lkaufman wrote: ↑Mon May 05, 2025 8:01 pm Hikaru Nakamura played 16 games with LeelaKnightOdds at 3'+2", winning the first game, drawing the last one and one other, and losing 13, including ten in a row! His score of 12.5% was actually much better than other top GMs against the same version; players in the World top 50 (FIDE) scored only 4% in 80 games before this, without a single win! He played only g1 odds, not b1. He only intended to play 3 games, but after losing games 2 and 3 wanted to keep playing until he got a second win, which never happened. I think he could do somewhat better with a bit of opening preparation and less talking, but he regularly wins Titled Tuesday while talking so it doesn't seem to hurt his play too much. I hope he will either try playing it in Rapid or else switch the handicap to Rook odds in blitz, which should be pretty competitive.
Hi Larry,
Without any doubt, Leela's performance against Nakamura is truly impressive! It's amazing how it manages to complicate the games right from the opening, avoiding exchanges and using every tactical trick to its advantage. The games were very entertaining. Just a few years ago, it would've been hard to imagine this level of performance against one of the strongest blitz players in the world.
I can't even imagine a net to do the same and trained specifically to beat humans with all 32 pieces on the board—it would be absolutely impossible to even draw a single game!!!
By the way, do you know if Leela’s Queen-odds network is available for download? I don’t have a very powerful GPU, but I assume it’s not a huge net and might be able to run on a normal GPU, or even on a single CPU. Do you know anything about that?
Also, is there any chance—however small—of seeing a new version of Komodo in the future?
lkaufman wrote: ↑Mon May 05, 2025 8:01 pm Hikaru Nakamura played 16 games with LeelaKnightOdds at 3'+2", winning the first game, drawing the last one and one other, and losing 13, including ten in a row! His score of 12.5% was actually much better than other top GMs against the same version; players in the World top 50 (FIDE) scored only 4% in 80 games before this, without a single win! He played only g1 odds, not b1. He only intended to play 3 games, but after losing games 2 and 3 wanted to keep playing until he got a second win, which never happened. I think he could do somewhat better with a bit of opening preparation and less talking, but he regularly wins Titled Tuesday while talking so it doesn't seem to hurt his play too much. I hope he will either try playing it in Rapid or else switch the handicap to Rook odds in blitz, which should be pretty competitive.
Hi Larry,
Without any doubt, Leela's performance against Nakamura is truly impressive! It's amazing how it manages to complicate the games right from the opening, avoiding exchanges and using every tactical trick to its advantage. The games were very entertaining. Just a few years ago, it would've been hard to imagine this level of performance against one of the strongest blitz players in the world.
I can't even imagine a net to do the same and trained specifically to beat humans with all 32 pieces on the board—it would be absolutely impossible to even draw a single game!!!
By the way, do you know if Leela’s Queen-odds network is available for download? I don’t have a very powerful GPU, but I assume it’s not a huge net and might be able to run on a normal GPU, or even on a single CPU. Do you know anything about that?
Also, is there any chance—however small—of seeing a new version of Komodo in the future?
Best wishes.
The LeelaQueenOdds net is available for download, but the KnightOdds/RookOdds net is not. They would both run fine on a normal GPU, just would take longer to reach the recommended node figure. A new version of Komodo is unlikely, unless perhaps some discovery is made by another engine that would add many Elo points to Komodo with minimal work. Chess.com could choose to resume work on Komodo, but I have no expectation that they will do so.
I'm pretty sure that the knight odds net as it is (with appropriate parameter settings) would win every game against any human in standard chess, unless they played many thousands of games in which case it might give up a draw.
lkaufman wrote: ↑Mon May 05, 2025 8:01 pm Hikaru Nakamura played 16 games with LeelaKnightOdds at 3'+2", winning the first game, drawing the last one and one other, and losing 13, including ten in a row! His score of 12.5% was actually much better than other top GMs against the same version; players in the World top 50 (FIDE) scored only 4% in 80 games before this, without a single win! He played only g1 odds, not b1. He only intended to play 3 games, but after losing games 2 and 3 wanted to keep playing until he got a second win, which never happened. I think he could do somewhat better with a bit of opening preparation and less talking, but he regularly wins Titled Tuesday while talking so it doesn't seem to hurt his play too much. I hope he will either try playing it in Rapid or else switch the handicap to Rook odds in blitz, which should be pretty competitive.
At 3 minute blitz, hardly surprising. Longer time controls would be different
Well, I would call 3'2" the rough equivalent of 5 minute blitz, not 3, but I agree with you. I suggested in my comments that 10' +5" would be about balanced for Hikaru at knight odds. For 3'2" blitz, rook odds is appropriate for Hikaru.
Is it possible that Hikaru (or anyone) is something like 338 Elo points stronger at 10'+5" than 3'+2"? That sounds like way too big a difference.
lkaufman wrote: ↑Mon May 05, 2025 8:01 pm Hikaru Nakamura played 16 games with LeelaKnightOdds at 3'+2", winning the first game, drawing the last one and one other, and losing 13, including ten in a row! His score of 12.5% was actually much better than other top GMs against the same version; players in the World top 50 (FIDE) scored only 4% in 80 games before this, without a single win! He played only g1 odds, not b1. He only intended to play 3 games, but after losing games 2 and 3 wanted to keep playing until he got a second win, which never happened. I think he could do somewhat better with a bit of opening preparation and less talking, but he regularly wins Titled Tuesday while talking so it doesn't seem to hurt his play too much. I hope he will either try playing it in Rapid or else switch the handicap to Rook odds in blitz, which should be pretty competitive.
At 3 minute blitz, hardly surprising. Longer time controls would be different
Well, I would call 3'2" the rough equivalent of 5 minute blitz, not 3, but I agree with you. I suggested in my comments that 10' +5" would be about balanced for Hikaru at knight odds. For 3'2" blitz, rook odds is appropriate for Hikaru.
Is it possible that Hikaru (or anyone) is something like 338 Elo points stronger at 10'+5" than 3'+2"? That sounds like way too big a difference.
Good point, I would estimate the elo gain for the tripling of time as roughly 200, but I think that the score is a bit exaggerated due to Hikaru resigning some games prematurely presumably because once he was way behind his only goal was to win a second game, so spending several minutes fighting for a draw was unappealing. If we assume that the score might have been 13 to 3 rather than 14 to 2 if he didn't care how long the games took and only wanted to maximize his score in a sixteen game match, then perhaps a match at 10'5" would at least be close.
Hikaru Nakamura played 16 games with LeelaKnightOdds at 3'+2", winning the first game, drawing the last one and one other, and losing 13, including ten in a row! His score of 12.5% was actually much better than other top GMs against the same version; players in the World top 50 (FIDE) scored only 4% in 80 games before this, without a single win! He played only g1 odds, not b1. He only intended to play 3 games, but after losing games 2 and 3 wanted to keep playing until he got a second win, which never happened. I think he could do somewhat better with a bit of opening preparation and less talking, but he regularly wins Titled Tuesday while talking so it doesn't seem to hurt his play too much. I hope he will either try playing it in Rapid or else switch the handicap to Rook odds in blitz, which should be pretty competitive.
This video is a valuable historical document: computers have been beating humans at fast time controls for decades, but here we are shown the thoughts of a top human who is expecting to win, plays what he thinks is a series of 5 rapid games for a shortish video, and wanting to end with a win. He keeps on thinking he's messed up, so plays "one more game" not realising that he's actually being outplayed by a wide enough margin to lose even at knight odds!
Human chess is partly about tactics and strategy, but mostly about memory
Hikaru Nakamura played 16 games with LeelaKnightOdds at 3'+2", winning the first game, drawing the last one and one other, and losing 13, including ten in a row! His score of 12.5% was actually much better than other top GMs against the same version; players in the World top 50 (FIDE) scored only 4% in 80 games before this, without a single win! He played only g1 odds, not b1. He only intended to play 3 games, but after losing games 2 and 3 wanted to keep playing until he got a second win, which never happened. I think he could do somewhat better with a bit of opening preparation and less talking, but he regularly wins Titled Tuesday while talking so it doesn't seem to hurt his play too much. I hope he will either try playing it in Rapid or else switch the handicap to Rook odds in blitz, which should be pretty competitive.
This video is a valuable historical document: computers have been beating humans at fast time controls for decades, but here we are shown the thoughts of a top human who is expecting to win, plays what he thinks is a series of 5 rapid games for a shortish video, and wanting to end with a win. He keeps on thinking he's messed up, so plays "one more game" not realising that he's actually being outplayed by a wide enough margin to lose even at knight odds!
I think he realized well before the halfway mark that he was a big underdog, but since he did win one game he felt he should be able to win one more. One strong GM (David Paravyan aka "Drop_Stone") played way over a hundred blitz games before finally winning one, eventually winning a few more out of hundreds. A strong GM will eventually win a game, but most don't play enough to get a single win. He has said that he still thinks he would win handily at 3'2" with rook odds, but based on results against other top GMs I think it would be a tossup.
Hikaru Nakamura played 16 games with LeelaKnightOdds at 3'+2", winning the first game, drawing the last one and one other, and losing 13, including ten in a row! His score of 12.5% was actually much better than other top GMs against the same version; players in the World top 50 (FIDE) scored only 4% in 80 games before this, without a single win! He played only g1 odds, not b1. He only intended to play 3 games, but after losing games 2 and 3 wanted to keep playing until he got a second win, which never happened. I think he could do somewhat better with a bit of opening preparation and less talking, but he regularly wins Titled Tuesday while talking so it doesn't seem to hurt his play too much. I hope he will either try playing it in Rapid or else switch the handicap to Rook odds in blitz, which should be pretty competitive.
This video is a valuable historical document: computers have been beating humans at fast time controls for decades, but here we are shown the thoughts of a top human who is expecting to win, plays what he thinks is a series of 5 rapid games for a shortish video, and wanting to end with a win. He keeps on thinking he's messed up, so plays "one more game" not realising that he's actually being outplayed by a wide enough margin to lose even at knight odds!
I think he realized well before the halfway mark that he was a big underdog, but since he did win one game he felt he should be able to win one more. One strong GM (David Paravyan aka "Drop_Stone") played way over a hundred blitz games before finally winning one, eventually winning a few more out of hundreds. A strong GM will eventually win a game, but most don't play enough to get a single win. He has said that he still thinks he would win handily at 3'2" with rook odds, but based on results against other top GMs I think it would be a tossup.
My wooden horse "Catecan" and I were riding along the path of life! Today we stopped for a moment at a roadside inn.
Catecan drank water from the trough and ate some grass with honey. In the meantime, I drank some panela water. It was then that I was able to see the first 10 riders on the list of 100, and Catecan and I were very happy to see the great Hissha resuming his giant stride. We can barely see them from afar; we will try to catch up.
I am thinking chess is in a coin.Human beings for ever playing in one face.Now I am playing in the other face:"Antichess". Computers are as a fortres where owner forgot to close a little door behind. You must enter across this door.Forget the front.
Hikaru Nakamura played 16 games with LeelaKnightOdds at 3'+2", winning the first game, drawing the last one and one other, and losing 13, including ten in a row! His score of 12.5% was actually much better than other top GMs against the same version; players in the World top 50 (FIDE) scored only 4% in 80 games before this, without a single win! He played only g1 odds, not b1. He only intended to play 3 games, but after losing games 2 and 3 wanted to keep playing until he got a second win, which never happened. I think he could do somewhat better with a bit of opening preparation and less talking, but he regularly wins Titled Tuesday while talking so it doesn't seem to hurt his play too much. I hope he will either try playing it in Rapid or else switch the handicap to Rook odds in blitz, which should be pretty competitive.
This video is a valuable historical document: computers have been beating humans at fast time controls for decades, but here we are shown the thoughts of a top human who is expecting to win, plays what he thinks is a series of 5 rapid games for a shortish video, and wanting to end with a win. He keeps on thinking he's messed up, so plays "one more game" not realising that he's actually being outplayed by a wide enough margin to lose even at knight odds!
I think he realized well before the halfway mark that he was a big underdog, but since he did win one game he felt he should be able to win one more. One strong GM (David Paravyan aka "Drop_Stone") played way over a hundred blitz games before finally winning one, eventually winning a few more out of hundreds. A strong GM will eventually win a game, but most don't play enough to get a single win. He has said that he still thinks he would win handily at 3'2" with rook odds, but based on results against other top GMs I think it would be a tossup.
My wooden horse "Catecan" and I were riding along the path of life! Today we stopped for a moment at a roadside inn.
Catecan drank water from the trough and ate some grass with honey. In the meantime, I drank some panela water. It was then that I was able to see the first 10 riders on the list of 100, and Catecan and I were very happy to see the great Hissha resuming his giant stride. We can barely see them from afar; we will try to catch up.
I am thinking chess is in a coin.Human beings for ever playing in one face.Now I am playing in the other face:"Antichess". Computers are as a fortres where owner forgot to close a little door behind. You must enter across this door.Forget the front.
Hikaru Nakamura played 16 games with LeelaKnightOdds at 3'+2", winning the first game, drawing the last one and one other, and losing 13, including ten in a row! His score of 12.5% was actually much better than other top GMs against the same version; players in the World top 50 (FIDE) scored only 4% in 80 games before this, without a single win! He played only g1 odds, not b1. He only intended to play 3 games, but after losing games 2 and 3 wanted to keep playing until he got a second win, which never happened. I think he could do somewhat better with a bit of opening preparation and less talking, but he regularly wins Titled Tuesday while talking so it doesn't seem to hurt his play too much. I hope he will either try playing it in Rapid or else switch the handicap to Rook odds in blitz, which should be pretty competitive.
This video is a valuable historical document: computers have been beating humans at fast time controls for decades, but here we are shown the thoughts of a top human who is expecting to win, plays what he thinks is a series of 5 rapid games for a shortish video, and wanting to end with a win. He keeps on thinking he's messed up, so plays "one more game" not realising that he's actually being outplayed by a wide enough margin to lose even at knight odds!
I think he realized well before the halfway mark that he was a big underdog, but since he did win one game he felt he should be able to win one more. One strong GM (David Paravyan aka "Drop_Stone") played way over a hundred blitz games before finally winning one, eventually winning a few more out of hundreds. A strong GM will eventually win a game, but most don't play enough to get a single win. He has said that he still thinks he would win handily at 3'2" with rook odds, but based on results against other top GMs I think it would be a tossup.
My wooden horse "Catecan" and I were riding along the path of life! Today we stopped for a moment at a roadside inn.
Catecan drank water from the trough and ate some grass with honey. In the meantime, I drank some panela water. It was then that I was able to see the first 10 riders on the list of 100, and Catecan and I were very happy to see the great Hissha resuming his giant stride. We can barely see them from afar; we will try to catch up.
Good evening, Mr. Larry Kaufman. As I'm close to bed and tired today, I can see that LeelaQueenOdds is still very popular. I hope the robot remains very popular. Do you think the board will one day be preceded by one of the three best bullet chess players in the world? Of course, the current leaders are also excellent human players and extraordinary combatants against the robot. By the way, I just noticed that Hissha reached 2707 in Elo! Congratulations!
I am thinking chess is in a coin.Human beings for ever playing in one face.Now I am playing in the other face:"Antichess". Computers are as a fortres where owner forgot to close a little door behind. You must enter across this door.Forget the front.
