Amazing results of LeelaKnightOdds-Dev

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, chrisw, Rebel

Father
Posts: 1510
Joined: Sun Mar 19, 2006 4:39 am
Location: Colombia
Full name: Pablo Ignacio Restrepo

Re: Reporte acerca de LeelaQueenOdds en control de tiempo 3'0 minutos tras 138 encuentros.

Post by Father »

Father wrote: Tue Oct 29, 2024 2:00 pm
lkaufman wrote: Sat Oct 26, 2024 12:52 am
Uri Blass wrote: Sat Oct 26, 2024 12:02 am
Father wrote: Fri Oct 25, 2024 4:25 am
Father wrote: Fri Oct 25, 2024 12:18 am
lkaufman wrote: Thu Oct 24, 2024 11:52 pm
Father wrote: Thu Oct 24, 2024 8:08 pm
Father wrote: Thu Oct 24, 2024 7:41 pm
lkaufman wrote: Thu Oct 24, 2024 6:16 pm
Father wrote: Thu Oct 24, 2024 3:14 pm
Father wrote: Thu Oct 24, 2024 6:29 am
lkaufman wrote: Mon Oct 07, 2024 5:01 pm
Father wrote: Mon Oct 07, 2024 4:47 pm
lkaufman wrote: Mon Oct 07, 2024 3:47 am
Father wrote: Mon Oct 07, 2024 1:57 am Mr. Larry Kaufman.
This performance in the fight of man against machine with odds is really wonderful.
I consider that in man-machine duels with odds, that in some way, there are real probabilities on the part of the human being to acquire progress in shortening the distances.
If at any time you wish or consider that it is useful for you to participate in man-versus-machine duels, I would be attentive to an invitation, in exchange for the sole pleasure of playing chess, without a prize of any kind.
Thank you. Your games have been helpful in highlighting the issue of allowing early repetition draws, we're working on ways to reduce them. If you want to know what I would consider most useful, I would like to see you play games with LeelaQueenForKnight, at whatever time limit allows you to win a decent percentage of the games. It's most helpful when players pick a handicap that is a bit difficult for them but where the win/loss ratio is not too one-sided. My guess is that you'll need a fairly slow blitz or even Rapid TC to do this; it's WAY harder than actual queen odds, though easier for the human than rook odds.
...excellent odds!! I have stated, ( 20 games ) I believe that as human beings playing against odds, we have the possibility of learning more about the odds and thus increase the probability of improving the results... I have a question; What could be the estimated elo of this odds? Thank you Mr. Larry Kaufman.
In LiChess blitz games that bot performs around 2400 Lichess level, which is considered to be equivalent to 2235 FIDE rating. Of course it would be lower at longer time controls, but you are playing blitz, so the above is appropriate for you.

1.733 / 5.000
Mr. Larry Kaufman. Just a few minutes ago, I finished my first match against "LeelaQueenOdds". I played ten matches, and the final result was the following: 6 points to the computer, 4 points to me. The computer won 5 matches against me, I won 3 matches against the computer, and there were two draws. I mistakenly believed that it would be a simple matter to beat the computer. But nothing could be further from the truth. I played the games with a 3-minute time control. The computer knocked me out four times in a row... I went into a deep shock. I was then forced to make a quick change of strategy, using all my experience. A round of only ten matches, and I was defeated four times. !! I asked myself what to do against this beast, who without legs and without one of its hands was annihilating me!! So, game number five was my first win, game number six a draw, game number seven my second win, game number eight a draw, game number nine represented the tragedy of my fifth loss, and game number ten represented my third win. If we divide the tournament into two stages, game number one to game number four shows a computer that takes me by surprise. And game number four to game ten, a modification of my flexible thinking, which gives a favorable result in my favor of four against two. But... I lost. Nothing to say, and instead, a lot to thank the creators of this ODDS for. Without a doubt, I recommend this odds. The reason is simple... it allows us to play chess, to improve ourselves, and it gives us the joy of playing against someone better than the human world champion, who gives us the opportunity without his queen, to play chess. Thank you.
...The sun that gives us light and life rises to the heavens. Today I have understood that we can be better, move forward, progress and evolve as human beings. Yesterday ending the day at night I lost 6-4. Today at the beginning of the day I beat the computer 7-3. I have tamed the furious horse, now it is time to ride the cybernetic horse. Our strategic thinking makes the conquering steed tame. Man and machine heading towards the north and ideals, crossing the seven seas. My Father and Catecan will conquer the world.

[pgn][Event "Casual blitz game"]
[Site "https://lichess.org/MtU1NIXV"]
[Date "2024.10.24"]
[White "LeelaQueenOdds"]
[Black "Catecan"]
[Result "0-1"]
[UTCDate "2024.10.24"]
[UTCTime "12:01:04"]
[WhiteElo "2000"]
[BlackElo "2082"]
[WhiteTitle "BOT"]
[Variant "From Position"]
[TimeControl "180+0"]
[ECO "?"]
[Opening "?"]
[Termination "Normal"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[SetUp "1"]
[Annotator "lichess.org"]

1. c4 e6 2. d3 d5 3. a3 c6 4. e4 dxc4 5. dxc4 e5 6. Nc3 Nf6 7. h3 a5 8. Be3 b6 9. Nf3 Nbd7 10. Rd1 Qc7 11. g4 h6 12. Rg1 Be7 13. Bd3 Ba6 14. b3 Bxa3 15. Ke2 Bc5 16. Bd2 Bb4 17. Ke3 Rd8 18. Bc2 Nc5 19. Ne2 Rxd2 20. Nxd2 Ke7 21. Nf3 Rd8 22. Rxd8 Qxd8 23. Rd1 Qc7 24. Bb1 g5 25. Ng3 Kf8 26. h4 Nxg4+ 27. Ke2 f6 28. hxg5 hxg5 29. Rh1 Kg8 30. Nh5 Qf7 31. Kf1 Ne6 32. Bd3 Nf4 33. Nxf4 exf4 34. e5 Nxe5 35. Nxe5 fxe5 36. Bh7+ Qxh7 37. Rxh7 Kxh7 38. Ke2 Kg6 39. Kd3 Kf6 40. Ke4 Bb7 41. c5 Bxc5 42. b4 axb4 43. Kd3 Ba6+ 44. Kc2 g4 45. Kb3 e4 46. Kc2 Bb5 47. Kb3 Ke5 48. Kc2 Bxf2 49. Kb3 Bc5 50. Kc2 e3 51. Kb2 f3 52. Kc2 g3 53. Kb2 Kf4 54. Kc2 Bd4 55. Kb3 c5 56. Ka2 Bc4+ 57. Kb1 b5 58. Kc2 b3+ 59. Kb1 e2 60. Kc1 Ke3 61. Kb1 f2 62. Kc1 g2 63. Kb1 e1=N 64. Kc1 f1=N 65. Kb1 g1=N 66. Kc1 b2+ 67. Kb1 Bd3+ 68. Ka2 b1=Q+ 69. Ka3 b4+ 70. Ka4 b3 71. Ka5 b2 72. Kb6 c4+ 73. Kc6 c3 74. Kd7 c2 75. Ke7 c1=N 76. Kd7 Qa1 77. Ke7 b1=N 78. Kf7 Qa7+ 79. Kf8 Bc5+ 80. Kg8 Bc4+ 81. Kh8 Bd4# { Black wins by checkmate. } 0-1[/pgn]

Note that right now, only the knight odds bot is specifically trained for odds, but it is likely that the other bots will be updated very soon to use the net trained for knight odds, as it seems to be stronger than the normal version even at larger odds. So once the switch is made, expect it to get even harder to win at queen odds!
... A question. Let's imagine an open tournament with about 200 participants. One of the 200 players is leelaOddsQueen. In the imaginary scenario, there are 5 grandmasters, 10 international masters, 30 top-class players, and 155 players of all kinds; The time control, 3 minutes per game, and 8 hours straight in the masnau. I wonder? Where in the table would OddsQueen probably be? Probable Elo? Thank you Mr. Larry Kaufman ...
Actually, that's a very easy question to answer. There are about 75 Lichess games at 3'0" at queen odds against players with Lichess ratings in the 2000 to 2200 range, with an almost exactly even score. You are the median player and very close to the mean at 2082. So therefore the expected Lichess blitz rating at queen odds should be very close to 2082. Now to convert that to a FIDE rating, the website https://chessgoals.com/rating-comparison/ is reliable and give 1916 FIDE as equivalent.
therefore "the queen" could mean or have a valley of 1200-1300 elo? ... This computer is certainly superior to the human world chess champion... therefore it seems that without a queen, it could statistically beat the world champion with relative ease... my score against this machine is 24.5 against 15.5 in my favor.
** Errata:
24.5
16.6
The average-strength human chess player seems to have a clear chance of success playing chess against the human world chess champion, with the champion having no queen.
- I clarify because the text was ambiguous.
In this case it means that the average human chess player is better than most people who play chess because I know that I can beat most people who know the rules without my queen and I believe that I can also beat most humans who know the chess rules without a queen and 2 rooks.

I have some experience in doing it against humans in a real board with no time limit and not only against children.

Of course adults who are too weak not to lose against me in these conditions do not play in chess tournaments but based on my experience in most cases it is not something easy to teach them to beat me.
I suppose he was talking about the average online player (1500 lichess blitz or 1200 chess.com blitz), although even then I think it is incorrect for 3'0" chess. For huge odds like queen odds, the result depends mostly on the elo of the receiver, so even Leela at say 3600 won't do a lot better than Magnus or Hikaru. Hikaru actually did a 3'0" queen odds run on chess.com and won the final game against a player rated slightly over 2000, although I think he recognized that it was a fluke and wouldn't expect to get a 2000 rating playing that way, maybe more like 1800 or so. But at longer time limits like Rapid it would be much lower, maybe 1500 Lichess or so, i.e. about the average. But for smaller handicaps like knight odds, the difference between what the engine can achieve and what Magnus or Hikaru can do is huge; in blitz it looks likely from the results that the LeelaKnightOdds bot can actually give Magnus or Hikaru knight odds at about 3'2" blitz and come out about even, though extrapolating from results against players around 2500 FIDE level is risky.
I would like to be useful to the LeelaQueenOdds programmers to share my own experience after 138 encounters:
1- Although each session of games against the computer is affected by my biorhythm, the levels of my fatigue and my emotions, I have observed °a changing personality in the nuances of the computer's thinking°
2- Sometimes the computer's thinking has been "on the downside", however I have come to consider that it has been part of the computer's plan, that of "making itself seem crazy" to deceive us, which is typical of historical leadership techniques. in the art of war and is called precisely this way: ° The theory of the madman °.
3- I have observed that for just over 24 hours, the profiles of the algorithm are growing in their power, to the point that I went from a session in which I considered in my first 10 games that I could have won ten to zero in my favor However, today's thirty-round session represented greater difficulty for me; I admit that I was more exhausted and sometimes distracted, but even so, I feel that the computer is not the same and that it is superior to that of other previous occasions in which it was very easy for me to overcome it.
4- The final result to date in my 138 matches is 83 in my favor against 55 for the computer, which represents a superiority in my level of strength compared to LeelaQueenOdds's 142 elo, however the scoreboard carries weights that At some point I have considered delimiting a difference of 420 elo in my favor. The doubt and question I have today is if the computer will really be increasing in power or if it will once again be easy prey for me.
6- In my opinion LeelaKnightOdds could be higher than 2000 Fide
..Correction. In my In my opinion LeelaQueenOdds - I wrote Knight- could be higher than 2000 Fide elo in these time controls, that is, having favorable final results in the total score the computer for Fide players of 2000 elo who participate in matches and combats in game levels of three minutes per game.
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.
Father
Posts: 1510
Joined: Sun Mar 19, 2006 4:39 am
Location: Colombia
Full name: Pablo Ignacio Restrepo

Re: Reporte acerca de LeelaQueenOdds en control de tiempo 3'0 minutos tras 138 encuentros.

Post by Father »

[pgn][pgn][/pgn][/pgn]
Father wrote: Tue Oct 29, 2024 2:15 pm
Father wrote: Tue Oct 29, 2024 2:00 pm
lkaufman wrote: Sat Oct 26, 2024 12:52 am
Uri Blass wrote: Sat Oct 26, 2024 12:02 am
Father wrote: Fri Oct 25, 2024 4:25 am
Father wrote: Fri Oct 25, 2024 12:18 am
lkaufman wrote: Thu Oct 24, 2024 11:52 pm
Father wrote: Thu Oct 24, 2024 8:08 pm
Father wrote: Thu Oct 24, 2024 7:41 pm
lkaufman wrote: Thu Oct 24, 2024 6:16 pm
Father wrote: Thu Oct 24, 2024 3:14 pm
Father wrote: Thu Oct 24, 2024 6:29 am
lkaufman wrote: Mon Oct 07, 2024 5:01 pm
Father wrote: Mon Oct 07, 2024 4:47 pm
lkaufman wrote: Mon Oct 07, 2024 3:47 am
Father wrote: Mon Oct 07, 2024 1:57 am Mr. Larry Kaufman.
This performance in the fight of man against machine with odds is really wonderful.
I consider that in man-machine duels with odds, that in some way, there are real probabilities on the part of the human being to acquire progress in shortening the distances.
If at any time you wish or consider that it is useful for you to participate in man-versus-machine duels, I would be attentive to an invitation, in exchange for the sole pleasure of playing chess, without a prize of any kind.
Thank you. Your games have been helpful in highlighting the issue of allowing early repetition draws, we're working on ways to reduce them. If you want to know what I would consider most useful, I would like to see you play games with LeelaQueenForKnight, at whatever time limit allows you to win a decent percentage of the games. It's most helpful when players pick a handicap that is a bit difficult for them but where the win/loss ratio is not too one-sided. My guess is that you'll need a fairly slow blitz or even Rapid TC to do this; it's WAY harder than actual queen odds, though easier for the human than rook odds.
...excellent odds!! I have stated, ( 20 games ) I believe that as human beings playing against odds, we have the possibility of learning more about the odds and thus increase the probability of improving the results... I have a question; What could be the estimated elo of this odds? Thank you Mr. Larry Kaufman.
In LiChess blitz games that bot performs around 2400 Lichess level, which is considered to be equivalent to 2235 FIDE rating. Of course it would be lower at longer time controls, but you are playing blitz, so the above is appropriate for you.

1.733 / 5.000
Mr. Larry Kaufman. Just a few minutes ago, I finished my first match against "LeelaQueenOdds". I played ten matches, and the final result was the following: 6 points to the computer, 4 points to me. The computer won 5 matches against me, I won 3 matches against the computer, and there were two draws. I mistakenly believed that it would be a simple matter to beat the computer. But nothing could be further from the truth. I played the games with a 3-minute time control. The computer knocked me out four times in a row... I went into a deep shock. I was then forced to make a quick change of strategy, using all my experience. A round of only ten matches, and I was defeated four times. !! I asked myself what to do against this beast, who without legs and without one of its hands was annihilating me!! So, game number five was my first win, game number six a draw, game number seven my second win, game number eight a draw, game number nine represented the tragedy of my fifth loss, and game number ten represented my third win. If we divide the tournament into two stages, game number one to game number four shows a computer that takes me by surprise. And game number four to game ten, a modification of my flexible thinking, which gives a favorable result in my favor of four against two. But... I lost. Nothing to say, and instead, a lot to thank the creators of this ODDS for. Without a doubt, I recommend this odds. The reason is simple... it allows us to play chess, to improve ourselves, and it gives us the joy of playing against someone better than the human world champion, who gives us the opportunity without his queen, to play chess. Thank you.
...The sun that gives us light and life rises to the heavens. Today I have understood that we can be better, move forward, progress and evolve as human beings. Yesterday ending the day at night I lost 6-4. Today at the beginning of the day I beat the computer 7-3. I have tamed the furious horse, now it is time to ride the cybernetic horse. Our strategic thinking makes the conquering steed tame. Man and machine heading towards the north and ideals, crossing the seven seas. My Father and Catecan will conquer the world.

[pgn][Event "Casual blitz game"]
[Site "https://lichess.org/MtU1NIXV"]
[Date "2024.10.24"]
[White "LeelaQueenOdds"]
[Black "Catecan"]
[Result "0-1"]
[UTCDate "2024.10.24"]
[UTCTime "12:01:04"]
[WhiteElo "2000"]
[BlackElo "2082"]
[WhiteTitle "BOT"]
[Variant "From Position"]
[TimeControl "180+0"]
[ECO "?"]
[Opening "?"]
[Termination "Normal"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[SetUp "1"]
[Annotator "lichess.org"]

1. c4 e6 2. d3 d5 3. a3 c6 4. e4 dxc4 5. dxc4 e5 6. Nc3 Nf6 7. h3 a5 8. Be3 b6 9. Nf3 Nbd7 10. Rd1 Qc7 11. g4 h6 12. Rg1 Be7 13. Bd3 Ba6 14. b3 Bxa3 15. Ke2 Bc5 16. Bd2 Bb4 17. Ke3 Rd8 18. Bc2 Nc5 19. Ne2 Rxd2 20. Nxd2 Ke7 21. Nf3 Rd8 22. Rxd8 Qxd8 23. Rd1 Qc7 24. Bb1 g5 25. Ng3 Kf8 26. h4 Nxg4+ 27. Ke2 f6 28. hxg5 hxg5 29. Rh1 Kg8 30. Nh5 Qf7 31. Kf1 Ne6 32. Bd3 Nf4 33. Nxf4 exf4 34. e5 Nxe5 35. Nxe5 fxe5 36. Bh7+ Qxh7 37. Rxh7 Kxh7 38. Ke2 Kg6 39. Kd3 Kf6 40. Ke4 Bb7 41. c5 Bxc5 42. b4 axb4 43. Kd3 Ba6+ 44. Kc2 g4 45. Kb3 e4 46. Kc2 Bb5 47. Kb3 Ke5 48. Kc2 Bxf2 49. Kb3 Bc5 50. Kc2 e3 51. Kb2 f3 52. Kc2 g3 53. Kb2 Kf4 54. Kc2 Bd4 55. Kb3 c5 56. Ka2 Bc4+ 57. Kb1 b5 58. Kc2 b3+ 59. Kb1 e2 60. Kc1 Ke3 61. Kb1 f2 62. Kc1 g2 63. Kb1 e1=N 64. Kc1 f1=N 65. Kb1 g1=N 66. Kc1 b2+ 67. Kb1 Bd3+ 68. Ka2 b1=Q+ 69. Ka3 b4+ 70. Ka4 b3 71. Ka5 b2 72. Kb6 c4+ 73. Kc6 c3 74. Kd7 c2 75. Ke7 c1=N 76. Kd7 Qa1 77. Ke7 b1=N 78. Kf7 Qa7+ 79. Kf8 Bc5+ 80. Kg8 Bc4+ 81. Kh8 Bd4# { Black wins by checkmate. } 0-1[/pgn]

Note that right now, only the knight odds bot is specifically trained for odds, but it is likely that the other bots will be updated very soon to use the net trained for knight odds, as it seems to be stronger than the normal version even at larger odds. So once the switch is made, expect it to get even harder to win at queen odds!
... A question. Let's imagine an open tournament with about 200 participants. One of the 200 players is leelaOddsQueen. In the imaginary scenario, there are 5 grandmasters, 10 international masters, 30 top-class players, and 155 players of all kinds; The time control, 3 minutes per game, and 8 hours straight in the masnau. I wonder? Where in the table would OddsQueen probably be? Probable Elo? Thank you Mr. Larry Kaufman ...
Actually, that's a very easy question to answer. There are about 75 Lichess games at 3'0" at queen odds against players with Lichess ratings in the 2000 to 2200 range, with an almost exactly even score. You are the median player and very close to the mean at 2082. So therefore the expected Lichess blitz rating at queen odds should be very close to 2082. Now to convert that to a FIDE rating, the website https://chessgoals.com/rating-comparison/ is reliable and give 1916 FIDE as equivalent.
therefore "the queen" could mean or have a valley of 1200-1300 elo? ... This computer is certainly superior to the human world chess champion... therefore it seems that without a queen, it could statistically beat the world champion with relative ease... my score against this machine is 24.5 against 15.5 in my favor.
** Errata:
24.5
16.6
The average-strength human chess player seems to have a clear chance of success playing chess against the human world chess champion, with the champion having no queen.
- I clarify because the text was ambiguous.
In this case it means that the average human chess player is better than most people who play chess because I know that I can beat most people who know the rules without my queen and I believe that I can also beat most humans who know the chess rules without a queen and 2 rooks.

I have some experience in doing it against humans in a real board with no time limit and not only against children.

Of course adults who are too weak not to lose against me in these conditions do not play in chess tournaments but based on my experience in most cases it is not something easy to teach them to beat me.
I suppose he was talking about the average online player (1500 lichess blitz or 1200 chess.com blitz), although even then I think it is incorrect for 3'0" chess. For huge odds like queen odds, the result depends mostly on the elo of the receiver, so even Leela at say 3600 won't do a lot better than Magnus or Hikaru. Hikaru actually did a 3'0" queen odds run on chess.com and won the final game against a player rated slightly over 2000, although I think he recognized that it was a fluke and wouldn't expect to get a 2000 rating playing that way, maybe more like 1800 or so. But at longer time limits like Rapid it would be much lower, maybe 1500 Lichess or so, i.e. about the average. But for smaller handicaps like knight odds, the difference between what the engine can achieve and what Magnus or Hikaru can do is huge; in blitz it looks likely from the results that the LeelaKnightOdds bot can actually give Magnus or Hikaru knight odds at about 3'2" blitz and come out about even, though extrapolating from results against players around 2500 FIDE level is risky.
I would like to be useful to the LeelaQueenOdds programmers to share my own experience after 138 encounters:
1- Although each session of games against the computer is affected by my biorhythm, the levels of my fatigue and my emotions, I have observed °a changing personality in the nuances of the computer's thinking°
2- Sometimes the computer's thinking has been "on the downside", however I have come to consider that it has been part of the computer's plan, that of "making itself seem crazy" to deceive us, which is typical of historical leadership techniques. in the art of war and is called precisely this way: ° The theory of the madman °.
3- I have observed that for just over 24 hours, the profiles of the algorithm are growing in their power, to the point that I went from a session in which I considered in my first 10 games that I could have won ten to zero in my favor However, today's thirty-round session represented greater difficulty for me; I admit that I was more exhausted and sometimes distracted, but even so, I feel that the computer is not the same and that it is superior to that of other previous occasions in which it was very easy for me to overcome it.
4- The final result to date in my 138 matches is 83 in my favor against 55 for the computer, which represents a superiority in my level of strength compared to LeelaQueenOdds's 142 elo, however the scoreboard carries weights that At some point I have considered delimiting a difference of 420 elo in my favor. The doubt and question I have today is if the computer will really be increasing in power or if it will once again be easy prey for me.
6- In my opinion LeelaKnightOdds could be higher than 2000 Fide
..Correction. In my In my opinion LeelaQueenOdds - I wrote Knight- could be higher than 2000 Fide elo in these time controls, that is, having favorable final results in the total score the computer for Fide players of 2000 elo who participate in matches and combats in game levels of three minutes per game.
[pgn] Event "Casual blitz game"]
[Site "https://lichess.org/xyBlQjn4"]
[Date "2024.10.29"]
[White "LeelaQueenOdds"]
[Black "Catecan"]
[Result "0-1"]
[UTCDate "2024.10.29"]
[UTCTime "18:51:54"]
[WhiteElo "2000"]
[BlackElo "2082"]
[WhiteTitle "BOT"]
[Variant "From Position"]
[TimeControl "180+0"]
[ECO "?"]
[Opening "?"]
[Termination "Normal"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[SetUp "1"]
[Annotator "lichess.org"]

1. d4 d5 2. Bf4 e6 3. e3 c5 4. Nf3 cxd4 5. c4 dxe3 6. Bxe3 Nc6 7. cxd5 exd5 8. Bd3 Nf6 9. a3 Be6 10. O-O Bd6 11. h3 Ne5 12. Nxe5 Bxe5 13. Nd2 O-O 14. Rfe1 a6 15. Re2 b5 16. Nf3 Bd6 17. Kf1 Rc8 18. Rd1 Qe7 19. Rdd2 Rc7 20. Ke1 Rfc8 21. Rd1 h6 22. Kf1 Ne4 23. Rde1 Nf6 24. a4 bxa4 25. Ra1 Bc5 26. Bf4 Rc6 27. Rxa4 Bd6 28. Be3 Ne4 29. Ra1 Bc5 30. Bf4 Bb4 31. Be3 a5 32. Rd1 a4 33. Kg1 a3 34. bxa3 Bxa3 35. Ra1 Bc5 36. Bf4 Bd6 37. Be3 Qc7 38. Bd4 Bc5 39. Be5 Qb6 40. Rb1 Bxf2+ 41. Kh2 Qc5 42. Rf1 Qa7 43. Rfxf2 Nxf2 44. Bd4 Qa3 45. Rxf2 Qxd3 46. Rd2 Qb5 47. Rb2 Qa5 48. Rb3 Rc2 49. Ne5 Qa2 50. Rg3 Rxg2+ 51. Rxg2 Qb1 52. Nf3 Rc2 53. Rxc2 Qxc2+ 54. Kg3 Qc4 55. h4 Bf5 56. Kf4 Be4 57. h5 Bxf3 58. Kxf3 Qxd4 59. Ke2 Qe4+ 60. Kd2 d4 61. Kc1 d3 62. Kd2 Qe2+ 63. Kc3 d2 64. Kb4 d1=Q 65. Kc5 Qec2+ 66. Kb6 Qdb1+ 67. Ka5 Qca2# { Black wins by checkmate. } 0-1[/pgn]
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.
Uri Blass
Posts: 10491
Joined: Thu Mar 09, 2006 12:37 am
Location: Tel-Aviv Israel

Re: Reporte acerca de LeelaQueenOdds en control de tiempo 3'0 minutos tras 138 encuentros.

Post by Uri Blass »

Father wrote: Tue Oct 29, 2024 2:15 pm
Father wrote: Tue Oct 29, 2024 2:00 pm
lkaufman wrote: Sat Oct 26, 2024 12:52 am
Uri Blass wrote: Sat Oct 26, 2024 12:02 am
Father wrote: Fri Oct 25, 2024 4:25 am
Father wrote: Fri Oct 25, 2024 12:18 am
lkaufman wrote: Thu Oct 24, 2024 11:52 pm
Father wrote: Thu Oct 24, 2024 8:08 pm
Father wrote: Thu Oct 24, 2024 7:41 pm
lkaufman wrote: Thu Oct 24, 2024 6:16 pm
Father wrote: Thu Oct 24, 2024 3:14 pm
Father wrote: Thu Oct 24, 2024 6:29 am
lkaufman wrote: Mon Oct 07, 2024 5:01 pm
Father wrote: Mon Oct 07, 2024 4:47 pm
lkaufman wrote: Mon Oct 07, 2024 3:47 am
Father wrote: Mon Oct 07, 2024 1:57 am Mr. Larry Kaufman.
This performance in the fight of man against machine with odds is really wonderful.
I consider that in man-machine duels with odds, that in some way, there are real probabilities on the part of the human being to acquire progress in shortening the distances.
If at any time you wish or consider that it is useful for you to participate in man-versus-machine duels, I would be attentive to an invitation, in exchange for the sole pleasure of playing chess, without a prize of any kind.
Thank you. Your games have been helpful in highlighting the issue of allowing early repetition draws, we're working on ways to reduce them. If you want to know what I would consider most useful, I would like to see you play games with LeelaQueenForKnight, at whatever time limit allows you to win a decent percentage of the games. It's most helpful when players pick a handicap that is a bit difficult for them but where the win/loss ratio is not too one-sided. My guess is that you'll need a fairly slow blitz or even Rapid TC to do this; it's WAY harder than actual queen odds, though easier for the human than rook odds.
...excellent odds!! I have stated, ( 20 games ) I believe that as human beings playing against odds, we have the possibility of learning more about the odds and thus increase the probability of improving the results... I have a question; What could be the estimated elo of this odds? Thank you Mr. Larry Kaufman.
In LiChess blitz games that bot performs around 2400 Lichess level, which is considered to be equivalent to 2235 FIDE rating. Of course it would be lower at longer time controls, but you are playing blitz, so the above is appropriate for you.

1.733 / 5.000
Mr. Larry Kaufman. Just a few minutes ago, I finished my first match against "LeelaQueenOdds". I played ten matches, and the final result was the following: 6 points to the computer, 4 points to me. The computer won 5 matches against me, I won 3 matches against the computer, and there were two draws. I mistakenly believed that it would be a simple matter to beat the computer. But nothing could be further from the truth. I played the games with a 3-minute time control. The computer knocked me out four times in a row... I went into a deep shock. I was then forced to make a quick change of strategy, using all my experience. A round of only ten matches, and I was defeated four times. !! I asked myself what to do against this beast, who without legs and without one of its hands was annihilating me!! So, game number five was my first win, game number six a draw, game number seven my second win, game number eight a draw, game number nine represented the tragedy of my fifth loss, and game number ten represented my third win. If we divide the tournament into two stages, game number one to game number four shows a computer that takes me by surprise. And game number four to game ten, a modification of my flexible thinking, which gives a favorable result in my favor of four against two. But... I lost. Nothing to say, and instead, a lot to thank the creators of this ODDS for. Without a doubt, I recommend this odds. The reason is simple... it allows us to play chess, to improve ourselves, and it gives us the joy of playing against someone better than the human world champion, who gives us the opportunity without his queen, to play chess. Thank you.
...The sun that gives us light and life rises to the heavens. Today I have understood that we can be better, move forward, progress and evolve as human beings. Yesterday ending the day at night I lost 6-4. Today at the beginning of the day I beat the computer 7-3. I have tamed the furious horse, now it is time to ride the cybernetic horse. Our strategic thinking makes the conquering steed tame. Man and machine heading towards the north and ideals, crossing the seven seas. My Father and Catecan will conquer the world.

[pgn][Event "Casual blitz game"]
[Site "https://lichess.org/MtU1NIXV"]
[Date "2024.10.24"]
[White "LeelaQueenOdds"]
[Black "Catecan"]
[Result "0-1"]
[UTCDate "2024.10.24"]
[UTCTime "12:01:04"]
[WhiteElo "2000"]
[BlackElo "2082"]
[WhiteTitle "BOT"]
[Variant "From Position"]
[TimeControl "180+0"]
[ECO "?"]
[Opening "?"]
[Termination "Normal"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[SetUp "1"]
[Annotator "lichess.org"]

1. c4 e6 2. d3 d5 3. a3 c6 4. e4 dxc4 5. dxc4 e5 6. Nc3 Nf6 7. h3 a5 8. Be3 b6 9. Nf3 Nbd7 10. Rd1 Qc7 11. g4 h6 12. Rg1 Be7 13. Bd3 Ba6 14. b3 Bxa3 15. Ke2 Bc5 16. Bd2 Bb4 17. Ke3 Rd8 18. Bc2 Nc5 19. Ne2 Rxd2 20. Nxd2 Ke7 21. Nf3 Rd8 22. Rxd8 Qxd8 23. Rd1 Qc7 24. Bb1 g5 25. Ng3 Kf8 26. h4 Nxg4+ 27. Ke2 f6 28. hxg5 hxg5 29. Rh1 Kg8 30. Nh5 Qf7 31. Kf1 Ne6 32. Bd3 Nf4 33. Nxf4 exf4 34. e5 Nxe5 35. Nxe5 fxe5 36. Bh7+ Qxh7 37. Rxh7 Kxh7 38. Ke2 Kg6 39. Kd3 Kf6 40. Ke4 Bb7 41. c5 Bxc5 42. b4 axb4 43. Kd3 Ba6+ 44. Kc2 g4 45. Kb3 e4 46. Kc2 Bb5 47. Kb3 Ke5 48. Kc2 Bxf2 49. Kb3 Bc5 50. Kc2 e3 51. Kb2 f3 52. Kc2 g3 53. Kb2 Kf4 54. Kc2 Bd4 55. Kb3 c5 56. Ka2 Bc4+ 57. Kb1 b5 58. Kc2 b3+ 59. Kb1 e2 60. Kc1 Ke3 61. Kb1 f2 62. Kc1 g2 63. Kb1 e1=N 64. Kc1 f1=N 65. Kb1 g1=N 66. Kc1 b2+ 67. Kb1 Bd3+ 68. Ka2 b1=Q+ 69. Ka3 b4+ 70. Ka4 b3 71. Ka5 b2 72. Kb6 c4+ 73. Kc6 c3 74. Kd7 c2 75. Ke7 c1=N 76. Kd7 Qa1 77. Ke7 b1=N 78. Kf7 Qa7+ 79. Kf8 Bc5+ 80. Kg8 Bc4+ 81. Kh8 Bd4# { Black wins by checkmate. } 0-1[/pgn]

Note that right now, only the knight odds bot is specifically trained for odds, but it is likely that the other bots will be updated very soon to use the net trained for knight odds, as it seems to be stronger than the normal version even at larger odds. So once the switch is made, expect it to get even harder to win at queen odds!
... A question. Let's imagine an open tournament with about 200 participants. One of the 200 players is leelaOddsQueen. In the imaginary scenario, there are 5 grandmasters, 10 international masters, 30 top-class players, and 155 players of all kinds; The time control, 3 minutes per game, and 8 hours straight in the masnau. I wonder? Where in the table would OddsQueen probably be? Probable Elo? Thank you Mr. Larry Kaufman ...
Actually, that's a very easy question to answer. There are about 75 Lichess games at 3'0" at queen odds against players with Lichess ratings in the 2000 to 2200 range, with an almost exactly even score. You are the median player and very close to the mean at 2082. So therefore the expected Lichess blitz rating at queen odds should be very close to 2082. Now to convert that to a FIDE rating, the website https://chessgoals.com/rating-comparison/ is reliable and give 1916 FIDE as equivalent.
therefore "the queen" could mean or have a valley of 1200-1300 elo? ... This computer is certainly superior to the human world chess champion... therefore it seems that without a queen, it could statistically beat the world champion with relative ease... my score against this machine is 24.5 against 15.5 in my favor.
** Errata:
24.5
16.6
The average-strength human chess player seems to have a clear chance of success playing chess against the human world chess champion, with the champion having no queen.
- I clarify because the text was ambiguous.
In this case it means that the average human chess player is better than most people who play chess because I know that I can beat most people who know the rules without my queen and I believe that I can also beat most humans who know the chess rules without a queen and 2 rooks.

I have some experience in doing it against humans in a real board with no time limit and not only against children.

Of course adults who are too weak not to lose against me in these conditions do not play in chess tournaments but based on my experience in most cases it is not something easy to teach them to beat me.
I suppose he was talking about the average online player (1500 lichess blitz or 1200 chess.com blitz), although even then I think it is incorrect for 3'0" chess. For huge odds like queen odds, the result depends mostly on the elo of the receiver, so even Leela at say 3600 won't do a lot better than Magnus or Hikaru. Hikaru actually did a 3'0" queen odds run on chess.com and won the final game against a player rated slightly over 2000, although I think he recognized that it was a fluke and wouldn't expect to get a 2000 rating playing that way, maybe more like 1800 or so. But at longer time limits like Rapid it would be much lower, maybe 1500 Lichess or so, i.e. about the average. But for smaller handicaps like knight odds, the difference between what the engine can achieve and what Magnus or Hikaru can do is huge; in blitz it looks likely from the results that the LeelaKnightOdds bot can actually give Magnus or Hikaru knight odds at about 3'2" blitz and come out about even, though extrapolating from results against players around 2500 FIDE level is risky.
I would like to be useful to the LeelaQueenOdds programmers to share my own experience after 138 encounters:
1- Although each session of games against the computer is affected by my biorhythm, the levels of my fatigue and my emotions, I have observed °a changing personality in the nuances of the computer's thinking°
2- Sometimes the computer's thinking has been "on the downside", however I have come to consider that it has been part of the computer's plan, that of "making itself seem crazy" to deceive us, which is typical of historical leadership techniques. in the art of war and is called precisely this way: ° The theory of the madman °.
3- I have observed that for just over 24 hours, the profiles of the algorithm are growing in their power, to the point that I went from a session in which I considered in my first 10 games that I could have won ten to zero in my favor However, today's thirty-round session represented greater difficulty for me; I admit that I was more exhausted and sometimes distracted, but even so, I feel that the computer is not the same and that it is superior to that of other previous occasions in which it was very easy for me to overcome it.
4- The final result to date in my 138 matches is 83 in my favor against 55 for the computer, which represents a superiority in my level of strength compared to LeelaQueenOdds's 142 elo, however the scoreboard carries weights that At some point I have considered delimiting a difference of 420 elo in my favor. The doubt and question I have today is if the computer will really be increasing in power or if it will once again be easy prey for me.
6- In my opinion LeelaKnightOdds could be higher than 2000 Fide
..Correction. In my In my opinion LeelaQueenOdds - I wrote Knight- could be higher than 2000 Fide elo in these time controls, that is, having favorable final results in the total score the computer for Fide players of 2000 elo who participate in matches and combats in game levels of three minutes per game.
I think fide rating should be used for fide time control and the interesting question is what fide rating can LeelaQueenOdds get with 90+30 time control and what is possible to get.

I do not believe it is possible to get fide rating 2000 in these conditions but I guess you can get more than fide rating 1000.
lkaufman
Posts: 6030
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA

Re: Reporte acerca de LeelaQueenOdds en control de tiempo 3'0 minutos tras 138 encuentros.

Post by lkaufman »

Uri Blass wrote: Wed Oct 30, 2024 1:24 pm
Father wrote: Tue Oct 29, 2024 2:15 pm
Father wrote: Tue Oct 29, 2024 2:00 pm
lkaufman wrote: Sat Oct 26, 2024 12:52 am
Uri Blass wrote: Sat Oct 26, 2024 12:02 am
Father wrote: Fri Oct 25, 2024 4:25 am
Father wrote: Fri Oct 25, 2024 12:18 am
lkaufman wrote: Thu Oct 24, 2024 11:52 pm
Father wrote: Thu Oct 24, 2024 8:08 pm
Father wrote: Thu Oct 24, 2024 7:41 pm
lkaufman wrote: Thu Oct 24, 2024 6:16 pm
Father wrote: Thu Oct 24, 2024 3:14 pm
Father wrote: Thu Oct 24, 2024 6:29 am
lkaufman wrote: Mon Oct 07, 2024 5:01 pm
Father wrote: Mon Oct 07, 2024 4:47 pm
lkaufman wrote: Mon Oct 07, 2024 3:47 am
Father wrote: Mon Oct 07, 2024 1:57 am Mr. Larry Kaufman.
This performance in the fight of man against machine with odds is really wonderful.
I consider that in man-machine duels with odds, that in some way, there are real probabilities on the part of the human being to acquire progress in shortening the distances.
If at any time you wish or consider that it is useful for you to participate in man-versus-machine duels, I would be attentive to an invitation, in exchange for the sole pleasure of playing chess, without a prize of any kind.
Thank you. Your games have been helpful in highlighting the issue of allowing early repetition draws, we're working on ways to reduce them. If you want to know what I would consider most useful, I would like to see you play games with LeelaQueenForKnight, at whatever time limit allows you to win a decent percentage of the games. It's most helpful when players pick a handicap that is a bit difficult for them but where the win/loss ratio is not too one-sided. My guess is that you'll need a fairly slow blitz or even Rapid TC to do this; it's WAY harder than actual queen odds, though easier for the human than rook odds.
...excellent odds!! I have stated, ( 20 games ) I believe that as human beings playing against odds, we have the possibility of learning more about the odds and thus increase the probability of improving the results... I have a question; What could be the estimated elo of this odds? Thank you Mr. Larry Kaufman.
In LiChess blitz games that bot performs around 2400 Lichess level, which is considered to be equivalent to 2235 FIDE rating. Of course it would be lower at longer time controls, but you are playing blitz, so the above is appropriate for you.

1.733 / 5.000
Mr. Larry Kaufman. Just a few minutes ago, I finished my first match against "LeelaQueenOdds". I played ten matches, and the final result was the following: 6 points to the computer, 4 points to me. The computer won 5 matches against me, I won 3 matches against the computer, and there were two draws. I mistakenly believed that it would be a simple matter to beat the computer. But nothing could be further from the truth. I played the games with a 3-minute time control. The computer knocked me out four times in a row... I went into a deep shock. I was then forced to make a quick change of strategy, using all my experience. A round of only ten matches, and I was defeated four times. !! I asked myself what to do against this beast, who without legs and without one of its hands was annihilating me!! So, game number five was my first win, game number six a draw, game number seven my second win, game number eight a draw, game number nine represented the tragedy of my fifth loss, and game number ten represented my third win. If we divide the tournament into two stages, game number one to game number four shows a computer that takes me by surprise. And game number four to game ten, a modification of my flexible thinking, which gives a favorable result in my favor of four against two. But... I lost. Nothing to say, and instead, a lot to thank the creators of this ODDS for. Without a doubt, I recommend this odds. The reason is simple... it allows us to play chess, to improve ourselves, and it gives us the joy of playing against someone better than the human world champion, who gives us the opportunity without his queen, to play chess. Thank you.
...The sun that gives us light and life rises to the heavens. Today I have understood that we can be better, move forward, progress and evolve as human beings. Yesterday ending the day at night I lost 6-4. Today at the beginning of the day I beat the computer 7-3. I have tamed the furious horse, now it is time to ride the cybernetic horse. Our strategic thinking makes the conquering steed tame. Man and machine heading towards the north and ideals, crossing the seven seas. My Father and Catecan will conquer the world.

[pgn][Event "Casual blitz game"]
[Site "https://lichess.org/MtU1NIXV"]
[Date "2024.10.24"]
[White "LeelaQueenOdds"]
[Black "Catecan"]
[Result "0-1"]
[UTCDate "2024.10.24"]
[UTCTime "12:01:04"]
[WhiteElo "2000"]
[BlackElo "2082"]
[WhiteTitle "BOT"]
[Variant "From Position"]
[TimeControl "180+0"]
[ECO "?"]
[Opening "?"]
[Termination "Normal"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[SetUp "1"]
[Annotator "lichess.org"]

1. c4 e6 2. d3 d5 3. a3 c6 4. e4 dxc4 5. dxc4 e5 6. Nc3 Nf6 7. h3 a5 8. Be3 b6 9. Nf3 Nbd7 10. Rd1 Qc7 11. g4 h6 12. Rg1 Be7 13. Bd3 Ba6 14. b3 Bxa3 15. Ke2 Bc5 16. Bd2 Bb4 17. Ke3 Rd8 18. Bc2 Nc5 19. Ne2 Rxd2 20. Nxd2 Ke7 21. Nf3 Rd8 22. Rxd8 Qxd8 23. Rd1 Qc7 24. Bb1 g5 25. Ng3 Kf8 26. h4 Nxg4+ 27. Ke2 f6 28. hxg5 hxg5 29. Rh1 Kg8 30. Nh5 Qf7 31. Kf1 Ne6 32. Bd3 Nf4 33. Nxf4 exf4 34. e5 Nxe5 35. Nxe5 fxe5 36. Bh7+ Qxh7 37. Rxh7 Kxh7 38. Ke2 Kg6 39. Kd3 Kf6 40. Ke4 Bb7 41. c5 Bxc5 42. b4 axb4 43. Kd3 Ba6+ 44. Kc2 g4 45. Kb3 e4 46. Kc2 Bb5 47. Kb3 Ke5 48. Kc2 Bxf2 49. Kb3 Bc5 50. Kc2 e3 51. Kb2 f3 52. Kc2 g3 53. Kb2 Kf4 54. Kc2 Bd4 55. Kb3 c5 56. Ka2 Bc4+ 57. Kb1 b5 58. Kc2 b3+ 59. Kb1 e2 60. Kc1 Ke3 61. Kb1 f2 62. Kc1 g2 63. Kb1 e1=N 64. Kc1 f1=N 65. Kb1 g1=N 66. Kc1 b2+ 67. Kb1 Bd3+ 68. Ka2 b1=Q+ 69. Ka3 b4+ 70. Ka4 b3 71. Ka5 b2 72. Kb6 c4+ 73. Kc6 c3 74. Kd7 c2 75. Ke7 c1=N 76. Kd7 Qa1 77. Ke7 b1=N 78. Kf7 Qa7+ 79. Kf8 Bc5+ 80. Kg8 Bc4+ 81. Kh8 Bd4# { Black wins by checkmate. } 0-1[/pgn]

Note that right now, only the knight odds bot is specifically trained for odds, but it is likely that the other bots will be updated very soon to use the net trained for knight odds, as it seems to be stronger than the normal version even at larger odds. So once the switch is made, expect it to get even harder to win at queen odds!
... A question. Let's imagine an open tournament with about 200 participants. One of the 200 players is leelaOddsQueen. In the imaginary scenario, there are 5 grandmasters, 10 international masters, 30 top-class players, and 155 players of all kinds; The time control, 3 minutes per game, and 8 hours straight in the masnau. I wonder? Where in the table would OddsQueen probably be? Probable Elo? Thank you Mr. Larry Kaufman ...
Actually, that's a very easy question to answer. There are about 75 Lichess games at 3'0" at queen odds against players with Lichess ratings in the 2000 to 2200 range, with an almost exactly even score. You are the median player and very close to the mean at 2082. So therefore the expected Lichess blitz rating at queen odds should be very close to 2082. Now to convert that to a FIDE rating, the website https://chessgoals.com/rating-comparison/ is reliable and give 1916 FIDE as equivalent.
therefore "the queen" could mean or have a valley of 1200-1300 elo? ... This computer is certainly superior to the human world chess champion... therefore it seems that without a queen, it could statistically beat the world champion with relative ease... my score against this machine is 24.5 against 15.5 in my favor.
** Errata:
24.5
16.6
The average-strength human chess player seems to have a clear chance of success playing chess against the human world chess champion, with the champion having no queen.
- I clarify because the text was ambiguous.
In this case it means that the average human chess player is better than most people who play chess because I know that I can beat most people who know the rules without my queen and I believe that I can also beat most humans who know the chess rules without a queen and 2 rooks.

I have some experience in doing it against humans in a real board with no time limit and not only against children.

Of course adults who are too weak not to lose against me in these conditions do not play in chess tournaments but based on my experience in most cases it is not something easy to teach them to beat me.
I suppose he was talking about the average online player (1500 lichess blitz or 1200 chess.com blitz), although even then I think it is incorrect for 3'0" chess. For huge odds like queen odds, the result depends mostly on the elo of the receiver, so even Leela at say 3600 won't do a lot better than Magnus or Hikaru. Hikaru actually did a 3'0" queen odds run on chess.com and won the final game against a player rated slightly over 2000, although I think he recognized that it was a fluke and wouldn't expect to get a 2000 rating playing that way, maybe more like 1800 or so. But at longer time limits like Rapid it would be much lower, maybe 1500 Lichess or so, i.e. about the average. But for smaller handicaps like knight odds, the difference between what the engine can achieve and what Magnus or Hikaru can do is huge; in blitz it looks likely from the results that the LeelaKnightOdds bot can actually give Magnus or Hikaru knight odds at about 3'2" blitz and come out about even, though extrapolating from results against players around 2500 FIDE level is risky.
I would like to be useful to the LeelaQueenOdds programmers to share my own experience after 138 encounters:
1- Although each session of games against the computer is affected by my biorhythm, the levels of my fatigue and my emotions, I have observed °a changing personality in the nuances of the computer's thinking°
2- Sometimes the computer's thinking has been "on the downside", however I have come to consider that it has been part of the computer's plan, that of "making itself seem crazy" to deceive us, which is typical of historical leadership techniques. in the art of war and is called precisely this way: ° The theory of the madman °.
3- I have observed that for just over 24 hours, the profiles of the algorithm are growing in their power, to the point that I went from a session in which I considered in my first 10 games that I could have won ten to zero in my favor However, today's thirty-round session represented greater difficulty for me; I admit that I was more exhausted and sometimes distracted, but even so, I feel that the computer is not the same and that it is superior to that of other previous occasions in which it was very easy for me to overcome it.
4- The final result to date in my 138 matches is 83 in my favor against 55 for the computer, which represents a superiority in my level of strength compared to LeelaQueenOdds's 142 elo, however the scoreboard carries weights that At some point I have considered delimiting a difference of 420 elo in my favor. The doubt and question I have today is if the computer will really be increasing in power or if it will once again be easy prey for me.
6- In my opinion LeelaKnightOdds could be higher than 2000 Fide
..Correction. In my In my opinion LeelaQueenOdds - I wrote Knight- could be higher than 2000 Fide elo in these time controls, that is, having favorable final results in the total score the computer for Fide players of 2000 elo who participate in matches and combats in game levels of three minutes per game.
I think fide rating should be used for fide time control and the interesting question is what fide rating can LeelaQueenOdds get with 90+30 time control and what is possible to get.

I do not believe it is possible to get fide rating 2000 in these conditions but I guess you can get more than fide rating 1000.
Well, I think the minimum FIDE rating now is 1400, which is about what I would guess the result of your test would be, based on my calculation of 1900 FIDE at 3'0" blitz and subtracting 500 estimated elo difference between fast blitz and classical. For the knight odds bot, I think things have reached the point where a 90' + 30" match with a 2500 FIDE rated GM would be a tossup. If you have any candidates in mind for such a match let me know, perhaps we can offer some prize money.
Komodo rules!
lkaufman
Posts: 6030
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA

Re: Amazing results of LeelaKnightOdds-Dev

Post by lkaufman »

A player with a lichess blitz rating of 2800 and Rapid 2753 finally won a game from LeelaKnightOdds. It took him 90 games to win one, sixty two blitz and 28 Rapid games. The Rapid score was +25 = 2 -1 at time limits ranging from 9 + 3 to 15 + 3. Based on his LiChess ratings his FIDE rating should be about 2560. That works out to a performance rating of just over 3000, giving knight odds in Rapid!! Of course that's probably a lucky result, and extrapolating like this is speculative, but it does suggest that already a knight odds Rapid match with any of the World top ten would be competitive.
Komodo rules!
Father
Posts: 1510
Joined: Sun Mar 19, 2006 4:39 am
Location: Colombia
Full name: Pablo Ignacio Restrepo

Re: Amazing results of LeelaKnightOdds-Dev

Post by Father »

lkaufman wrote: Wed Oct 30, 2024 7:12 pm A player with a lichess blitz rating of 2800 and Rapid 2753 finally won a game from LeelaKnightOdds. It took him 90 games to win one, sixty two blitz and 28 Rapid games. The Rapid score was +25 = 2 -1 at time limits ranging from 9 + 3 to 15 + 3. Based on his LiChess ratings his FIDE rating should be about 2560. That works out to a performance rating of just over 3000, giving knight odds in Rapid!! Of course that's probably a lucky result, and extrapolating like this is speculative, but it does suggest that already a knight odds Rapid match with any of the World top ten would be competitive.
Mr. Larry Kaufman, good afternoon. There are some mathematical operations in determining the power at play of a human against a computer, which are causing me confusion in calculation. Let's see: Nakamura estimates that he has a playing power of 1800 playing without his queen; On the other hand, the hardware and software that drive LeelaQueenOdds is superior to the performance that Nakamura has against the human playing Nakamura without his queen; So then Nakamura could be estimated without his queen at 1800 elo, and LeelaQueenOdds, I would estimate, who could be 100 elo higher than Nakamura also playing without his queen the machine. So then, when I face LeelaQueenOdds I am facing a player with playing power between 1800 to 1900 elo. Taking these assumptions as a basis, and as an example my results today against LeelaQueenOdss which are 5 in my favor against 1 for the computer LeelaQueenOdds, we would have that my elo performance today has been +466 compared to the firepower of the computer, from which , my tournament elo today is 2222-2366. On the other hand, I wonder if scientifically, if it is possible or not possible, or better asked, if it is probable or if it is not probable, to convert LeelaQueenOdds into a player with a playing strength of 2000 Fide. In a purely anecdotal way, in approximately 2004, I remember that the performance in a line of 20 matches of the 5th player in the world ranking within a web club was 4 points out of 20 at that time, and today everything has changed ; At that same place and time yesterday my performance was 12 points out of 20 on or against the same computer and software. What I want to show with all this is that the measurement of human playing power is under prejudice.
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.
lkaufman
Posts: 6030
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA

Re: Amazing results of LeelaKnightOdds-Dev

Post by lkaufman »

Father wrote: Wed Oct 30, 2024 9:16 pm
lkaufman wrote: Wed Oct 30, 2024 7:12 pm A player with a lichess blitz rating of 2800 and Rapid 2753 finally won a game from LeelaKnightOdds. It took him 90 games to win one, sixty two blitz and 28 Rapid games. The Rapid score was +25 = 2 -1 at time limits ranging from 9 + 3 to 15 + 3. Based on his LiChess ratings his FIDE rating should be about 2560. That works out to a performance rating of just over 3000, giving knight odds in Rapid!! Of course that's probably a lucky result, and extrapolating like this is speculative, but it does suggest that already a knight odds Rapid match with any of the World top ten would be competitive.
Mr. Larry Kaufman, good afternoon. There are some mathematical operations in determining the power at play of a human against a computer, which are causing me confusion in calculation. Let's see: Nakamura estimates that he has a playing power of 1800 playing without his queen; On the other hand, the hardware and software that drive LeelaQueenOdds is superior to the performance that Nakamura has against the human playing Nakamura without his queen; So then Nakamura could be estimated without his queen at 1800 elo, and LeelaQueenOdds, I would estimate, who could be 100 elo higher than Nakamura also playing without his queen the machine. So then, when I face LeelaQueenOdds I am facing a player with playing power between 1800 to 1900 elo. Taking these assumptions as a basis, and as an example my results today against LeelaQueenOdss which are 5 in my favor against 1 for the computer LeelaQueenOdds, we would have that my elo performance today has been +466 compared to the firepower of the computer, from which , my tournament elo today is 2222-2366. On the other hand, I wonder if scientifically, if it is possible or not possible, or better asked, if it is probable or if it is not probable, to convert LeelaQueenOdds into a player with a playing strength of 2000 Fide. In a purely anecdotal way, in approximately 2004, I remember that the performance in a line of 20 matches of the 5th player in the world ranking within a web club was 4 points out of 20 at that time, and today everything has changed ; At that same place and time yesterday my performance was 12 points out of 20 on or against the same computer and software. What I want to show with all this is that the measurement of human playing power is under prejudice.
It is difficult for me to follow what you are saying, but if your point is that the meaning of a given rating like 2000 is different than it was twenty years ago, you are correct. My studies indicate that ratings for a given level of play have dropped about a hundred elo in that time period, so for example a player with a 1900 FIDE rating today plays about as well as a player with a 2000 FIDE rating did twenty years ago. Regarding your performance rating, a 5 to 1 score = +280 elo, not 466, I don't know where you got that number. Your overall results suggest that you play blitz at about the same level as a 2000 FIDE player today, or a 2100 FIDE player twenty years ago. But this is blitz, it doesn't tell us what your rating would be in classical chess.
Komodo rules!
Father
Posts: 1510
Joined: Sun Mar 19, 2006 4:39 am
Location: Colombia
Full name: Pablo Ignacio Restrepo

Re: Amazing results of LeelaKnightOdds-Dev

Post by Father »

lkaufman wrote: Thu Oct 31, 2024 12:18 am
Father wrote: Wed Oct 30, 2024 9:16 pm
lkaufman wrote: Wed Oct 30, 2024 7:12 pm A player with a lichess blitz rating of 2800 and Rapid 2753 finally won a game from LeelaKnightOdds. It took him 90 games to win one, sixty two blitz and 28 Rapid games. The Rapid score was +25 = 2 -1 at time limits ranging from 9 + 3 to 15 + 3. Based on his LiChess ratings his FIDE rating should be about 2560. That works out to a performance rating of just over 3000, giving knight odds in Rapid!! Of course that's probably a lucky result, and extrapolating like this is speculative, but it does suggest that already a knight odds Rapid match with any of the World top ten would be competitive.
Mr. Larry Kaufman, good afternoon. There are some mathematical operations in determining the power at play of a human against a computer, which are causing me confusion in calculation. Let's see: Nakamura estimates that he has a playing power of 1800 playing without his queen; On the other hand, the hardware and software that drive LeelaQueenOdds is superior to the performance that Nakamura has against the human playing Nakamura without his queen; So then Nakamura could be estimated without his queen at 1800 elo, and LeelaQueenOdds, I would estimate, who could be 100 elo higher than Nakamura also playing without his queen the machine. So then, when I face LeelaQueenOdds I am facing a player with playing power between 1800 to 1900 elo. Taking these assumptions as a basis, and as an example my results today against LeelaQueenOdss which are 5 in my favor against 1 for the computer LeelaQueenOdds, we would have that my elo performance today has been +466 compared to the firepower of the computer, from which , my tournament elo today is 2222-2366. On the other hand, I wonder if scientifically, if it is possible or not possible, or better asked, if it is probable or if it is not probable, to convert LeelaQueenOdds into a player with a playing strength of 2000 Fide. In a purely anecdotal way, in approximately 2004, I remember that the performance in a line of 20 matches of the 5th player in the world ranking within a web club was 4 points out of 20 at that time, and today everything has changed ; At that same place and time yesterday my performance was 12 points out of 20 on or against the same computer and software. What I want to show with all this is that the measurement of human playing power is under prejudice.
It is difficult for me to follow what you are saying, but if your point is that the meaning of a given rating like 2000 is different than it was twenty years ago, you are correct. My studies indicate that ratings for a given level of play have dropped about a hundred elo in that time period, so for example a player with a 1900 FIDE rating today plays about as well as a player with a 2000 FIDE rating did twenty years ago. Regarding your performance rating, a 5 to 1 score = +280 elo, not 466, I don't know where you got that number. Your overall results suggest that you play blitz at about the same level as a 2000 FIDE player today, or a 2100 FIDE player twenty years ago. But this is blitz, it doesn't tell us what your rating would be in classical chess.
Thank you Mr. Larry Kaufman. I would like to see and it would give me a lot of joy, to see and know, a LeelaQueenOdds computer, with an elo of 2000 Fide in play with a level of classic time control and of course also the machine with even higher power in blitz time control. I don't know what the limit of science is.
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.
Father
Posts: 1510
Joined: Sun Mar 19, 2006 4:39 am
Location: Colombia
Full name: Pablo Ignacio Restrepo

Re: Amazing results of LeelaKnightOdds-Dev

Post by Father »

lkaufman wrote: Thu Oct 31, 2024 12:18 am
Father wrote: Wed Oct 30, 2024 9:16 pm
lkaufman wrote: Wed Oct 30, 2024 7:12 pm A player with a lichess blitz rating of 2800 and Rapid 2753 finally won a game from LeelaKnightOdds. It took him 90 games to win one, sixty two blitz and 28 Rapid games. The Rapid score was +25 = 2 -1 at time limits ranging from 9 + 3 to 15 + 3. Based on his LiChess ratings his FIDE rating should be about 2560. That works out to a performance rating of just over 3000, giving knight odds in Rapid!! Of course that's probably a lucky result, and extrapolating like this is speculative, but it does suggest that already a knight odds Rapid match with any of the World top ten would be competitive.
Mr. Larry Kaufman, good afternoon. There are some mathematical operations in determining the power at play of a human against a computer, which are causing me confusion in calculation. Let's see: Nakamura estimates that he has a playing power of 1800 playing without his queen; On the other hand, the hardware and software that drive LeelaQueenOdds is superior to the performance that Nakamura has against the human playing Nakamura without his queen; So then Nakamura could be estimated without his queen at 1800 elo, and LeelaQueenOdds, I would estimate, who could be 100 elo higher than Nakamura also playing without his queen the machine. So then, when I face LeelaQueenOdds I am facing a player with playing power between 1800 to 1900 elo. Taking these assumptions as a basis, and as an example my results today against LeelaQueenOdss which are 5 in my favor against 1 for the computer LeelaQueenOdds, we would have that my elo performance today has been +466 compared to the firepower of the computer, from which , my tournament elo today is 2222-2366. On the other hand, I wonder if scientifically, if it is possible or not possible, or better asked, if it is probable or if it is not probable, to convert LeelaQueenOdds into a player with a playing strength of 2000 Fide. In a purely anecdotal way, in approximately 2004, I remember that the performance in a line of 20 matches of the 5th player in the world ranking within a web club was 4 points out of 20 at that time, and today everything has changed ; At that same place and time yesterday my performance was 12 points out of 20 on or against the same computer and software. What I want to show with all this is that the measurement of human playing power is under prejudice.
It is difficult for me to follow what you are saying, but if your point is that the meaning of a given rating like 2000 is different than it was twenty years ago, you are correct. My studies indicate that ratings for a given level of play have dropped about a hundred elo in that time period, so for example a player with a 1900 FIDE rating today plays about as well as a player with a 2000 FIDE rating did twenty years ago. Regarding your performance rating, a 5 to 1 score = +280 elo, not 466, I don't know where you got that number. Your overall results suggest that you play blitz at about the same level as a 2000 FIDE player today, or a 2100 FIDE player twenty years ago. But this is blitz, it doesn't tell us what your rating would be in classical chess.
.. also for me it is simply incredible and wonderful that it exists. a computer that without the queen has an elo of 1800. I want to tell you that despite having lived with a great imagination, the standards of LeelaQueenOdds exceed all expectations and dreams that my mind had in the eighties of the last century. . My favorite queen is LeelaQueeOdds, because this Queen allows me to play chess, and I love chess more than trench warfare behind the stone wall.
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.
lkaufman
Posts: 6030
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA

Re: Amazing results of LeelaKnightOdds-Dev

Post by lkaufman »

Father wrote: Thu Oct 31, 2024 2:47 am
lkaufman wrote: Thu Oct 31, 2024 12:18 am
Father wrote: Wed Oct 30, 2024 9:16 pm
lkaufman wrote: Wed Oct 30, 2024 7:12 pm A player with a lichess blitz rating of 2800 and Rapid 2753 finally won a game from LeelaKnightOdds. It took him 90 games to win one, sixty two blitz and 28 Rapid games. The Rapid score was +25 = 2 -1 at time limits ranging from 9 + 3 to 15 + 3. Based on his LiChess ratings his FIDE rating should be about 2560. That works out to a performance rating of just over 3000, giving knight odds in Rapid!! Of course that's probably a lucky result, and extrapolating like this is speculative, but it does suggest that already a knight odds Rapid match with any of the World top ten would be competitive.
Mr. Larry Kaufman, good afternoon. There are some mathematical operations in determining the power at play of a human against a computer, which are causing me confusion in calculation. Let's see: Nakamura estimates that he has a playing power of 1800 playing without his queen; On the other hand, the hardware and software that drive LeelaQueenOdds is superior to the performance that Nakamura has against the human playing Nakamura without his queen; So then Nakamura could be estimated without his queen at 1800 elo, and LeelaQueenOdds, I would estimate, who could be 100 elo higher than Nakamura also playing without his queen the machine. So then, when I face LeelaQueenOdds I am facing a player with playing power between 1800 to 1900 elo. Taking these assumptions as a basis, and as an example my results today against LeelaQueenOdss which are 5 in my favor against 1 for the computer LeelaQueenOdds, we would have that my elo performance today has been +466 compared to the firepower of the computer, from which , my tournament elo today is 2222-2366. On the other hand, I wonder if scientifically, if it is possible or not possible, or better asked, if it is probable or if it is not probable, to convert LeelaQueenOdds into a player with a playing strength of 2000 Fide. In a purely anecdotal way, in approximately 2004, I remember that the performance in a line of 20 matches of the 5th player in the world ranking within a web club was 4 points out of 20 at that time, and today everything has changed ; At that same place and time yesterday my performance was 12 points out of 20 on or against the same computer and software. What I want to show with all this is that the measurement of human playing power is under prejudice.
It is difficult for me to follow what you are saying, but if your point is that the meaning of a given rating like 2000 is different than it was twenty years ago, you are correct. My studies indicate that ratings for a given level of play have dropped about a hundred elo in that time period, so for example a player with a 1900 FIDE rating today plays about as well as a player with a 2000 FIDE rating did twenty years ago. Regarding your performance rating, a 5 to 1 score = +280 elo, not 466, I don't know where you got that number. Your overall results suggest that you play blitz at about the same level as a 2000 FIDE player today, or a 2100 FIDE player twenty years ago. But this is blitz, it doesn't tell us what your rating would be in classical chess.
Thank you Mr. Larry Kaufman. I would like to see and it would give me a lot of joy, to see and know, a LeelaQueenOdds computer, with an elo of 2000 Fide in play with a level of classic time control and of course also the machine with even higher power in blitz time control. I don't know what the limit of science is.
You will be glad to hear that I expect the rook odds and queen odds bots to be upgraded soon, to make use of the net trained for knight odds. Although we didn't expect that a net trained for knight odds would be better than the normal one at rook or queen odds, it turns out that it is noticeably stronger even when giving these odds. Apparently it has learned to play odds chess well in general, not just knight odds. I think you can expect that it will perform at about FIDE 2000 level in 3'0" blitz, but that's probably only about equal to 1500 FIDE at classical time limits. Once a net is specifically trained for queen odds, that may add another 100 elo or so. Beyond that it won't be easy to improve.
Komodo rules!