GM Joel Benjamin vs Leela Knight Odds Classical match

[lkaufman]

A match between GM Joel Benjamin and LeelaKnightOdds at a time limit in the standard or classical range has been scheduled for Jan. 25 thru 27. Eight games at one hour plus 30 seconds per move for Joel, with Leela playing bullet speed, three games per day for two days, two the final day. This is the first time that any engine has attempted to give knight odds to a GM (or even IM) at a time limit longer than Rapid. Joel is a three time U.S. Champion, and although now 60 years old, is still active, still rated near the FIDE threshold for GM (he is 2485, 2500 needed to earn the title initially), and recently finished third in the over-50 U.S. Championship. Furthermore he has more experience playing serious matches with engines than anyone, starting with Deep Blue around 1996-1997, Rybka in 2007 (pawn odds 8 games, draw and move odds 8 games), and Komodo around 2016 at various odds. This should be the ultimate test of whether an engine can really give knight odds to a GM with adequate time to think. Hard to predict the result, in Rapid Leela would be expected to win, but at this time limit, who knows?

[Father]

A match between GM Joel Benjamin and LeelaKnightOdds at a time limit in the standard or classical range has been scheduled for Jan. 25 thru 27. Eight games at one hour plus 30 seconds per move for Joel, with Leela playing bullet speed, three games per day for two days, two the final day. This is the first time that any engine has attempted to give knight odds to a GM (or even IM) at a time limit longer than Rapid. Joel is a three time U.S. Champion, and although now 60 years old, is still active, still rated near the FIDE threshold for GM (he is 2485, 2500 needed to earn the title initially), and recently finished third in the over-50 U.S. Championship. Furthermore he has more experience playing serious matches with engines than anyone, starting with Deep Blue around 1996-1997, Rybka in 2007 (pawn odds 8 games, draw and move odds 8 games), and Komodo around 2016 at various odds. This should be the ultimate test of whether an engine can really give knight odds to a GM with adequate time to think. Hard to predict the result, in Rapid Leela would be expected to win, but at this time limit, who knows?

Good evening, Mr. Larry Kaufman. Extraordinary news from the world of chess, science, the human brain and Artificial Intelligence. I wish you all, GM Benjamin Joel, the organizers and collaborators of the meeting, and of course the human team that makes Leela and Odds and Bots possible, the best winds in this year 2025 that is about to begin on its way. "Walker, there is no path, the path is made by walking, and when you look ahead, you see with faith and hope the wonderful paths that we will have to walk."

[Uri Blass]

A match between GM Joel Benjamin and LeelaKnightOdds at a time limit in the standard or classical range has been scheduled for Jan. 25 thru 27. Eight games at one hour plus 30 seconds per move for Joel, with Leela playing bullet speed, three games per day for two days, two the final day. This is the first time that any engine has attempted to give knight odds to a GM (or even IM) at a time limit longer than Rapid. Joel is a three time U.S. Champion, and although now 60 years old, is still active, still rated near the FIDE threshold for GM (he is 2485, 2500 needed to earn the title initially), and recently finished third in the over-50 U.S. Championship. Furthermore he has more experience playing serious matches with engines than anyone, starting with Deep Blue around 1996-1997, Rybka in 2007 (pawn odds 8 games, draw and move odds 8 games), and Komodo around 2016 at various odds. This should be the ultimate test of whether an engine can really give knight odds to a GM with adequate time to think. Hard to predict the result, in Rapid Leela would be expected to win, but at this time limit, who knows?

60+30 is slightly faster than the normal time control that I play OTB that is 90+30 so I do not think that is exactly classical time control.

https://www.chess.com/club/the-classical-time-control

In important events people use even slower time control.
90 minutes for the first 40 moves, followed by 30 minutes for the rest of the game, with an addition of 30 seconds per move starting from move one.

[Uri Blass]

A match between GM Joel Benjamin and LeelaKnightOdds at a time limit in the standard or classical range has been scheduled for Jan. 25 thru 27. Eight games at one hour plus 30 seconds per move for Joel, with Leela playing bullet speed, three games per day for two days, two the final day. This is the first time that any engine has attempted to give knight odds to a GM (or even IM) at a time limit longer than Rapid. Joel is a three time U.S. Champion, and although now 60 years old, is still active, still rated near the FIDE threshold for GM (he is 2485, 2500 needed to earn the title initially), and recently finished third in the over-50 U.S. Championship. Furthermore he has more experience playing serious matches with engines than anyone, starting with Deep Blue around 1996-1997, Rybka in 2007 (pawn odds 8 games, draw and move odds 8 games), and Komodo around 2016 at various odds. This should be the ultimate test of whether an engine can really give knight odds to a GM with adequate time to think. Hard to predict the result, in Rapid Leela would be expected to win, but at this time limit, who knows?

I checked his rating and his fide rating is 2473 and not 2485 when the facts are that his rating went down recently.
https://ratings.fide.com/profile/2000091/calculations

[lkaufman]

A match between GM Joel Benjamin and LeelaKnightOdds at a time limit in the standard or classical range has been scheduled for Jan. 25 thru 27. Eight games at one hour plus 30 seconds per move for Joel, with Leela playing bullet speed, three games per day for two days, two the final day. This is the first time that any engine has attempted to give knight odds to a GM (or even IM) at a time limit longer than Rapid. Joel is a three time U.S. Champion, and although now 60 years old, is still active, still rated near the FIDE threshold for GM (he is 2485, 2500 needed to earn the title initially), and recently finished third in the over-50 U.S. Championship. Furthermore he has more experience playing serious matches with engines than anyone, starting with Deep Blue around 1996-1997, Rybka in 2007 (pawn odds 8 games, draw and move odds 8 games), and Komodo around 2016 at various odds. This should be the ultimate test of whether an engine can really give knight odds to a GM with adequate time to think. Hard to predict the result, in Rapid Leela would be expected to win, but at this time limit, who knows?

60+30 is slightly faster than the normal time control that I play OTB that is 90+30 so I do not think that is exactly classical time control.

https://www.chess.com/club/the-classical-time-control

In important events people use even slower time control.
90 minutes for the first 40 moves, followed by 30 minutes for the rest of the game, with an addition of 30 seconds per move starting from move one.

The match was originally set for 90 minutes plus 30 seconds, a more standard FIDE time control, but Joel said that he didn't think he would do any worse with an hour instead of 90 minutes at knight odds since it's mostly about making sure nothing was overlooked rather than trying to find the "best" move, which has no objective meaning in a position where many moves are winning. This made it possible to complete the match in 3 days rather than 4, making scheduling easier. Perhaps "standard" is a better term for the time control than "classical". This time control was used recently in the Carlsen vs Caruana chess960 match.

Regarding FIDE rating, Joel was 2485 when the match was agreed. I didn't re-check it before posting the announcement. Although he is below the 2500 rating required to earn the GM title, I think that his vast experience playing the best engines at various odds makes him a tougher opponent than a randomly chosen 2500 FIDE player. He has already played a huge number of Rapid games with LeelaKnightOdds (and also LeelaRookOdds) as preparation for the match; clearly he wants to do as well as possible.

[lkaufman]

The match is all set to start this Saturday Jan 25, with Matthew Sadler streaming it with his commentary on his "Silicon Road" YouTube channel. Schedule is noon, 2:30 pm, and 5 pm EST on Saturday and Sunday, noon and 4 p.m. EST Monday. Alternating b1 and g1 knight odds, one hour plus 30" increment time limit but with Leela playing at bullet chess speed (1'1" or faster) and also playing up to a dozen games at once at various odds, all on a single GPU! Joel plans to use the account "SuperEasy" which he has used to play hundreds of games with the Leela Odds bots. He is in the number one spot on the "LeelaQueenOdds" Leaderboard with a 94% score (all at 10'3"), and at that time control has scored 43% at rook and move odds and 28% at rook odds. At knight odds at this time control his score stands at 62 losses, 14 draws, and just one win, but time pressure played a huge role in many games, so this isn't predictive of the result with about seven times as much time to think. In general, his results in these various odds games have been better than most players with similar FIDE ratings, and he certainly has played more Rapid odds games than any other GM (except perhaps me). So this should be a rather severe test of the question of whether Leela can successfully give knight odds to a player of current GM standard at a time control in the Classical/standard range.

[Uri Blass]

The match is all set to start this Saturday Jan 25, with Matthew Sadler streaming it with his commentary on his "Silicon Road" YouTube channel. Schedule is noon, 2:30 pm, and 5 pm EST on Saturday and Sunday, noon and 4 p.m. EST Monday. Alternating b1 and g1 knight odds, one hour plus 30" increment time limit but with Leela playing at bullet chess speed (1'1" or faster) and also playing up to a dozen games at once at various odds, all on a single GPU! Joel plans to use the account "SuperEasy" which he has used to play hundreds of games with the Leela Odds bots. He is in the number one spot on the "LeelaQueenOdds" Leaderboard with a 94% score (all at 10'3"), and at that time control has scored 43% at rook and move odds and 28% at rook odds. At knight odds at this time control his score stands at 62 losses, 14 draws, and just one win, but time pressure played a huge role in many games, so this isn't predictive of the result with about seven times as much time to think. In general, his results in these various odds games have been better than most players with similar FIDE ratings, and he certainly has played more Rapid odds games than any other GM (except perhaps me). So this should be a rather severe test of the question of whether Leela can successfully give knight odds to a player of current GM standard at a time control in the Classical/standard range.

I wonder if you did not find a way to take advantage of time to play slower than 1+1
Playing faster is of course better to give the opponent less time to think but I think there is no big difference in the ability of the opponent at 60+30 if Leela use twice more time and the question is if there is no way to improve Leela significantly by using more time.

One way may be simply to find wins in a winning position because Leela may miss a win in a winning position and draw the game.

I also believe that even if you go for traps in a losing position then more search time should be productive with correct search to find more ideas for good traps against the opponent.

[Uri Blass]

For wins of him against Leelaknight odds I do not see wins after 11.11.2024(search for dates 12.11.2024 and later found no wins).

I could see in 2025 34 losses and 7 draws that is slightly worse than 62 losses and 14 draws even if I ignore the win.

[lkaufman]

The match is all set to start this Saturday Jan 25, with Matthew Sadler streaming it with his commentary on his "Silicon Road" YouTube channel. Schedule is noon, 2:30 pm, and 5 pm EST on Saturday and Sunday, noon and 4 p.m. EST Monday. Alternating b1 and g1 knight odds, one hour plus 30" increment time limit but with Leela playing at bullet chess speed (1'1" or faster) and also playing up to a dozen games at once at various odds, all on a single GPU! Joel plans to use the account "SuperEasy" which he has used to play hundreds of games with the Leela Odds bots. He is in the number one spot on the "LeelaQueenOdds" Leaderboard with a 94% score (all at 10'3"), and at that time control has scored 43% at rook and move odds and 28% at rook odds. At knight odds at this time control his score stands at 62 losses, 14 draws, and just one win, but time pressure played a huge role in many games, so this isn't predictive of the result with about seven times as much time to think. In general, his results in these various odds games have been better than most players with similar FIDE ratings, and he certainly has played more Rapid odds games than any other GM (except perhaps me). So this should be a rather severe test of the question of whether Leela can successfully give knight odds to a player of current GM standard at a time control in the Classical/standard range.

I wonder if you did not find a way to take advantage of time to play slower than 1+1
Playing faster is of course better to give the opponent less time to think but I think there is no big difference in the ability of the opponent at 60+30 if Leela use twice more time and the question is if there is no way to improve Leela significantly by using more time.

One way may be simply to find wins in a winning position because Leela may miss a win in a winning position and draw the game.

I also believe that even if you go for traps in a losing position then more search time should be productive with correct search to find more ideas for good traps against the opponent.

These are good ideas, the programmers have similar ideas, they just haven't had time to implement them yet.

[lkaufman]

For wins of him against Leelaknight odds I do not see wins after 11.11.2024(search for dates 12.11.2024 and later found no wins).

I could see in 2025 34 losses and 7 draws that is slightly worse than 62 losses and 14 draws even if I ignore the win.

It seems that Joel played nine Rapid games exactly on November 11(which you excluded), including his one win, which took 129 moves! I believe that Nov.11 was the upgrade date for the queen odds net; the knight odds net was upgraded Nov.2 if I remember correctly. As for any slight decline in results for Joel since new year's, it is most likely because the opening book gets upgraded regularly for all the bots based on results, so even without a new net they gradually get stronger as bad lines are pruned.