Dragon vs GM Nakamura Analog Handicap Match

[Uri Blass]

Tang is one of the best in bullet but it seems when we look at blitz results
He is not one of the best.

I do not consider bullet to be normal chess because there are tricks that work against humans in bullet that are not going to work against engines.

For example when your opponent is in time trouble in bullet with no increment you can use some stupid move in order to win on time or win material and finally the game because the opponent has 2 options:
1)to play fast a move that he prepared against a logical move that is going to be a blunder against your stupid move.
2)not to play fast and look at your move and in this case he can expect to lose on time.

[lkaufman]

Dragon MCTS beat Komodo 14.1 in standard play ?

It should, although I'll have to run a match to be precise about that.

Excellent, if it can be done.

OK, it was a massacre. After 2060 bullet (30" + 0.30" inc) games, Dragon MCTS was +75 elo vs. Komodo 14.1!

[lkaufman]

Tang is one of the best in bullet but it seems when we look at blitz results
He is not one of the best.

I do not consider bullet to be normal chess because there are tricks that work against humans in bullet that are not going to work against engines.

For example when your opponent is in time trouble in bullet with no increment you can use some stupid move in order to win on time or win material and finally the game because the opponent has 2 options:
1)to play fast a move that he prepared against a logical move that is going to be a blunder against your stupid move.
2)not to play fast and look at your move and in this case he can expect to lose on time.

On chess.com blitz, he is currently 2973, 170 below Hikaru. That would make him 2730 FIDE blitz. Not "one of the best" in the world, but much better than his FIDE blitz rating.

[Nordlandia]

Is there anything that can be done to get Carlsen to play against Dragon. Carlsen has previously said he does not like playing engines. "It's like playing against an idiot who is unbeatable."

[lkaufman]

Dragon vs. my son Ray was drawn 2 to 2 at knight odds, same tc as the Naka match. So if the difference in the odds offsets the strength difference between Naka and Ray, the Dragon vs Naka match should be close. It should be noted that standard Dragon lost 1.5 to 0.5 to Ray, while MCTS Dragon won by 1.5 to 0.5. Of course that's not a statistically significant difference, just a hint that it's okay to use MCTS for at least some games.

Knight odds are trickier to simulate due to importance of some opening lines. With a naive opening set and hardly a use of the human opening advantage, I got +4 -1 =3 for Dragon against Lc0 (human). Anyway, all this is highly speculative, as is the prediction for Naka match.

You are correct. Nakamura knows the positions, and knows the program. And this gives Nakamura the advantage of prepping the opening.

I'm trying to follow your example by simulating knight odds between Dragon running on 7 fast cores/threads and Lc0 j92-330 at 15' + 10". I set Lc0 to max of one node per second. So far, White (Dragon) is down one game after 13 games. I'm new to using this feature of Lc0, so I have a couple questions.
1. Does it matter whether I use GPU (CUDA) Lc0 or CPU version? With max nps set, I wouldn't think it would matter, but I might be missing something.
2. What human rating at 15' + 10" would you estimate for j92-330 with max NPS = 1? How much higher if you double it to NPS = 2?
3. There are several drawscore options. How would you set drawscore for this?
4. Any other tricks or settings I should know about for this sort of thing?

[Chessqueen]

Dragon vs. my son Ray was drawn 2 to 2 at knight odds, same tc as the Naka match. So if the difference in the odds offsets the strength difference between Naka and Ray, the Dragon vs Naka match should be close. It should be noted that standard Dragon lost 1.5 to 0.5 to Ray, while MCTS Dragon won by 1.5 to 0.5. Of course that's not a statistically significant difference, just a hint that it's okay to use MCTS for at least some games.

Knight odds are trickier to simulate due to importance of some opening lines. With a naive opening set and hardly a use of the human opening advantage, I got +4 -1 =3 for Dragon against Lc0 (human). Anyway, all this is highly speculative, as is the prediction for Naka match.

You are correct. Nakamura knows the positions, and knows the program. And this gives Nakamura the advantage of prepping the opening.

I'm trying to follow your example by simulating knight odds between Dragon running on 7 fast cores/threads and Lc0 j92-330 at 15' + 10". I set Lc0 to max of one node per second. So far, White (Dragon) is down one game after 13 games. I'm new to using this feature of Lc0, so I have a couple questions.
1. Does it matter whether I use GPU (CUDA) Lc0 or CPU version? With max nps set, I wouldn't think it would matter, but I might be missing something.
2. What human rating at 15' + 10" would you estimate for j92-330 with max NPS = 1? How much higher if you double it to NPS = 2?
3. There are several drawscore options. How would you set drawscore for this?
4. Any other tricks or settings I should know about for this sort of thing?

You should ask your son to play again Dragon MCTS vs. your son Ray at knight odds using twice the amount of time control? Please do NOT use any endgame tablebase greater than 4 men it is a great disadvantage against any human. I do have the same version of Rebel that play versus Anand and drew and I also have a AMD 486 DX2 66 MHz in the baseman collecting dust all that I need is your latest Dragon MCTS.