest level means that the opponent made a mistake. It's more apparent now that engines have gotten stronger, through NNUE.

No, that's not my point. The engines give a too high eval for white, and only realize that the position is actually 0.00 if you continue for a few more moves. And we are not talking many moves here, and none of the black moves have any deep point.

And they show this behaviour even at a very long time control.

What does "the engines" mean here? Zahak doesn't give a too high advantage to any sides throughout your openning line. Eval in itself doesn't really mean anything especially with NNUE (the numbers used to correlate to the worth of a pawn, but this is no longer the case). We try to keep them close, but we never get them right and this is OK. After all the engine is not required to tell you if a position is +0.1 for white or +0.2, but rather if move 1 leads to a better position than move2 or what

[Peter Berger]

No, he just realized that winning in chess at highest level means that the opponent made a mistake. It's more apparent now that engines have gotten stronger, through NNUE.

No, that's not my point. The engines give a too high eval for white, and only realize that the position is actually 0.00 if you continue for a few more moves. And we are not talking many moves here, and none of the black moves have any deep point.

And they show this behaviour even at a very long time control.

What does "the engines" mean here? Zahak doesn't give a too high advantage to any sides throughout your openning line. Eval in itself doesn't really mean anything especially with NNUE (the numbers used to correlate to the worth of a pawn, but this is no longer the case). We try to keep them close, but we never get them right and this is OK. After all the engine is not required to tell you if a position is +0.1 for white or +0.2, but rather if move 1 leads to a better position than move2 or what

Sorry, you are right. We are obviously just talking lc0 and stockfish ( since it is being trained with lc0 data) here. Of course there will be several engines who just show sth like 0.00 I assume. But you have to understand, that this is not "my opening line" but the line that every chess player who ever tried has been made to think about for at least a year now - this is what you get, when you start them and let them analyze and do their own thing.

[MonteCarlo]

There's a lot of very vague language in the original post. I'm curious to replicate, so for my benefit what do "really like", "too high eval", and "quite some time" mean in terms of evals and depths?

I ask because I usually let new versions of SF analyze very early opening positions just to see whether/how much such things change from version to version, and I don't remember seeing any especially high eval in these lines.

It has been a while, so I may be misremembering, and I don't think I've done this with SF 14.1, just SF14, so maybe there are some differences there.

I suspect the main gaps will be in the definitions of those terms, though.

Cheers!

[supersharp77]

For quite some time we have been getting this line by Stockfish and lc0 , and they really like it for white for quite some time – what is it that they don’t understand? Or maybe it is us humans who don’t understand?

For human eyes even of very mediocre players the position after 9. d4 already looks trivial to understand from black’s point of view: rooks get traded on the e-file, knight goes back to e8, d5 gets played , Bishop goes to f5, knight back to d6 – and this should be dead equal and very hard to lose.
Once the knight lands on d6, Stockfish kind of agrees and the eval becomes equal.

A sample line Stockfish-me:
1. e4 e5 2. Nf3 Nc6 3. Bb5 Nf6 4. O-O Nxe4 5. Re1 Nd6 6. Nxe5 Nxe5 7. Rxe5+ Be7 8. Bf1 O-O 9. d4 Bf6 10. Re1 Re8 11. c3 Rxe1 12. Qxe1 Ne8 13. Bf4 d5 14. Nd2 Bf5 15. a4 a5 16. Qe2 Nd6 17. Nf3 h6

Yes...This line recently appeared in some extremely high level Stockfish...Cfish...Eman...SF Development..NNUE Matches
I don't recall SF Classical search engines focusing on this particular Line...I believe it is some sort of NNUE bug or flaw of some sort...All the games were 100% dead even boring draws...Some of these engines were using opening books (bin) and or GUI opening books...It seem these NNUE engines are ignoring the books completely...I have not seen LC0 engines interested in such opening sequences as this....LC0 seems to prefer more complex lines with White or with black..more aggressive

[towforce]

Ultimately, choice of opening comes down to the question of what constitutes good chess:

Level 1:

Try to get an advantage, and then attack.


Level 2:

Understand your opponent very well, then guide them into a position in which their way of playing chess doesn't work. For computers, this might mean complex positions where there's too much going on, or too many choices, which would prevent their opponents from being able to see very far ahead.


Level 3:

Both yourself and your opponent have excellent knowledge of the game, and neither player is likely to ever make a move which would allow the other player to win. Death by draw.


Peter seems to be saying that in the opening, computers indicate an advantage, then two or three moves later only see a draw. Using my levels above, that would put us firmly in level one.