Amazing results of LeelaKnightOdds-Dev

[Uri Blass]

I think that the question if to allow the opponent to make a good sacrifice is dependent on the style of the opponent because it may be a good practical decision if the opponent is afraid to sacrifice.

Different players at the same rating can play differently so
leela may learn from history of games against the specific player if it is probably a good decision or not.

Learning from the game history of the opponent is currently beyond the capability of Leela or any other engine, but who knows, some day this may be possible. Already the "impossible" has happened with these bots.

Good evening Mr. Larry Kaufman. I would like to know what is the meaning of the bullet time control? 1+ 0 is the same as 1+1 and 1+2; 3+ 0 is the same as 3+2 and 3+5. ...I would only be on my return to the streets in fights with bare hands and without boxing gloves, willing to fight or 1+0: 1+1 and 3 +0. How do time controls affect the elo classification of the top 100 table and how many games must be played to access the elo? In advance again, thank you. I also don't understand the reason why there can't be a top 100 table on LeelaknightOdds. The better will be the human who scores the most points in a series of confrontations, regardless of how few the human's points are... excuse my ignorance, I remind you that I am an apprentice poet...

For the meaning of bullet time control.
Lichess time controls are based on estimated game duration = (clock initial time in seconds) + 40 × (clock increment).
For instance, the estimated duration of a 5+3 game is 5 × 60 + 40 × 3 = 420 seconds.

≤ 29s = UltraBullet
≤ 179s = Bullet
≤ 479s = Blitz
≤ 1499s = Rapid
≥ 1500s = Classical

For the elo classification of the top 100 table if you play at longer time control then it means you play against an opponent with bigger rating.

LeelaQueenOdds as white 1+0 (Reference) 2611.8
LeelaQueenOdds as black 1+0 (Reference) 2561.8
LeelaQueenOdds as white 1+1 (Reference) 2461.0
LeelaQueenOdds as white 3+0 (Reference) 2447.1
LeelaQueenOdds as black 1+1 (Reference) 2411.0
LeelaQueenOdds as black 3+0 (Reference) 2397.1
LeelaQueenOdds as white 3+2 (Reference) 2349.6
LeelaQueenOdds as black 3+2 (Reference) 2299.6
LeelaQueenOdds as white 5+3 (Reference) 2291.4
LeelaQueenOdds as black 5+3 (Reference) 2241.4
LeelaQueenOdds as white 10+5 (Reference) 2210.9
LeelaQueenOdds as black 10+5 (Reference) 2160.9
LeelaQueenOdds as white 15+10 (Reference) 2155.2
LeelaQueenOdds as black 15+10 (Reference) 2105.2

The reason there is no top 100 table with knightodds is that the person who does the table is interested only in queen odd table.
It is more interesting to have a table when humans players can do more than 50% against the bot and it seems to be impossible with knight odds except maybe top players who do not compete.

Of course you are free to use 3+0 and try to make draws to get to the top of the table with queen odds.

Note that some humans who played against dragonkomodo without a knight for money in israel told me that they are going to play the bot.

I am afraid that for 3+0 time control or 1+1 time control lc0 will have to do queen and rook odds bot to allow the best players who practically play to get more than 50%

[lkaufman]

I think that the question if to allow the opponent to make a good sacrifice is dependent on the style of the opponent because it may be a good practical decision if the opponent is afraid to sacrifice.

Different players at the same rating can play differently so
leela may learn from history of games against the specific player if it is probably a good decision or not.

Learning from the game history of the opponent is currently beyond the capability of Leela or any other engine, but who knows, some day this may be possible. Already the "impossible" has happened with these bots.

Good evening Mr. Larry Kaufman. I would like to know what is the meaning of the bullet time control? 1+ 0 is the same as 1+1 and 1+2; 3+ 0 is the same as 3+2 and 3+5. ...I would only be on my return to the streets in fights with bare hands and without boxing gloves, willing to fight or 1+0: 1+1 and 3 +0. How do time controls affect the elo classification of the top 100 table and how many games must be played to access the elo? In advance again, thank you. I also don't understand the reason why there can't be a top 100 table on LeelaknightOdds. The better will be the human who scores the most points in a series of confrontations, regardless of how few the human's points are... excuse my ignorance, I remind you that I am an apprentice poet...

For the meaning of bullet time control.
Lichess time controls are based on estimated game duration = (clock initial time in seconds) + 40 × (clock increment).
For instance, the estimated duration of a 5+3 game is 5 × 60 + 40 × 3 = 420 seconds.

≤ 29s = UltraBullet
≤ 179s = Bullet
≤ 479s = Blitz
≤ 1499s = Rapid
≥ 1500s = Classical

For the elo classification of the top 100 table if you play at longer time control then it means you play against an opponent with bigger rating.

LeelaQueenOdds as white 1+0 (Reference) 2611.8
LeelaQueenOdds as black 1+0 (Reference) 2561.8
LeelaQueenOdds as white 1+1 (Reference) 2461.0
LeelaQueenOdds as white 3+0 (Reference) 2447.1
LeelaQueenOdds as black 1+1 (Reference) 2411.0
LeelaQueenOdds as black 3+0 (Reference) 2397.1
LeelaQueenOdds as white 3+2 (Reference) 2349.6
LeelaQueenOdds as black 3+2 (Reference) 2299.6
LeelaQueenOdds as white 5+3 (Reference) 2291.4
LeelaQueenOdds as black 5+3 (Reference) 2241.4
LeelaQueenOdds as white 10+5 (Reference) 2210.9
LeelaQueenOdds as black 10+5 (Reference) 2160.9
LeelaQueenOdds as white 15+10 (Reference) 2155.2
LeelaQueenOdds as black 15+10 (Reference) 2105.2

The reason there is no top 100 table with knightodds is that the person who does the table is interested only in queen odd table.
It is more interesting to have a table when humans players can do more than 50% against the bot and it seems to be impossible with knight odds except maybe top players who do not compete.

Of course you are free to use 3+0 and try to make draws to get to the top of the table with queen odds.

Note that some humans who played against dragonkomodo without a knight for money in israel told me that they are going to play the bot.

I am afraid that for 3+0 time control or 1+1 time control lc0 will have to do queen and rook odds bot to allow the best players who practically play to get more than 50%

Actually the person who made the does the table is the one who made the knight odds net, not the queen odds net! But we all think that a leaderboard is not very useful if you can be at the top without winning a game. The idea of odds in chess is to equalize (roughly) the chances, so it is like normal chess in that both sides try to win while switching to playing for a draw if things aren't going well. So we usually tell the engine to play as if the opponent was strong enough to score 50% at the chosen handicap. If no one plays at time controls where this is plausible, then players may offer very early repetitions and get draws due to the engine "thinking" that they are strong players. We could fix this by assuming weaker play, and in fact the queen for knight bot already does this because it uses the queen odds net, so we could do a leaderboard for that handicap if it became more popular with master level players. Expect another leaderboard before long for at least some other handicap or handicaps, and some new handicaps.

Regarding play at 3+0 or 1+1, it looks like I already have to admit that I was wrong just a week or two ago when I doubted that any engine would be able to give queen odds successfully at 3 + 0 to 2500 rated Lichess players. The current estimate is about 2450, and may even be a slight underestimate. We may not be at 2500 today, but clearly we are close enough to expect to reach it within a few months at most. Happy to be wrong about this!

Regarding bullet chess, while 1 + 0, 1 + 1, and 2 + 1 or 1 +2 are all considered "bullet", the first is primarily a test of non-chess skills like premoving, mouse speed, etc., while the others at least feel a bit like real chess, the result is usually decided on the board. At 1 + 0 even top GMs might struggle to win with queen odds, but with a 1 second increment added they should usually win at queen odds, but not at rook odds.

[Uri Blass]

I see players resign too early against lc0.
In this game black resigned in a drawn endgame and black had a lot of time on the clock.
https://lichess.org/95jKl1vM#151

[Uri Blass]

finally a strong chess player from israel beat the QueenOdd bot at 2+3 time control.

Fide master Ori Taichman
Fide rating 2345.

7-3 and 1 draw for Ori Taichman

[Uri Blass]

Ori said afrer winning the last 3 games that it is not as easy as he thought but he believes that he can win 10 games in a row not in bullet and with at least 2 seconds increment.

He said that he played 2+3 but practically he almost did not use the 2 minutes in all games because he did not take it completely seriously like title tuesday.

[Uri Blass]

I find that in one of the game that leela lost she could get a draw.

https://lichess.org/TRZb7e3D#157

80.Rxh2+ is a draw
There are cases when leela won drawn positions thanks to a losing move but of course not in this case.
Does leela use tablebases and if yes for how many pieces?

[Father]

I find that in one of the game that leela lost she could get a draw.

https://lichess.org/TRZb7e3D#157

80.Rxh2+ is a draw
There are cases when leela won drawn positions thanks to a losing move but of course not in this case.
Does leela use tablebases and if yes for how many pieces?

… The best humans playing against chess computers will be those who obtain the highest score; The way in which the score is acquired by the human lacks all importance and relevance, likewise, whether or not the human is a coffee player or a federated player. For example, at this moment Uriblass has a higher and better position on the table than the one I occupy and some Masters have. There is nothing to discuss on this issue and any system of protectorate is manipulation and Manichaeism. Protecting the elo of machines or any human or non-human elite, for and only because of who they are, is simply endorsing lies. Democracy in chess is the result of the advent of machines, software and hardware, as well as thanks to its Fathers, the programmers and creative teams.

[Uri Blass]

I find that in one of the game that leela lost she could get a draw.

https://lichess.org/TRZb7e3D#157

80.Rxh2+ is a draw
There are cases when leela won drawn positions thanks to a losing move but of course not in this case.
Does leela use tablebases and if yes for how many pieces?

… The best humans playing against chess computers will be those who obtain the highest score; The way in which the score is acquired by the human lacks all importance and relevance, likewise, whether or not the human is a coffee player or a federated player. For example, at this moment Uriblass has a higher and better position on the table than the one I occupy and some Masters have. There is nothing to discuss on this issue and any system of protectorate is manipulation and Manichaeism. Protecting the elo of machines or any human or non-human elite, for and only because of who they are, is simply endorsing lies. Democracy in chess is the result of the advent of machines, software and hardware, as well as thanks to its Fathers, the programmers and creative teams.

I do not agree.
There are better players than me against chess computers that simply did not play enough games to get higher score than me.

For example Orichess88 that is number 10 in the list because he played only 11 games.

[Uri Blass]

It seems that leela needs some improvement in the endgame to play perfect in winning positions(there is no reason to play with the idea that the opponent is going to go wrong in this case)

https://lichess.org/hSvoGdjq/black

She won this game but 45...Kd4 was a mistake because it draws against perfect play when 45...Kf4 is winning and there is no reason to let the opponent get a draw in a winning position.

[Uri Blass]

I look at more games that leela won in the endgame to see what blunders you can hope human opponent are going to make and I found the following game when it is unbelievable that bishop and 3 pawns lost against the rook.

https://lichess.org/D713U2Yk/black#125