ideas for odd humans championship against engines

[Uri Blass]

In all cases If 2 players get exactly the same number of points then the player who used less time to get the points is the winner.

Option 1:Play only a+b time control against Leela with some fixed odds(rook odds or different odds).

players are free to choose a when the sum of a and b in all their game is 180(or a different number).
Options for champions in different categories
Category 1:The player who get the maximal number of points
Category 2:The player who get the maximal number of points with more than 50% result.
Category 3:The player who get the maximal number of wins.



Option 2:Decide about time control of 10+5 with knight odds and 10 games and the player who get the maximal number of points is the champion.

Option 3:Play only 90+30 time control and a match of 6 games and the player who can win with the smallest odd is the champion.
If the best 2 players win with the same odds then the player who score more points is the champion and if they get the same number of points then the player who used less time is the champion.

I think that for this option if we want world championship then Leela should be able to give smaller odds than knight odds because maybe some players in the world can win with a knight for a pawn f7 odds or 2 pawn odds.

What is your opinion?

[lkaufman]

At the present time option 2 makes the most sense. The very best players in the world can still likely make a plus score with knight odds at 10'5" but not easily, so the best score at this should be a valid test of who is the best Rapid player in the World. All we need is a sponsor! As for Classical chess, option 3, the odds would need to be two pawns or knight for pawn (or pawn f7 and 3 moves perhaps), which are not quite so common/standard odds. Probably the knight odds net is also quite good at giving knight for pawn, but probably not as good as normal engines at giving two pawns or pawn and 3 moves as they are so different from knight odds. Of course a special net could be trained for any odds if there is enough motivation.

In general, knight odds is the most suitable for such competitions, due to minimal change from normal chess (no pawns missing, castling on both sides, knights are interchangeable), also the option of b1 and g1 odds doubles the opening possibilities, minimizing the value of preparing openings. My experience is that the larger odds, even rook odds, may allow the human to win by deep calculation as the engine takes a lot of risk and assumes that the human won't see anything that is not very obvious. But with knight odds there is more need to play well positionally and strategically as well as tactically. However knight odds is too much for the top players at classical time limits now, though probably fair for "par" grandmasters. Perhaps some day engines will even be able to give knight odds in classical to the very best players, but this would require another breakthru. A year ago I would have said this will never happen, but now I won't say that.

[Father]

At the present time option 2 makes the most sense. The very best players in the world can still likely make a plus score with knight odds at 10'5" but not easily, so the best score at this should be a valid test of who is the best Rapid player in the World. All we need is a sponsor! As for Classical chess, option 3, the odds would need to be two pawns or knight for pawn (or pawn f7 and 3 moves perhaps), which are not quite so common/standard odds. Probably the knight odds net is also quite good at giving knight for pawn, but probably not as good as normal engines at giving two pawns or pawn and 3 moves as they are so different from knight odds. Of course a special net could be trained for any odds if there is enough motivation.

In general, knight odds is the most suitable for such competitions, due to minimal change from normal chess (no pawns missing, castling on both sides, knights are interchangeable), also the option of b1 and g1 odds doubles the opening possibilities, minimizing the value of preparing openings. My experience is that the larger odds, even rook odds, may allow the human to win by deep calculation as the engine takes a lot of risk and assumes that the human won't see anything that is not very obvious. But with knight odds there is more need to play well positionally and strategically as well as tactically. However knight odds is too much for the top players at classical time limits now, though probably fair for "par" grandmasters. Perhaps some day engines will even be able to give knight odds in classical to the very best players, but this would require another breakthru. A year ago I would have said this will never happen, but now I won't say that.

In my opinion, chess machines will beat all humans, even in the case of LeelaQueenOdds.

[towforce]

In my opinion, chess machines will beat all humans, even in the case of LeelaQueenOdds.

This appears to overestimate the limit of chess itself. However, we can work this out:

1. The last time I checked, extrapolating elo level vs draw rate seemed to put the upper limit of chess at around 4200. Probably time to update this estimate

2. Anyone happen to know what elo difference is required to offer queen odds?

[lkaufman]

In my opinion, chess machines will beat all humans, even in the case of LeelaQueenOdds.

This appears to overestimate the limit of chess itself. However, we can work this out:

1. The last time I checked, extrapolating elo level vs draw rate seemed to put the upper limit of chess at around 4200. Probably time to update this estimate

2. Anyone happen to know what elo difference is required to offer queen odds?

On point 1, if we are talking about classical time limits and playing only the best openings (ones often seen in top level human events or in engine competitions where tournament books are used), the best guide is the SSDF rating list. It has Leela on top at 3586 followed by Stockfish 16 at 3582, probably Stockfish 17 would be a few elo higher. Progress has slowed to a few elo per year as nearly all games are drawn under these conditions between the top engines. The upper limit probably won't be much over 3600 under these conditions. But this isn't quite fair, since the stronger engine should choose less drawish openings when the elo gap is meaningful, and also should modify its play to take more chances, so if we allow for that the limit goes up. As a rough guess, FRC data suggests that the rating differences might be increased by a third or so, so if we assume a 3700 max under current rules and a 2800 fixed rating point, we get a max rating of about 4000.

On point 2, the rating difference for odds depends on the level of the players and the time control, but I can calculate the maximum, which should apply for the current use. Once the level reaches the point where the weaker player can only aim for a draw (without odds), the win/draw threshold is 191 elo, and it is known that this threshold corresponds to 0.7 to 0.75 pawns. So a "clean" pawn up is about 260 elo. In the opening, the approximate rounded piece values are pawn = 1, knight = 3.5, bishop = 4, rook = 5, and queen = 9.5. First move is about a quarter pawn. So (with odds giver White) knight odds = 845 elo, rook odds = 1235 elo, and queen odds = 2405. Actual handicap values in blitz games between humans are only about half of these values, but we are talking about classical games at the highest possible level, so these numbers should be roughly correct.

The conclusion seems clear: Classical time limit knight odds against the world's best human player isn't reasonable now, but does look plausible in the future. Classical time limit rook odds against the world's best human player should never be balanced if you take these numbers as the absolute truth, but they are estimates and subject to various errors, so I would have to say that there is some non-trivial chance that some day an engine will win a classical time limit match from the best human player at rook odds! But that should be the absolute limit, these values aren't subject to enough error to allow for larger handicaps to be plausible.

[towforce]

In my opinion, chess machines will beat all humans, even in the case of LeelaQueenOdds.

This appears to overestimate the limit of chess itself. However, we can work this out:

1. The last time I checked, extrapolating elo level vs draw rate seemed to put the upper limit of chess at around 4200. Probably time to update this estimate

2. Anyone happen to know what elo difference is required to offer queen odds?

On point 1, if we are talking about classical time limits and playing only the best openings (ones often seen in top level human events or in engine competitions where tournament books are used), the best guide is the SSDF rating list. It has Leela on top at 3586 followed by Stockfish 16 at 3582, probably Stockfish 17 would be a few elo higher. Progress has slowed to a few elo per year as nearly all games are drawn under these conditions between the top engines. The upper limit probably won't be much over 3600 under these conditions. But this isn't quite fair, since the stronger engine should choose less drawish openings when the elo gap is meaningful, and also should modify its play to take more chances, so if we allow for that the limit goes up. As a rough guess, FRC data suggests that the rating differences might be increased by a third or so, so if we assume a 3700 max under current rules and a 2800 fixed rating point, we get a max rating of about 4000.

On point 2, the rating difference for odds depends on the level of the players and the time control, but I can calculate the maximum, which should apply for the current use. Once the level reaches the point where the weaker player can only aim for a draw (without odds), the win/draw threshold is 191 elo, and it is known that this threshold corresponds to 0.7 to 0.75 pawns. So a "clean" pawn up is about 260 elo. In the opening, the approximate rounded piece values are pawn = 1, knight = 3.5, bishop = 4, rook = 5, and queen = 9.5. First move is about a quarter pawn. So (with odds giver White) knight odds = 845 elo, rook odds = 1235 elo, and queen odds = 2405. Actual handicap values in blitz games between humans are only about half of these values, but we are talking about classical games at the highest possible level, so these numbers should be roughly correct.

The conclusion seems clear: Classical time limit knight odds against the world's best human player isn't reasonable now, but does look plausible in the future. Classical time limit rook odds against the world's best human player should never be balanced if you take these numbers as the absolute truth, but they are estimates and subject to various errors, so I would have to say that there is some non-trivial chance that some day an engine will win a classical time limit match from the best human player at rook odds! But that should be the absolute limit, these values aren't subject to enough error to allow for larger handicaps to be plausible.

Larry - thank you for that very informative response.

[Father]

In my opinion, chess machines will beat all humans, even in the case of LeelaQueenOdds.

This appears to overestimate the limit of chess itself. However, we can work this out:

1. The last time I checked, extrapolating elo level vs draw rate seemed to put the upper limit of chess at around 4200. Probably time to update this estimate

2. Anyone happen to know what elo difference is required to offer queen odds?

On point 1, if we are talking about classical time limits and playing only the best openings (ones often seen in top level human events or in engine competitions where tournament books are used), the best guide is the SSDF rating list. It has Leela on top at 3586 followed by Stockfish 16 at 3582, probably Stockfish 17 would be a few elo higher. Progress has slowed to a few elo per year as nearly all games are drawn under these conditions between the top engines. The upper limit probably won't be much over 3600 under these conditions. But this isn't quite fair, since the stronger engine should choose less drawish openings when the elo gap is meaningful, and also should modify its play to take more chances, so if we allow for that the limit goes up. As a rough guess, FRC data suggests that the rating differences might be increased by a third or so, so if we assume a 3700 max under current rules and a 2800 fixed rating point, we get a max rating of about 4000.

On point 2, the rating difference for odds depends on the level of the players and the time control, but I can calculate the maximum, which should apply for the current use. Once the level reaches the point where the weaker player can only aim for a draw (without odds), the win/draw threshold is 191 elo, and it is known that this threshold corresponds to 0.7 to 0.75 pawns. So a "clean" pawn up is about 260 elo. In the opening, the approximate rounded piece values are pawn = 1, knight = 3.5, bishop = 4, rook = 5, and queen = 9.5. First move is about a quarter pawn. So (with odds giver White) knight odds = 845 elo, rook odds = 1235 elo, and queen odds = 2405. Actual handicap values in blitz games between humans are only about half of these values, but we are talking about classical games at the highest possible level, so these numbers should be roughly correct.

The conclusion seems clear: Classical time limit knight odds against the world's best human player isn't reasonable now, but does look plausible in the future. Classical time limit rook odds against the world's best human player should never be balanced if you take these numbers as the absolute truth, but they are estimates and subject to various errors, so I would have to say that there is some non-trivial chance that some day an engine will win a classical time limit match from the best human player at rook odds! But that should be the absolute limit, these values aren't subject to enough error to allow for larger handicaps to be plausible.

Good morning Mr. Larry Kaufman. Good morning to everyone in the chat, I wish you a happy new year 2025 and many motivations to continue with the chess game of existence. I wish you the best plays, good health, provision, goals and dreams to fulfill and achieve. As for LeelaQueenOdds, the queen of boots, I hope that my dreams become reality, and soon LeelaQueenOdds climbs a few more steps towards glory, transforming the impossible into the possible and the magically unattainable into the plausible. For everyone: Good winds and good seas in the stormy journey of my life, may we all reach a good port. Happy new year 2025.

[Father]

In my opinion, chess machines will beat all humans, even in the case of LeelaQueenOdds.

** the stormy journey of the life
This appears to overestimate the limit of chess itself. However, we can work this out:

1. The last time I checked, extrapolating elo level vs draw rate seemed to put the upper limit of chess at around 4200. Probably time to update this estimate

2. Anyone happen to know what elo difference is required to offer queen odds?

On point 1, if we are talking about classical time limits and playing only the best openings (ones often seen in top level human events or in engine competitions where tournament books are used), the best guide is the SSDF rating list. It has Leela on top at 3586 followed by Stockfish 16 at 3582, probably Stockfish 17 would be a few elo higher. Progress has slowed to a few elo per year as nearly all games are drawn under these conditions between the top engines. The upper limit probably won't be much over 3600 under these conditions. But this isn't quite fair, since the stronger engine should choose less drawish openings when the elo gap is meaningful, and also should modify its play to take more chances, so if we allow for that the limit goes up. As a rough guess, FRC data suggests that the rating differences might be increased by a third or so, so if we assume a 3700 max under current rules and a 2800 fixed rating point, we get a max rating of about 4000.

On point 2, the rating difference for odds depends on the level of the players and the time control, but I can calculate the maximum, which should apply for the current use. Once the level reaches the point where the weaker player can only aim for a draw (without odds), the win/draw threshold is 191 elo, and it is known that this threshold corresponds to 0.7 to 0.75 pawns. So a "clean" pawn up is about 260 elo. In the opening, the approximate rounded piece values are pawn = 1, knight = 3.5, bishop = 4, rook = 5, and queen = 9.5. First move is about a quarter pawn. So (with odds giver White) knight odds = 845 elo, rook odds = 1235 elo, and queen odds = 2405. Actual handicap values in blitz games between humans are only about half of these values, but we are talking about classical games at the highest possible level, so these numbers should be roughly correct.

The conclusion seems clear: Classical time limit knight odds against the world's best human player isn't reasonable now, but does look plausible in the future. Classical time limit rook odds against the world's best human player should never be balanced if you take these numbers as the absolute truth, but they are estimates and subject to various errors, so I would have to say that there is some non-trivial chance that some day an engine will win a classical time limit match from the best human player at rook odds! But that should be the absolute limit, these values aren't subject to enough error to allow for larger handicaps to be plausible.

Good morning Mr. Larry Kaufman. Good morning to everyone in the chat, I wish you a happy new year 2025 and many motivations to continue with the chess game of existence. I wish you the best plays, good health, provision, goals and dreams to fulfill and achieve. As for LeelaQueenOdds, the queen of boots, I hope that my dreams become reality, and soon LeelaQueenOdds climbs a few more steps towards glory, transforming the impossible into the possible and the magically unattainable into the plausible. For everyone: Good winds and good seas in the stormy journey of my life, may we all reach a good port. Happy new year 2025.

[Father]

In my opinion, chess machines will beat all humans, even in the case of LeelaQueenOdds.

** the stormy journey of the life
This appears to overestimate the limit of chess itself. However, we can work this out:

1. The last time I checked, extrapolating elo level vs draw rate seemed to put the upper limit of chess at around 4200. Probably time to update this estimate

2. Anyone happen to know what elo difference is required to offer queen odds?

On point 1, if we are talking about classical time limits and playing only the best openings (ones often seen in top level human events or in engine competitions where tournament books are used), the best guide is the SSDF rating list. It has Leela on top at 3586 followed by Stockfish 16 at 3582, probably Stockfish 17 would be a few elo higher. Progress has slowed to a few elo per year as nearly all games are drawn under these conditions between the top engines. The upper limit probably won't be much over 3600 under these conditions. But this isn't quite fair, since the stronger engine should choose less drawish openings when the elo gap is meaningful, and also should modify its play to take more chances, so if we allow for that the limit goes up. As a rough guess, FRC data suggests that the rating differences might be increased by a third or so, so if we assume a 3700 max under current rules and a 2800 fixed rating point, we get a max rating of about 4000.

On point 2, the rating difference for odds depends on the level of the players and the time control, but I can calculate the maximum, which should apply for the current use. Once the level reaches the point where the weaker player can only aim for a draw (without odds), the win/draw threshold is 191 elo, and it is known that this threshold corresponds to 0.7 to 0.75 pawns. So a "clean" pawn up is about 260 elo. In the opening, the approximate rounded piece values are pawn = 1, knight = 3.5, bishop = 4, rook = 5, and queen = 9.5. First move is about a quarter pawn. So (with odds giver White) knight odds = 845 elo, rook odds = 1235 elo, and queen odds = 2405. Actual handicap values in blitz games between humans are only about half of these values, but we are talking about classical games at the highest possible level, so these numbers should be roughly correct.

The conclusion seems clear: Classical time limit knight odds against the world's best human player isn't reasonable now, but does look plausible in the future. Classical time limit rook odds against the world's best human player should never be balanced if you take these numbers as the absolute truth, but they are estimates and subject to various errors, so I would have to say that there is some non-trivial chance that some day an engine will win a classical time limit match from the best human player at rook odds! But that should be the absolute limit, these values aren't subject to enough error to allow for larger handicaps to be plausible.

Good morning Mr. Larry Kaufman. Good morning to everyone in the chat, I wish you a happy new year 2025 and many motivations to continue with the chess game of existence. I wish you the best plays, good health, provision, goals and dreams to fulfill and achieve. As for LeelaQueenOdds, the queen of boots, I that my dreams become reality, and soon LeelaQueenOdds climbs a few more steps towards glory, transforming the impossible into the possible and the magically unattainable into the plausible. For everyone: Good winds and good seas in the stormy journey of my life, may we all reach a good port. Happy new year 2025

** the stormy journey of the life

[Father]

In my opinion, chess machines will beat all humans, even in the case of LeelaQueenOdds.

** the stormy journey of the life
This appears to overestimate the limit of chess itself. However, we can work this out:

1. The last time I checked, extrapolating elo level vs draw rate seemed to put the upper limit of chess at around 4200. Probably time to update this estimate

2. Anyone happen to know what elo difference is required to offer queen odds?

On point 1, if we are talking about classical time limits and playing only the best openings (ones often seen in top level human events or in engine competitions where tournament books are used), the best guide is the SSDF rating list. It has Leela on top at 3586 followed by Stockfish 16 at 3582, probably Stockfish 17 would be a few elo higher. Progress has slowed to a few elo per year as nearly all games are drawn under these conditions between the top engines. The upper limit probably won't be much over 3600 under these conditions. But this isn't quite fair, since the stronger engine should choose less drawish openings when the elo gap is meaningful, and also should modify its play to take more chances, so if we allow for that the limit goes up. As a rough guess, FRC data suggests that the rating differences might be increased by a third or so, so if we assume a 3700 max under current rules and a 2800 fixed rating point, we get a max rating of about 4000.

On point 2, the rating difference for odds depends on the level of the players and the time control, but I can calculate the maximum, which should apply for the current use. Once the level reaches the point where the weaker player can only aim for a draw (without odds), the win/draw threshold is 191 elo, and it is known that this threshold corresponds to 0.7 to 0.75 pawns. So a "clean" pawn up is about 260 elo. In the opening, the approximate rounded piece values are pawn = 1, knight = 3.5, bishop = 4, rook = 5, and queen = 9.5. First move is about a quarter pawn. So (with odds giver White) knight odds = 845 elo, rook odds = 1235 elo, and queen odds = 2405. Actual handicap values in blitz games between humans are only about half of these values, but we are talking about classical games at the highest possible level, so these numbers should be roughly correct.

The conclusion seems clear: Classical time limit knight odds against the world's best human player isn't reasonable now, but does look plausible in the future. Classical time limit rook odds against the world's best human player should never be balanced if you take these numbers as the absolute truth, but they are estimates and subject to various errors, so I would have to say that there is some non-trivial chance that some day an engine will win a classical time limit match from the best human player at rook odds! But that should be the absolute limit, these values aren't subject to enough error to allow for larger handicaps to be plausible.

Good morning Mr. Larry Kaufman. Good morning to everyone in the chat, I wish you a happy new year 2025 and many motivations to continue with the chess game of existence. I wish you the best plays, good health, provision, goals and dreams to fulfill and achieve. As for LeelaQueenOdds, the queen of boots, I that my dreams become reality, and soon LeelaQueenOdds climbs a few more steps towards glory, transforming the impossible into the possible and the magically unattainable into the plausible. For everyone: Good winds and good seas in the stormy journey of my life, may we all reach a good port. Happy new year 2025

** the stormy journey of the life

..One question please... I thought I understood that the number of games I currently play against LeelaQueenOdds is not decisive for my elo score... for example since the last
Matching my rating until today, there are 634 new games, all bullet at 1'0" and me with the black tools...by the way, I don't know if it is a false assessment on my part, but I understand the algorithm especially "anti- Father" ... is it just my imagination, or does LeelaQueenOdds play better against me? ...