Knight odds test blitz match dragon vs GM Lenderman

[lkaufman]

This Thursday, April 8, from 1 pm to 5 pm Eastern Daylight time, we plan to play a four hour series of blitz games between the latest Komodo Dragon (MCTS mode) and GM Alex Lenderman (FIDE 2624 standard, 2642 Rapid, 2651 Blitz) on chess.com at knight odds. This is not a show, just a test of a proposed format for future matches, but anyone can watch the games (observe games of Playkomodo or AlexanderL). The format is this: the first game is at 9' + 1" increment (the slowest time control that chess.com classifies as blitz), at knight and move odds (meaning Nb8 removed). This is repeated if necessary until Lenderman wins a game, then he plays Black (so Nb1 removed), again repeating if necessary until he wins. Once he wins with Black, time limit is cut to 8' + 1", and he again must win a game as White and another as Black to promote to 7' + 1". After each pair of wins another minute is deducted. If we reaches 1' + 1" and wins with both colors, he'll just continue to play as Black at that handicap until the four hours is up. Dragon will be playing on my 32 core Threadripper, will have a tiny opening book to insure variety at knight odds, and will play quite a bit faster than the time limit would allow. I expect him to win most of the early games, but to get bogged down once the time limit gets a bit low. He'll get $20 for each win so he will be motivated; draws don't help him, it's effectively "knight odds Armageddon". What time limit will he reach in the four hours allotted?

[JJJ]

Nice ! I was waiting since a long time for this !

[Chessqueen]

This Thursday, April 8, from 1 pm to 5 pm Eastern Daylight time, we plan to play a four hour series of blitz games between the latest Komodo Dragon (MCTS mode) and GM Alex Lenderman (FIDE 2624 standard, 2642 Rapid, 2651 Blitz) on chess.com at knight odds. This is not a show, just a test of a proposed format for future matches, but anyone can watch the games (observe games of Playkomodo or AlexanderL). The format is this: the first game is at 9' + 1" increment (the slowest time control that chess.com classifies as blitz), at knight and move odds (meaning Nb8 removed). This is repeated if necessary until Lenderman wins a game, then he plays Black (so Nb1 removed), again repeating if necessary until he wins. Once he wins with Black, time limit is cut to 8' + 1", and he again must win a game as White and another as Black to promote to 7' + 1". After each pair of wins another minute is deducted. If we reaches 1' + 1" and wins with both colors, he'll just continue to play as Black at that handicap until the four hours is up. Dragon will be playing on my 32 core Threadripper, will have a tiny opening book to insure variety at knight odds, and will play quite a bit faster than the time limit would allow. I expect him to win most of the early games, but to get bogged down once the time limit gets a bit low. He'll get $20 for each win so he will be motivated; draws don't help him, it's effectively "knight odds Armageddon". What time limit will he reach in the four hours allotted?

Since Dragon has improved a lot in the last 11 months since the last match against GM Lenderman when it lost at Knight Odds at 3'+2", I believe that at 3'+1" it is time for an improved Dragon to get its revenge. So my answer is that GM Lenderman will reach 3'+1" time limit in his 4 hours alloted.

[Uri Blass]

I think that it may be interesing if there is not only promotion if you win but also demotion if you lose and of course for promotion you do not win money unless you promote to level that you were not in it earlier so if you demote from 9+1 to 10+1 you need to win 2 games without losses to get more money.

Of course it is harder to win money in this case so you may start with slower time control or increase the money that you give for promotion to a level that you were not in it earlier as compensation.

[Uri Blass]

I can add that without demotion it is better to resign if you see that you do not win even if you can probably still draw because you do not want to lose time.

I can also add that the best strategy for dragon to prevent the GM to earn money is never to resign and to use almost all the time that it has(except 0.01 second) in case it see that the opponent mate next move.

[lkaufman]

I can add that without demotion it is better to resign if you see that you do not win even if you can probably still draw because you do not want to lose time.

Yes, I thought of this and debated which way to go. Since the computer will always repeat the position if still down a piece or close to it, I think Armageddon knight odds makes more sense and so I went with the given format. It might be harder to get a top player to do this if he has to increase the odds after a loss. Your way is also interesting, but it seems reasonable to require that the piece-up side has to win the game, a draw is just too easy to achieve by repeating when still well ahead. In theory Contempt could be made large enough to avoid this, but my experience is that if Contempt is set too high, beyond about 2 pawns, Komodo will start to play noticeably worse.

[Nordlandia]

Is Armageddon mode planned to be used for this event ?

Not sure if odds + armageddon have been tried before prior to this match.

Monte Carlo + Armageddon is arguable worst for humans in such matches perhaps.

[lkaufman]

Is Armageddon mode planned to be used for this event ?

Not sure if odds + armageddon have been tried before prior to this match.

Monte Carlo + Armageddon is arguable worst for humans in such matches perhaps.

Theoretically Armageddon mode should be used for this match, but there is no real practical need for it, as Dragon will be aiming for draws anyway with Contempt set to only half or so of the handicap value. Of course, if dragon cuts the deficit to a pawn or so in the middlegame it may avoid draws, but there is almost no chance that it would lose such positions to a human opponent in blitz. By refusing a draw Dragon will actually improve its chances of minimizing losses given the four hour time limit, so I'm not inclined to use Armageddon mode, though I haven't ruled it out. As you say, it hasn't been tested much if at all with odds.

[lkaufman]

I can add that without demotion it is better to resign if you see that you do not win even if you can probably still draw because you do not want to lose time.

I can also add that the best strategy for dragon to prevent the GM to earn money is never to resign and to use almost all the time that it has(except 0.01 second) in case it see that the opponent mate next move.

That's true, but our main goal is to learn from the games and to entertain spectators so we won't have Dragon do this. Also we committed to make Dragon play much faster than necessary, it should rarely use more than half the allowed based time, though it will use the increment since MCTS mode is not really designed to play properly with much less than one second per move.

[lkaufman]

This Thursday, April 8, from 1 pm to 5 pm Eastern Daylight time, we plan to play a four hour series of blitz games between the latest Komodo Dragon (MCTS mode) and GM Alex Lenderman (FIDE 2624 standard, 2642 Rapid, 2651 Blitz) on chess.com at knight odds. This is not a show, just a test of a proposed format for future matches, but anyone can watch the games (observe games of Playkomodo or AlexanderL). The format is this: the first game is at 9' + 1" increment (the slowest time control that chess.com classifies as blitz), at knight and move odds (meaning Nb8 removed). This is repeated if necessary until Lenderman wins a game, then he plays Black (so Nb1 removed), again repeating if necessary until he wins. Once he wins with Black, time limit is cut to 8' + 1", and he again must win a game as White and another as Black to promote to 7' + 1". After each pair of wins another minute is deducted. If we reaches 1' + 1" and wins with both colors, he'll just continue to play as Black at that handicap until the four hours is up. Dragon will be playing on my 32 core Threadripper, will have a tiny opening book to insure variety at knight odds, and will play quite a bit faster than the time limit would allow. I expect him to win most of the early games, but to get bogged down once the time limit gets a bit low. He'll get $20 for each win so he will be motivated; draws don't help him, it's effectively "knight odds Armageddon". What time limit will he reach in the four hours allotted?

Since Dragon has improved a lot in the last 11 months since the last match against GM Lenderman when it lost at Knight Odds at 3'+2", I believe that at 3'+1" it is time for an improved Dragon to get its revenge. So my answer is that GM Lenderman will reach 3'+1" time limit in his 4 hours alloted.

That was also my opinion a week ago, now I'm a little more optimistic based on some refinements so I'll predict he'll just reach 4' + 1". Even 9' + 1" is not enough time for me to beat Dragon at knight odds, but I'm a long way from Lenderman's level in blitz, and even further with only a one second increment, so I can't really judge from my own games with Dragon. I would guess that whatever level he reaches, Hikaru could probably reach half that base time level at least if he plays it the same way.