stockfish 10 vs. Mephisto III S Glasgow

[JohnW]

There are no engines that have a plan in chess.

Imo it would increase the quality of the games.

Dont you think that THIS is a reason LC0 was such a sensation?! Because it played better chess.
And as a result of this todays AB thinkers use nnue.

I think it was a sensation because at one time it was the strongest engine out there and it was beating the other top engines.
I am not sure how you would define what a quality game is. How does a quality game from Lc0 differ from a non quality game from a stronger Stockfish?

[Raphexon]

There are no engines that have a plan in chess.

Imo it would increase the quality of the games.

Dont you think that THIS is a reason LC0 was such a sensation?! Because it played better chess.
And as a result of this todays AB thinkers use nnue.

NNUE doesn't have anything to do with Lc0 really.
The Shogi guys developed it independently and is more of an evolution of the NNs they used before (which also predates A0)

Lc0 was a sensation because it was beating Stockfish, often in positions were SF was completely blind.

[Uri Blass]

There are no engines that have a plan in chess.

Imo it would increase the quality of the games.

Dont you think that THIS is a reason LC0 was such a sensation?! Because it played better chess.
And as a result of this todays AB thinkers use nnue.

NNUE doesn't have anything to do with Lc0 really.
The Shogi guys developed it independently and is more of an evolution of the NNs they used before (which also predates A0)

Lc0 was a sensation because it was beating Stockfish, often in positions were SF was completely blind.

Stockfish needed NNUE to beat LC0 so I disagree that NNUE does not have anything to do with LC0
It seems that NNUE was enough to prevent losses of stockfish against LC0 and fixed the weakness of SF against LC0

I do not know if SF was going to use NNUE in case A0 and LC0 did not exist
and basically both NNUE and LC0 have the same idea of writing some evaluation that humans do not understand
and I think that this make the engines less human friendly for analysis because it is frustrating to see some evaluation of a position with no idea what is the reason for the evaluation.

[Sopel]

There are no engines that have a plan in chess.

Imo it would increase the quality of the games.

Dont you think that THIS is a reason LC0 was such a sensation?! Because it played better chess.
And as a result of this todays AB thinkers use nnue.

NNUE doesn't have anything to do with Lc0 really.
The Shogi guys developed it independently and is more of an evolution of the NNs they used before (which also predates A0)

Lc0 was a sensation because it was beating Stockfish, often in positions were SF was completely blind.

Stockfish needed NNUE to beat LC0 so I disagree that NNUE does not have anything to do with LC0
It seems that NNUE was enough to prevent losses of stockfish against LC0 and fixed the weakness of SF against LC0

I do not know if SF was going to use NNUE in case A0 and LC0 did not exist
and basically both NNUE and LC0 have the same idea of writing some evaluation that humans do not understand
and I think that this make the engines less human friendly for analysis because it is frustrating to see some evaluation of a position with no idea what is the reason for the evaluation.

If you dumb down the approaches all the way to "some evaluation that humans do not understand" then sure, they look the same.

and I think that this make the engines less human friendly for analysis because it is frustrating to see some evaluation of a position with no idea what is the reason for the evaluation

you're trying to say that classical eval, that went through a tree of billions of nodes, is somehow more human friendly?

Stockfish needed NNUE to beat LC0 so I disagree that NNUE does not have anything to do with LC0

That's not how relationships are defined

[Sopel]

There are no engines that have a plan in chess.

Imo it would increase the quality of the games.

Dont you think that THIS is a reason LC0 was such a sensation?! Because it played better chess.
And as a result of this todays AB thinkers use nnue.

Do you not consider a principal variation "a plan"? Why?

[Raphexon]

There are no engines that have a plan in chess.

Imo it would increase the quality of the games.

Dont you think that THIS is a reason LC0 was such a sensation?! Because it played better chess.
And as a result of this todays AB thinkers use nnue.

NNUE doesn't have anything to do with Lc0 really.
The Shogi guys developed it independently and is more of an evolution of the NNs they used before (which also predates A0)

Lc0 was a sensation because it was beating Stockfish, often in positions were SF was completely blind.

Stockfish needed NNUE to beat LC0 so I disagree that NNUE does not have anything to do with LC0
It seems that NNUE was enough to prevent losses of stockfish against LC0 and fixed the weakness of SF against LC0

I do not know if SF was going to use NNUE in case A0 and LC0 did not exist
and basically both NNUE and LC0 have the same idea of writing some evaluation that humans do not understand
and I think that this make the engines less human friendly for analysis because it is frustrating to see some evaluation of a position with no idea what is the reason for the evaluation.

"Stockfish needed NNUE to beat LC0 so I disagree that NNUE does not have anything to do with LC0"
By that same logic you'd disagree that search patches had nothing to do with Lc0 because SF needed them to not get crushed.
And that logic is wrong.

I do not know if SF was going to use NNUE in case A0 and LC0 did not exist

The devs wouldn't have ignored 40 elo, and that was a very early net and unoptimized code.

[Madeleine Birchfield]

Stockfish needed NNUE to beat LC0 so I disagree that NNUE does not have anything to do with LC0
It seems that NNUE was enough to prevent losses of stockfish against LC0 and fixed the weakness of SF against LC0

I do not know if SF was going to use NNUE in case A0 and LC0 did not exist
and basically both NNUE and LC0 have the same idea of writing some evaluation that humans do not understand
and I think that this make the engines less human friendly for analysis because it is frustrating to see some evaluation of a position with no idea what is the reason for the evaluation.

The only thing that would change if Leela and AlphaZero didn't exist is that the SV nets and StockfiNN nets would never have been used in Stockfish NNUE, as Sergio Vieri and Mark Jordan would have likely never joined the computer chess community without the Leela project. Instead, it'll probably be Henk Drost or Joerg Oster or Dietrich Kappe or Tomasz Sobczyk or some other net trainer who would continue to generate the nets like what happened in June 2020.

Scratch that, there is one other thing that would change. Norman Schmidt would have to generate his own net for Fire, as the current net in Fire is an old SV net.

[Madeleine Birchfield]

I think it was a sensation because at one time it was the strongest engine out there and it was beating the other top engines.

Leela was a sensation because TCEC abandoned its rule of equal hardware for all engines that participate in its tournaments in Season 13 and decided to give Leela a bunch of strong GPUs to use. Then everybody was extremely impressed that Leela on stronger hardware was able to defeat the Stockfish of that time. Leela CPU is nowhere near the top engines, even in the days of season 17 when Leela GPU was destroying Stockfish in the superfinal.

[Madeleine Birchfield]

Dont you think that THIS is a reason LC0 was such a sensation?! Because it played better chess.
And as a result of this todays AB thinkers use nnue.

Lc0 did not play better chess. It had a better evaluation function that allowed it to play better in the opening and in positional middlegames, but its search function was crap that made it extremely vulnerable to tactics and endgame shenanigans. One would think playing better chess would also entail not blundering the game away due to some MCTS failure. And Lc0 only did better than the top engines on the GPU, which made its algorithms somewhat useless for CPU based engines.

The CPU based engines started using NNUE (which was derived from the Stockfish clones used in Shogi) because it proved to be an instant 200+ elo gain for every engine, and training a net is much easier and requires less chess-specific knowledge than spending hours tinkering with each eval patch that might fail upon test. They didn't really care about Lc0, and the Lc0 community doesn't really care about the CPU based engines.

[Madeleine Birchfield]

What if a chess engine with a plan beats the engines without it ?!

That would mean the notion of „no plan“ is just a workaround for lack of higher quality of chess because the programmers have no clue about chess.

A 32-man tablebase would mathematically solve chess; no other engine would be able to beat it by definition. Would a chess engine with a 32-man tablebase have "a plan"?