GM Awonder Liang vs Leela Knight Odds Rapid match

[Uri Blass]

1)Sir_Patzer (2541) 3+2 117-1 and 7 draws
2)DannyStoll(2284?) 15+10 6-1
3)FM veni_vidi_vici2007(2682?) 10+10 20-2 and 1 draw when most of the games were 10+10 or 15+10 and lost also 3.5-0.5 with 15+10
4)Monchi(2471) 15+10lost 28-1 and most games were 15+10
5)CurtVonBardeleben(1962?) 15+10 3 times won all 3 game.
6)MassivePackage(2345) 5+3 lost 18-1`
7)BuckT(1500?) 3+10 32-4 and 3 draws mos of them 3+10
8)slomka88(2325) 5+3 lost 14-1
9)ScheveningenSea(2717) lost 62-4 and 10 draws(most games at 3+2 and the wins are at 3+2)
10)turton(1500?) lost 41-4 and 1 draw played 15+10
11)jlhammer(GM Jon Ludvig Hammer)(2444?) 5+3 won the only try.
12)FM OjaiJoao(2546) 5+5 FM Alex King lost 18.5-1.5 at 5+5
13)Stockblunder(1500?) 10+2 lost 47-1
14)No_Move_Left(2564) 3+2 lost 3-1
15)nebulosastella(1760?) 15+10 lost 79-1 and 3 draws and most games are 15+10
16)Bonitta 2502(3+2) lost 44-1 with 6 draws

[lkaufman]

Probably IM can score around 50% at rook odds in 15+10, but most of the games are blitz where the human score like 10% or so

In your next email, you show:
3)FM veni_vidi_vici2007(2682?) 10+10 20-2 and 1 draw when most of the games were 10+10 or 15+10 and lost also 3.5-0.5 with 15+10.

Although he is not an IM, only FM, all of his ratings are above average for an IM, and yet he lost by the above overwhelming margins. So I think it is clear that a typical IM has no chance in a Rapid rook odds match. Of course there may be 1 or 2 IMs in the world who are of GM strength and very good in Rapid play who might have a chance, but clearly GM level, if not the title, is required. I know of one GM who has a minus record in Rapid at Rook and Move odds after many games (i.e. he plays White), I don't mean myself, but a player who still plays at GM level. Furthermore the bot was upgraded Dec. 11 to use the new net trained for both knight odds and rook odds, used in the Awonder match (and the Giri games). Your compilation shows that the performance of the Rook odds bot in Rapid is so high that I believe only a 2600+ FIDE player (or someone of similar strength in Rapid) could win a match at 15'10" at rook odds against the current bot. Given that it seems to require something approaching CM level to win a Rapid match at queen odds, this makes sense to me.

[Father]

Probably IM can score around 50% at rook odds in 15+10, but most of the games are blitz where the human score like 10% or so

In your next email, you show:
3)FM veni_vidi_vici2007(2682?) 10+10 20-2 and 1 draw when most of the games were 10+10 or 15+10 and lost also 3.5-0.5 with 15+10.

Although he is not an IM, only FM, all of his ratings are above average for an IM, and yet he lost by the above overwhelming margins. So I think it is clear that a typical IM has no chance in a Rapid rook odds match. Of course there may be 1 or 2 IMs in the world who are of GM strength and very good in Rapid play who might have a chance, but clearly GM level, if not the title, is required. I know of one GM who has a minus record in Rapid at Rook and Move odds after many games (i.e. he plays White), I don't mean myself, but a player who still plays at GM level. Furthermore the bot was upgraded Dec. 11 to use the new net trained for both knight odds and rook odds, used in the Awonder match (and the Giri games). Your compilation shows that the performance of the Rook odds bot in Rapid is so high that I believe only a 2600+ FIDE player (or someone of similar strength in Rapid) could win a match at 15'10" at rook odds against the current bot. Given that it seems to require something approaching CM level to win a Rapid match at queen odds, this makes sense to me.

..this Odds bot is new to me... we just did the first two laps of the racetrack, and I could see that the galloping and racing style is different... it is definitely different from the OddsQueenforKnight and the LeelaQueenOdds bot...

[pgn][Event "Casual blitz game"]
[Site "https://lichess.org/dVJ7dB8z"]
[Date "2024.12.16"]
[White "Catecan"]
[Black "LeelaRookOdds"]
[Result "1/2-1/2"]
[UTCDate "2024.12.16"]
[UTCTime "17:51:04"]
[WhiteElo "2082"]
[BlackElo "2000"]
[BlackTitle "BOT"]
[Variant "From Position"]
[TimeControl "180+2"]
[ECO "?"]
[Opening "?"]
[Termination "Normal"]
[FEN "1nbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQk - 0 1"]
[SetUp "1"]
[Annotator "lichess.org"]

1. d4 Nf6 2. f4 c5 3. e3 e6 4. Nf3 d5 5. c3 Bd6 6. Bd3 O-O 7. O-O b6 8. a3 Nc6 9. Qe2 Bb7 10. Ba6 Ba8 11. Bd3 Bb7 12. Ba6 Ba8 13. Bd3 Ne7 14. Re1 Ne4 15. g3 h6 16. Nbd2 Qc8 17. Nf1 Bb7 18. Qd1 f6 19. Re2 Bc7 20. Rg2 e5 21. Qc2 c4 22. Be2 Re8 23. Bd2 b5 24. Be1 a5 25. Qc1 a4 26. Qc2 Rf8 27. Rc1 h5 28. N3d2 Qh3 29. Nf3 Bc6 30. Bd1 Be8 31. Qe2 Bd7 32. Qc2 g6 33. Qe2 Bg4 34. Qc2 Bd7 35. Qe2 Bg4 36. Qc2 Bc8 37. Qe2 Bf5 38. Ra1 Bg4 39. Rc1 Kg7 40. Ra1 Rh8 41. Qc2 Rf8 42. Qe2 Bb8 43. Qc2 Bc7 44. Qe2 { The game is a draw. } 1/2-1/2[/pgn]

[Uri Blass]

Probably IM can score around 50% at rook odds in 15+10, but most of the games are blitz where the human score like 10% or so

In your next email, you show:
3)FM veni_vidi_vici2007(2682?) 10+10 20-2 and 1 draw when most of the games were 10+10 or 15+10 and lost also 3.5-0.5 with 15+10.

Although he is not an IM, only FM, all of his ratings are above average for an IM, and yet he lost by the above overwhelming margins. So I think it is clear that a typical IM has no chance in a Rapid rook odds match. Of course there may be 1 or 2 IMs in the world who are of GM strength and very good in Rapid play who might have a chance, but clearly GM level, if not the title, is required. I know of one GM who has a minus record in Rapid at Rook and Move odds after many games (i.e. he plays White), I don't mean myself, but a player who still plays at GM level. Furthermore the bot was upgraded Dec. 11 to use the new net trained for both knight odds and rook odds, used in the Awonder match (and the Giri games). Your compilation shows that the performance of the Rook odds bot in Rapid is so high that I believe only a 2600+ FIDE player (or someone of similar strength in Rapid) could win a match at 15'10" at rook odds against the current bot. Given that it seems to require something approaching CM level to win a Rapid match at queen odds, this makes sense to me.

I do not think you need CM level or something close to it to win a Rapid match at queen odds.
I am not CM level that is 2200 fide rating and I believe I can win a rapid match 15+10 at queen odds.

My best fide rating was 2051 and today I am below 2000(I may be better than my 1908 fide rating and the reason for my failure in the last tournament when I lost 20 elo is trying new openings that I did not know well that I wanted to get experience in them but I guess that I am not at level of above 2000 fide rating).

I did not play recently against the bot because I prefer to use my time in chess to study some opening theory and not to play again and again against the bot.

[Uri Blass]

Probably IM can score around 50% at rook odds in 15+10, but most of the games are blitz where the human score like 10% or so

In your next email, you show:
3)FM veni_vidi_vici2007(2682?) 10+10 20-2 and 1 draw when most of the games were 10+10 or 15+10 and lost also 3.5-0.5 with 15+10.

Although he is not an IM, only FM, all of his ratings are above average for an IM, and yet he lost by the above overwhelming margins. So I think it is clear that a typical IM has no chance in a Rapid rook odds match. Of course there may be 1 or 2 IMs in the world who are of GM strength and very good in Rapid play who might have a chance, but clearly GM level, if not the title, is required. I know of one GM who has a minus record in Rapid at Rook and Move odds after many games (i.e. he plays White), I don't mean myself, but a player who still plays at GM level. Furthermore the bot was upgraded Dec. 11 to use the new net trained for both knight odds and rook odds, used in the Awonder match (and the Giri games). Your compilation shows that the performance of the Rook odds bot in Rapid is so high that I believe only a 2600+ FIDE player (or someone of similar strength in Rapid) could win a match at 15'10" at rook odds against the current bot. Given that it seems to require something approaching CM level to win a Rapid match at queen odds, this makes sense to me.

I have some problem to believe that you need 2600 to beat the bot with rook odds with 15+10 time control and I will explain my reasons:
1)When I consider the result of liang it seems that liang should be able at least to beat the engine at 15+10 time control with knight odds(he could beat it with knight odds with 10+5 but not enough games to be sure if he can win in this time control).
2)I believe that the difference between rooks odd and knight odds is more than 100 elo.
3)Liang has rating of less than 2700 so rating of less than 2600 should be enough to win at rapid with rook odds.

Maybe there is a big difference between different players when there may be players with difference of 400 elo who have the same results with rook odds against leela so some 2600 player lose against the rook odds bot at 15+10 but some 2300 player can win against the bot.

[lkaufman]

Probably IM can score around 50% at rook odds in 15+10, but most of the games are blitz where the human score like 10% or so

In your next email, you show:
3)FM veni_vidi_vici2007(2682?) 10+10 20-2 and 1 draw when most of the games were 10+10 or 15+10 and lost also 3.5-0.5 with 15+10.

Although he is not an IM, only FM, all of his ratings are above average for an IM, and yet he lost by the above overwhelming margins. So I think it is clear that a typical IM has no chance in a Rapid rook odds match. Of course there may be 1 or 2 IMs in the world who are of GM strength and very good in Rapid play who might have a chance, but clearly GM level, if not the title, is required. I know of one GM who has a minus record in Rapid at Rook and Move odds after many games (i.e. he plays White), I don't mean myself, but a player who still plays at GM level. Furthermore the bot was upgraded Dec. 11 to use the new net trained for both knight odds and rook odds, used in the Awonder match (and the Giri games). Your compilation shows that the performance of the Rook odds bot in Rapid is so high that I believe only a 2600+ FIDE player (or someone of similar strength in Rapid) could win a match at 15'10" at rook odds against the current bot. Given that it seems to require something approaching CM level to win a Rapid match at queen odds, this makes sense to me.

I have some problem to believe that you need 2600 to beat the bot with rook odds with 15+10 time control and I will explain my reasons:
1)When I consider the result of liang it seems that liang should be able at least to beat the engine at 15+10 time control with knight odds(he could beat it with knight odds with 10+5 but not enough games to be sure if he can win in this time control).
2)I believe that the difference between rooks odd and knight odds is more than 100 elo.
3)Liang has rating of less than 2700 so rating of less than 2600 should be enough to win at rapid with rook odds.

Maybe there is a big difference between different players when there may be players with difference of 400 elo who have the same results with rook odds against leela so some 2600 player lose against the rook odds bot at 15+10 but some 2300 player can win against the bot.

Yes, I follow your reasoning and it makes sense, but the actual data shows performance in slow Rapid (base+inc = 20 to 25) not far from 3000, way above FIDE 2600 equivalent. I suppose it is mostly the difference between having played hundreds of practice games and studied the openings, as Awonder did, vs. just playing with little or no preparation. The other big lesson from all of this is that time is much more important than most of us thought. You can generally win in slow Rapid at queen odds with a FIDE rating in the low 1900s, whereas today a player "Derrick3110" with a 2882 bullet rating and a 2830 blitz rating (typical of strong GMs) played 12 games with it at 1'1", getting just a single draw with eleven losses!! I would have thought that a 2830 Lichess blitz player would play better at 1'1" than you do at 15'10", but it seems I would be very wrong, at least if queen odds play is any indication of play without odds. It seems that the time adjustment used in the leaderboard is too mild, time is absolutely critical!

[Uri Blass]

Probably IM can score around 50% at rook odds in 15+10, but most of the games are blitz where the human score like 10% or so

In your next email, you show:
3)FM veni_vidi_vici2007(2682?) 10+10 20-2 and 1 draw when most of the games were 10+10 or 15+10 and lost also 3.5-0.5 with 15+10.

Although he is not an IM, only FM, all of his ratings are above average for an IM, and yet he lost by the above overwhelming margins. So I think it is clear that a typical IM has no chance in a Rapid rook odds match. Of course there may be 1 or 2 IMs in the world who are of GM strength and very good in Rapid play who might have a chance, but clearly GM level, if not the title, is required. I know of one GM who has a minus record in Rapid at Rook and Move odds after many games (i.e. he plays White), I don't mean myself, but a player who still plays at GM level. Furthermore the bot was upgraded Dec. 11 to use the new net trained for both knight odds and rook odds, used in the Awonder match (and the Giri games). Your compilation shows that the performance of the Rook odds bot in Rapid is so high that I believe only a 2600+ FIDE player (or someone of similar strength in Rapid) could win a match at 15'10" at rook odds against the current bot. Given that it seems to require something approaching CM level to win a Rapid match at queen odds, this makes sense to me.

I have some problem to believe that you need 2600 to beat the bot with rook odds with 15+10 time control and I will explain my reasons:
1)When I consider the result of liang it seems that liang should be able at least to beat the engine at 15+10 time control with knight odds(he could beat it with knight odds with 10+5 but not enough games to be sure if he can win in this time control).
2)I believe that the difference between rooks odd and knight odds is more than 100 elo.
3)Liang has rating of less than 2700 so rating of less than 2600 should be enough to win at rapid with rook odds.

Maybe there is a big difference between different players when there may be players with difference of 400 elo who have the same results with rook odds against leela so some 2600 player lose against the rook odds bot at 15+10 but some 2300 player can win against the bot.

Yes, I follow your reasoning and it makes sense, but the actual data shows performance in slow Rapid (base+inc = 20 to 25) not far from 3000, way above FIDE 2600 equivalent. I suppose it is mostly the difference between having played hundreds of practice games and studied the openings, as Awonder did, vs. just playing with little or no preparation. The other big lesson from all of this is that time is much more important than most of us thought. You can generally win in slow Rapid at queen odds with a FIDE rating in the low 1900s, whereas today a player "Derrick3110" with a 2882 bullet rating and a 2830 blitz rating (typical of strong GMs) played 12 games with it at 1'1", getting just a single draw with eleven losses!! I would have thought that a 2830 Lichess blitz player would play better at 1'1" than you do at 15'10", but it seems I would be very wrong, at least if queen odds play is any indication of play without odds. It seems that the time adjustment used in the leaderboard is too mild, time is absolutely critical!

Looking at the games of Derrick I see strange moves and it seems that he does not understand basic principles.

Examples:

game 1:
https://lichess.org/3AIsysyo/black

I do not understand why 3...Nf6 and not 3...d5
I do not understand 9...Ke7

game 2:
https://lichess.org/lFNpyMIL/black

I do not understand 6...Be7 that does not develop pieces that i even have no reason to consider.
Black want to trade pieces and not to avoid trades.

game 3:
https://lichess.org/9rGhpgfn/black

I do not understand 1...e6 and not e5 or d5 and I do not understand the combination of 2..d5 and 2..b6 that seems to be passive.

It seems that the player does not follow basic principles in order to get a good position so that he can later force trades.

[Father]

[pgn][/pgn]

Giving up the rook will also get you checkmated eventually with perfect play, so why is it a better move than ...Kg7?

Well, Re8, I think, is better because ...Kg7 also leads to a quick defeat with perfect play, but at least it doesn’t give White the chance to get mated immediately. So the question is: why Kg7 is a better move than ...Re8?

Alex

... Mr. Larry Kaufman ... If you ask me, describe with one word a signal that marks the algorithmic profile of the LeelaRookOdds machine, I summarize it like this: "Ambitious."

[pgn][Event "Casual blitz game"]
[Site "https://lichess.org/xXh3Ktht"]
[Date "2024.12.16"]
[White "Catecan"]
[Black "LeelaRookOdds"]
[Result "1/2-1/2"]
[UTCDate "2024.12.16"]
[UTCTime "21:34:19"]
[WhiteElo "2082"]
[BlackElo "2000"]
[BlackTitle "BOT"]
[Variant "From Position"]
[TimeControl "180+2"]
[ECO "?"]
[Opening "?"]
[Termination "Normal"]
[FEN "1nbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQk - 0 1"]
[SetUp "1"]
[Annotator "lichess.org"]

1. d4 d6 2. e3 f5 3. f4 e6 4. Nf3 h6 5. c3 g5 6. Bd3 g4 7. Nfd2 d5 8. O-O h5 9. Re1 h4 10. Nf1 Nf6 11. Qe2 h3 12. g3 a5 13. a3 Kf7 14. b4 Kg6 15. Bb2 b6 16. Nbd2 Nbd7 17. Rec1 Bb7 18. Ba6 Ba8 19. Bd3 c5 20. Rab1 c4 21. Bc2 axb4 22. axb4 b5 23. Ra1 Bb7 24. Ra2 Qc7 25. Rca1 Kh5 26. Qe1 Ne4 27. Nxe4 dxe4 28. Bd1 Nf6 29. Be2 Nd5 30. Qd2 Qc6 31. Bc1 Bd6 32. Qe1 Bc7 33. Bd2 Bb6 34. Bd1 Kg6 35. Be2 e5 36. fxe5 Nc7 37. Qf2 Rd8 38. Be1 Nd5 39. Bd2 Rd7 40. Be1 Bd8 41. Bd2 Nc7 42. Be1 Ne6 43. Bd2 Rf7 44. Be1 Kg7 45. Bd2 Bg5 46. Bc1 Nc7 47. Bd2 Bd8 48. Be1 Ne6 49. Bd2 Qd5 50. Bc1 Rf8 51. Bd2 Rf7 52. Bc1 Rf8 53. Bd2 Ng5 54. Bc1 Qc6 55. Bd2 Rf7 56. Bc1 Rf8 57. Bd2 Bc7 58. Bc1 Bb6 59. Bd2 Rf7 60. Bc1 Bd8 61. Bd2 Kg6 62. Bc1 Bb6 63. Bd2 Qd5 64. Bc1 Bd8 65. Bd2 Qc6 66. Bc1 Kh7 67. Bd2 Qd5 68. Bc1 Qe6 69. Bd2 Bb6 70. Bc1 Bd8 71. Bd2 Qd5 72. Bc1 Kg7 73. Bd2 Qe6 74. Bc1 Qd5 75. Bd2 Qc6 76. Bc1 Kg6 77. Bd2 Bb6 78. Bc1 Bd8 79. Bd2 Kh7 80. Bc1 Bc7 81. Bd2 Kg8 82. Bc1 Bd8 83. Bd2 Kh7 84. Bc1 Bc7 85. Bd2 Kg8 86. Bc1 { The game is a draw. } 1/2-1/2[/pgn]



Because finding the sacrificial checkmate combination, with no hint that there is one, is more difficult than winning the game if Black just gives up a rook. To put it another way, giving up the rook is like resigning (unless there is a hope of flagging the opponent), whereas evan a GM might very well miss the brilliant mate in a practical game. If two moves are equally good against perfect play, the "better" move (as generally understood) is the one that maximizes your expected result (win or draw). It may depend on the opponent, but in this case it's pretty clear which is the best practical chance.

[lkaufman]

Looking at the games of Derrick I see strange moves and it seems that he does not understand basic principles.

Examples:

game 1:
https://lichess.org/3AIsysyo/black

I do not understand why 3...Nf6 and not 3...d5
I do not understand 9...Ke7

game 2:
https://lichess.org/lFNpyMIL/black

I do not understand 6...Be7 that does not develop pieces that i even have no reason to consider.
Black want to trade pieces and not to avoid trades.

game 3:
https://lichess.org/9rGhpgfn/black

I do not understand 1...e6 and not e5 or d5 and I do not understand the combination of 2..d5 and 2..b6 that seems to be passive.

It seems that the player does not follow basic principles in order to get a good position so that he can later force trades.

In game 1 3...Nf6 seems like a pre-move or instant bullet move; 9...Ke7 was surely a mouse-slip. In game 2 6...Be7 probably intended to trade with ...Bg5 but changed his mind. In game 3 the moves are not the ones I would make but not bad, probably it's just his normal opening in even game chess with ...b6 motivated by White's g4. Most likely a young tactical player with less refined judgment, or just playing super-fast due to bullet tc. Other players with such ratings have done better at 1'1".

[Father]

Looking at the games of Derrick I see strange moves and it seems that he does not understand basic principles.

Examples:

game 1:
https://lichess.org/3AIsysyo/black

I do not understand why 3...Nf6 and not 3...d5
I do not understand 9...Ke7

game 2:
https://lichess.org/lFNpyMIL/black

I do not understand 6...Be7 that does not develop pieces that i even have no reason to consider.
Black want to trade pieces and not to avoid trades.

game 3:
https://lichess.org/9rGhpgfn/black

I do not understand 1...e6 and not e5 or d5 and I do not understand the combination of 2..d5 and 2..b6 that seems to be passive.

It seems that the player does not follow basic principles in order to get a good position so that he can later force trades.

In game 1 3...Nf6 seems like a pre-move or instant bullet move; 9...Ke7 was surely a mouse-slip. In game 2 6...Be7 probably intended to trade with ...Bg5 but changed his mind. In game 3 the moves are not the ones I would make but not bad, probably it's just his normal opening in even game chess with ...b6 motivated by White's g4. Most likely a young tactical player with less refined judgment, or just playing super-fast due to bullet tc. Other players with such ratings have done better at 1'1".

… Good evening to the Leela team, good evening Mr. Larry Kaufman. Just as sleep and fatigue began to take over, I thought about the leaderboard of the top 100 human scores facing LeelaQueenOdds. And I thought about the humans who would like to be part of the leaderboard but who can't be there. And then I said to myself: Wouldn't it be more fair if instead of the top 100 there were, for example, the top 1000?