Dragon 3.2 Released at KomodoChess.com

[Wolfgang]

Be careful what you write.

Dietrich will be really really scared now.....

[dkappe]

Do you understand what the purpose of an opening book for testing is?

It depends upon what you're wanting to achieve.

Having said that, most of us are hobbyists, and one can decide which results to look at, which rating lists to look at, which competitions to take an interest in, which type of opening books we prefer, and so on.

We've all had our share of criticism, which is okay as long as any such conversations are conducted in a respectful manner.

Well said. Not wanting to hijack this thread, I’ll start a new one with some questions I have about testing and opening books.

[lkaufman]

It's not an obvious thing to many, you see the same argument used by others all over the board that books or openings should be balanced. Instead of insulting it could do with a minimal explanation what you guys mean to achieve with unbalanced openings.

I think it's pretty simple. Unbalanced openings are only "fair" if each side plays White once. The argument against them is that this is not "real" chess, the opening positions would rarely occur in serious games between top modern GMs (even if they are taken from human master games, which may be old or blitz or Rapid games or below GM level games). It is not a question of "fairness", it is just a question of whether such openings are "real" chess. The benefit is of course that this allows for measuring engine progress, even though it seems that with "good" openings, good hardware, non-blitz time controls, and top - 3 engines, the draw rate is well over 99%; maybe soon we will never see another loss by current Stockfish or Dragon with these conditions. I admit that using unbalanced openings is not an ideal or even a desirable solution, but it is the best one I know of. I'm sure that we would not have made the ten elo or so gain we have made (with "good" openings) if we didn't test with unbalanced ones. Otherwise it's time to move on to another game for programming; at least Go and shogi have not reached anywhere near this point yet, although engines are both far beyond human level.

[CornfedForever]

It's not an obvious thing to many, you see the same argument used by others all over the board that books or openings should be balanced. Instead of insulting it could do with a minimal explanation what you guys mean to achieve with unbalanced openings.

I think it's pretty simple. Unbalanced openings are only "fair" if each side plays White once. The argument against them is that this is not "real" chess, the opening positions would rarely occur in serious games between top modern GMs (even if they are taken from human master games, which may be old or blitz or Rapid games or below GM level games).

+1. I mean, Magnus plays 1. a3 and such against major league human chess players. Quick TC's, but it's still a game of regular chess... as long as the starting position is normal.

[Peter Berger]

It's not an obvious thing to many, you see the same argument used by others all over the board that books or openings should be balanced. Instead of insulting it could do with a minimal explanation what you guys mean to achieve with unbalanced openings.

I think it's pretty simple. Unbalanced openings are only "fair" if each side plays White once. The argument against them is that this is not "real" chess, the opening positions would rarely occur in serious games between top modern GMs (even if they are taken from human master games, which may be old or blitz or Rapid games or below GM level games). It is not a question of "fairness", it is just a question of whether such openings are "real" chess. The benefit is of course that this allows for measuring engine progress, even though it seems that with "good" openings, good hardware, non-blitz time controls, and top - 3 engines, the draw rate is well over 99%; maybe soon we will never see another loss by current Stockfish or Dragon with these conditions. I admit that using unbalanced openings is not an ideal or even a desirable solution, but it is the best one I know of. I'm sure that we would not have made the ten elo or so gain we have made (with "good" openings) if we didn't test with unbalanced ones. Otherwise it's time to move on to another game for programming; at least Go and shogi have not reached anywhere near this point yet, although engines are both far beyond human level.

This is all very reasonable, of course.
The question is whether it is true.
We can probably all agree that correspondence chess is dead right now. This may be because the game is a draw, or just because humans are just too weak to be a positive factor in the engine games if they can avoid operator mistakes. Classical chess is the next obvious target in case this were due to the very nature of the game, but we don’t know.
But there is some merit in looking at things as Krzysztof does. It is artificial to use very unbalanced openings just to get a difference in performance between two engines who’d otherwise draw every single game.
Who knows whether scoring better at unbalanced openings makes you any better with equal starting position?! To me this is not completely obvious.
A similar argument could be made when it is about these ultra-fast time controls everyone uses right now.
It may be that Stockfish 20 will get 70% of the points against Stockfish 15 using balanced openings (or no book). Then we will say how dumb engines of the past used to be. Capablanca and Lasker used to claim the draw death of chess in the 20ies of the last century – we don’t believe they had a point anymore.

[Peter Berger]

OK, I see now that this is already being discussed in another thread. I didn't know, so please ignore my post.

[Graham Banks]

Dragon by Komodo 3.2 64-bit 4CPU Gauntlet for CCRL 40/15 underway in 16063.

[cc2150dx]

Dragon by Komodo 3.2 64-bit 4CPU Gauntlet for CCRL 40/15 underway in 16063.

Nice. Will you be also testing MCTS ?

[Graham Banks]

Dragon by Komodo 3.2 64-bit 4CPU Gauntlet for CCRL 40/15 underway in 16063.

Nice. Will you be also testing MCTS ?

Hadn't planned to do so.

[daniel71]

Hello Graham,
Do you have a link to watch the games? Thanks