Eureka I found a better Contempt for Knight Odds For Dragon

[Rebel]

Larry, instead of material handicap games against GM's what would you expect playing time handicap games, for instance:

GM : 40m/2h
Komodo : one core, 5m blitz, ponder off.

[Fritz 0]

I think that even at 1 minute bullet Dragon or Komodo would destroy any human playing classical.

[Chessqueen]

Here is the 2nd game

2nd game against Colossus 2022a rated around 2764 CCRL with the same Opening Contempt = 45

[pgn][Event "Computer chess game"]
[Site "DESKTOP-OFQ3C0P"]
[Date "2022.05.26"]
[Round "2"]
[White "Dragon-2.6.1-64bit-avx2"]
[Black "Colossus 2022a"]
[Result "1/2-1/2"]
[BlackElo "2764"]
[Time "10:50:32"]
[WhiteElo "3600"]
[TimeControl "900+10"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/P1PPP1PP/RNBQKBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "327"]
[WhiteType "Program"]
[BlackType "Program"]

1. Bb2 Nf6 2. Nf3 Nc6 3. e3 g6 4. c4 Bg7 5. Bd3 d5 6. cxd5 Qxd5 7. Be2 O-O
8. O-O a6 9. Nc3 Qd6 10. Qc1 Bf5 11. Nd1 Rfe8 12. Nf2 e5 13. Ng5 h6 14.
Nxf7 Kxf7 15. g4 Be6 16. Ne4 Qb4 17. Bd3 Bd5 18. Ba3 Qa4 19. Bc2 Qa5 20.
Rb1 Rad8 21. g5 hxg5 22. Rxb7 Nd4 23. exd4 Bxb7 24. Bb3+ Bd5 25. Qc6 Qb5
26. Nxf6 Qxc6 27. Nxd5+ Ke6 28. Ne7+ Kd7 29. Nxc6 Kxc6 30. Ba4+ Kd5 31. Rf7
Bf8 32. Bxe8 Bxa3 33. Ba4 Kxd4 34. Rxc7 Bb4 35. Rd7+ Rxd7 36. Bxd7 Bxd2 37.
Kf2 Be3+ 38. Ke2 Bg1 39. h3 Ke4 40. Be8 Kf5 41. Kf3 e4+ 42. Ke2 Bd4 43. Bc6
e3 44. Kf3 a5 45. Be4+ Kf6 46. Bd3 a4 47. a3 Bc5 48. Bb5 Bxa3 49. Bxa4 Bc5
50. Bd7 Bd4 51. Bc8 Kf7 52. Bd7 Bc5 53. Bb5 Kg7 54. Ke2 Bb6 55. Bd3 Bd4 56.
Ba6 Kh6 57. Bc8 Bb6 58. Ba6 Kh5 59. Bc8 Bd4 60. Be6 Kh4 61. Bf7 Kxh3 62.
Bxg6 g4 63. Bf5 Kh4 64. Be4 g3 65. Bg2 Kg4 66. Kd3 Kf4 67. Ke2 Bb2 68. Bc6
Ba3 69. Kd3 Bb4 70. Ke2 Bc3 71. Kd3 Bd4 72. Ke2 Bc5 73. Bg2 Bd6 74. Bd5 Ke5
75. Bc6 Kd4 76. Bg2 Bb4 77. Kf3 Be7 78. Ke2 Bc5 79. Kf3 Bb4 80. Ke2 Bd6 81.
Kf3 Bc7 82. Ke2 Bf4 83. Ba8 Kc4 84. Bg2 Kb3 85. Kd3 Ka3 86. Bf3 Ka2 87. Bc6
Kb2 88. Bg2 Kc1 89. Bf3 Bh6 90. Ke2 Kc2 91. Be4+ Kb2 92. Bf3 Bf4 93. Bg2
Bg5 94. Kf3 Kc3 95. Bf1 Bf4 96. Ke2 Kb3 97. Kf3 Ka2 98. Bb5 Ka3 99. Be2 Kb4
100. Bf1 Kc5 101. Ke2 Bg5 102. Bg2 Kd6 103. Kf3 Bf4 104. Bh3 Kc7 105. Be6
Kb8 106. Bg4 Kb7 107. Bf5 Kc6 108. Be6 Kc5 109. Bh3 Kc4 110. Ke2 Bg5 111.
Bg2 Kd4 112. Bf1 Ke5 113. Kf3 e2 114. Bxe2 Bf4 115. Ba6 Kd4 116. Bc8 Be5
117. Bb7 Bd6 118. Ke2 Bf4 119. Kf1 Ke3 120. Ba8 Bd6 121. Bg2 Bb4 122. Bc6
Bc3 123. Kg2 Be5 124. Bb5 Kf4 125. Kf1 Bd4 126. Be8 Bf2 127. Kg2 Be3 128.
Ba4 Bd4 129. Bd1 Bb2 130. Be2 Bc3 131. Bh5 Bb4 132. Bg6 Ba3 133. Be8 Bb2
134. Bd7 Be5 135. Be8 Bd6 136. Kf1 Bc5 137. Bc6 Ke3 138. Bb7 Bd4 139. Ba8
Kf4 140. Bc6 Kg4 141. Kg2 Bc5 142. Bd5 Ba3 143. Bb3 Bc1 144. Bd5 Bf4 145.
Bb7 Bc7 146. Bc6 Ba5 147. Be8 Be1 148. Bd7+ Kf4 149. Ba4 Kg5 150. Bb3 Kf6
151. Ba4 Bf2 152. Bb5 Ke7 153. Kf1 Be3 154. Ba6 Bd4 155. Bc8 Kd8 156. Be6
Be3 157. Kg2 Bf4 158. Kf3 Be5 159. Bh3 Bd6 160. Bg2 Kc8 161. Bh1 Bc7 162.
Kg4 Kd7 163. Kh3 g2 164. Kxg2 {Insufficient material} 1/2-1/2[/pgn]

[lkaufman]

Larry, instead of material handicap games against GM's what would you expect playing time handicap games, for instance:

GM : 40m/2h
Komodo : one core, 5m blitz, ponder off.

We played many games with GM Alex Lenderman (FIDE 2600) using one thread of Dragon (a couple months ago), where he played with standard Rapid (15' + 10") time control. The ratio of time odds (no ponder, variety book, small variety of play added) needed for a fair match was about 5,000 to 1! Dragon needed an average of about 4 milliseconds per move!

[Fritz 0]

Larry, instead of material handicap games against GM's what would you expect playing time handicap games, for instance:

GM : 40m/2h
Komodo : one core, 5m blitz, ponder off.

We played many games with GM Alex Lenderman (FIDE 2600) using one thread of Dragon (a couple months ago), where he played with standard Rapid (15' + 10") time control. The ratio of time odds (no ponder, variety book, small variety of play added) needed for a fair match was about 5,000 to 1! Dragon needed an average of about 4 milliseconds per move!

I'm runing a match between Dragon 3 at full strength playing 1 minute for the whole game and Dragon 3 Elo 3000 (which should simulate a 2800 Elo player playing classical). After 354 games Dragon 3 full strength leads by 494 Elo (+317 =35 -2). I'm using one thread on my i5.

Btw, how do you set the engine to play a move in 4 miliseconds?

[lkaufman]

Larry, instead of material handicap games against GM's what would you expect playing time handicap games, for instance:

GM : 40m/2h
Komodo : one core, 5m blitz, ponder off.

We played many games with GM Alex Lenderman (FIDE 2600) using one thread of Dragon (a couple months ago), where he played with standard Rapid (15' + 10") time control. The ratio of time odds (no ponder, variety book, small variety of play added) needed for a fair match was about 5,000 to 1! Dragon needed an average of about 4 milliseconds per move!

I'm runing a match between Dragon 3 at full strength playing 1 minute for the whole game and Dragon 3 Elo 3000 (which should simulate a 2800 Elo player playing classical). After 354 games Dragon 3 full strength leads by 494 Elo (+317 =35 -2). I'm using one thread on my i5.

Btw, how do you set the engine to play a move in 4 miliseconds?

The games with Lenderman were played while calibrating the parameters for the Elo settings. At the higher levels, the ratings are basically defined by number of nodes searched at the end of an iteration, so quite similar to a "time" based rule with nodes per second reasonably constant on a given machine (variety also is used but the effect on elo is tiny at 2600 level as the value used is small). So we tried settings for 2600, and adjusted them as needed until it was reasonably close, and then I just looked at the average time per move taken, which was just a bit over 4 milliseconds.

[lkaufman]

Larry, instead of material handicap games against GM's what would you expect playing time handicap games, for instance:

GM : 40m/2h
Komodo : one core, 5m blitz, ponder off.

We played many games with GM Alex Lenderman (FIDE 2600) using one thread of Dragon (a couple months ago), where he played with standard Rapid (15' + 10") time control. The ratio of time odds (no ponder, variety book, small variety of play added) needed for a fair match was about 5,000 to 1! Dragon needed an average of about 4 milliseconds per move!

I'm runing a match between Dragon 3 at full strength playing 1 minute for the whole game and Dragon 3 Elo 3000 (which should simulate a 2800 Elo player playing classical). After 354 games Dragon 3 full strength leads by 494 Elo (+317 =35 -2). I'm using one thread on my i5.

I decided to see if I could confirm your result, by running the same test except against Dragon 3 Elo 3400 instead of 3000. After about 6700 games, the result for normal Dragon taking one minute was +94 Elo, in effect EXACTLY matching your +494 elo result vs Elo 3000!! Obviously this matching performance was just lucky, as there is a substantial margin of error around your result based on 354 games and of course the Dragon Elos are not exactly right, but it is indeed a remarkable coincidence! The 3400 elo version plays somewhat faster than the game/1 minute time limit (roughly game in 36 seconds on average) and uses its time less efficiently than standard Dragon, so the +94 elo result for normal Dragon is plausible.

These two results show why we place so much emphasis on material handicaps for human vs engine matches. If we wanted to make for an even match between Dragon on one thread and a human playing standard chess, let's consider how we could reduce the nearly 500 elo gap between Dragon on one thread playing bullet without ponder and human playing Classical time limit. First, human gets White every game (40 elo). Second, no opening book for Dragon (except for a couple moves of randomly choosing between good moves to avoid repeat games) (est. 50 elo). Also, no tablebases (maybe ten elo). So we're down to about 400 elo now. Then, give the human double the normal classical time limit (making total game time normal for classical play), and halve the time for the engine so it is playing hyper-bullet (game in 30 sec), maybe close to another 100 elo combined, still at least 300 elo gap. Next, choose Magnus C. as the opponent, bringing the gap down to about 240 elo due to his rating being 2864 rather than 2800. Finally, the big one, games are Armageddon, Black must win, so Magnus wins all draws, maybe 160 elo or so I estimate. So Dragon remains an estimated 80 elo favorite, but at least that's close enough to be interesting. So if anyone wants to sponsor a match between Dragon on one thread playing hyper-bullet with Black every game and no opening book and no ponder and no TBs and losing all drawn games against Magnus Carlsen playing White every game with 3 hours plus one minute increment per move and draw odds, I'll still bet on Dragon, but I might lose my bet. It would be very interesting, but it's a lot cheaper to play a knight odds match with a random 2500 rated GM!

[pepechuy]

I decided to see if I could confirm your result, by running the same test except against Dragon 3 Elo 3400 instead of 3000. After about 6700 games, the result for normal Dragon taking one minute was +94 Elo, in effect EXACTLY matching your +494 elo result vs Elo 3000!! Obviously this matching performance was just lucky, as there is a substantial margin of error around your result based on 354 games and of course the Dragon Elos are not exactly right, but it is indeed a remarkable coincidence! The 3400 elo version plays somewhat faster than the game/1 minute time limit (roughly game in 36 seconds on average) and uses its time less efficiently than standard Dragon, so the +94 elo result for normal Dragon is plausible.

These two results show why we place so much emphasis on material handicaps for human vs engine matches. If we wanted to make for an even match between Dragon on one thread and a human playing standard chess, let's consider how we could reduce the nearly 500 elo gap between Dragon on one thread playing bullet without ponder and human playing Classical time limit. First, human gets White every game (40 elo). Second, no opening book for Dragon (except for a couple moves of randomly choosing between good moves to avoid repeat games) (est. 50 elo). Also, no tablebases (maybe ten elo). So we're down to about 400 elo now. Then, give the human double the normal classical time limit (making total game time normal for classical play), and halve the time for the engine so it is playing hyper-bullet (game in 30 sec), maybe close to another 100 elo combined, still at least 300 elo gap. Next, choose Magnus C. as the opponent, bringing the gap down to about 240 elo due to his rating being 2864 rather than 2800. Finally, the big one, games are Armageddon, Black must win, so Magnus wins all draws, maybe 160 elo or so I estimate. So Dragon remains an estimated 80 elo favorite, but at least that's close enough to be interesting. So if anyone wants to sponsor a match between Dragon on one thread playing hyper-bullet with Black every game and no opening book and no ponder and no TBs and losing all drawn games against Magnus Carlsen playing White every game with 3 hours plus one minute increment per move and draw odds, I'll still bet on Dragon, but I might lose my bet. It would be very interesting, but it's a lot cheaper to play a knight odds match with a random 2500 rated GM!

Hi.

Carlsen has clearly stated that he is not interested in playing odds matches (perhaps he has played them in private, we will never know).
I doubt any amount of money will convince him.
And I believe he already earns more than what you could offer him, anyway (:

Greetings.

[Chessqueen]

I decided to see if I could confirm your result, by running the same test except against Dragon 3 Elo 3400 instead of 3000. After about 6700 games, the result for normal Dragon taking one minute was +94 Elo, in effect EXACTLY matching your +494 elo result vs Elo 3000!! Obviously this matching performance was just lucky, as there is a substantial margin of error around your result based on 354 games and of course the Dragon Elos are not exactly right, but it is indeed a remarkable coincidence! The 3400 elo version plays somewhat faster than the game/1 minute time limit (roughly game in 36 seconds on average) and uses its time less efficiently than standard Dragon, so the +94 elo result for normal Dragon is plausible.

These two results show why we place so much emphasis on material handicaps for human vs engine matches. If we wanted to make for an even match between Dragon on one thread and a human playing standard chess, let's consider how we could reduce the nearly 500 elo gap between Dragon on one thread playing bullet without ponder and human playing Classical time limit. First, human gets White every game (40 elo). Second, no opening book for Dragon (except for a couple moves of randomly choosing between good moves to avoid repeat games) (est. 50 elo). Also, no tablebases (maybe ten elo). So we're down to about 400 elo now. Then, give the human double the normal classical time limit (making total game time normal for classical play), and halve the time for the engine so it is playing hyper-bullet (game in 30 sec), maybe close to another 100 elo combined, still at least 300 elo gap. Next, choose Magnus C. as the opponent, bringing the gap down to about 240 elo due to his rating being 2864 rather than 2800. Finally, the big one, games are Armageddon, Black must win, so Magnus wins all draws, maybe 160 elo or so I estimate. So Dragon remains an estimated 80 elo favorite, but at least that's close enough to be interesting. So if anyone wants to sponsor a match between Dragon on one thread playing hyper-bullet with Black every game and no opening book and no ponder and no TBs and losing all drawn games against Magnus Carlsen playing White every game with 3 hours plus one minute increment per move and draw odds, I'll still bet on Dragon, but I might lose my bet. It would be very interesting, but it's a lot cheaper to play a knight odds match with a random 2500 rated GM!

Hi.

Carlsen has clearly stated that he is not interested in playing odds matches (perhaps he has played them in private, we will never know).
I doubt any amount of money will convince him.
And I believe he already earns more than what you could offer him, anyway (:

Greetings.

Never ever say that you doubt that any amount of money will convince him, I guarantee you that if a Billionaire offer Carlsen $500,000 to play 5 odds games verse Komodo Dragon which is $100,000 per game and $100,000 more than what Kasparov made playing versus Deep Blue back in 1997, Carlsen will not refuse the offer and play since he will make 10 times more than in any match of 5 games that he can play today.

Note: Kasparov bested Deep Blue in the match with three wins and two ties and took home the $400,000 prize in 1997. An estimated 6 million people worldwide followed the action online.

[Fritz 0]

Larry, instead of material handicap games against GM's what would you expect playing time handicap games, for instance:

GM : 40m/2h
Komodo : one core, 5m blitz, ponder off.

We played many games with GM Alex Lenderman (FIDE 2600) using one thread of Dragon (a couple months ago), where he played with standard Rapid (15' + 10") time control. The ratio of time odds (no ponder, variety book, small variety of play added) needed for a fair match was about 5,000 to 1! Dragon needed an average of about 4 milliseconds per move!

I'm runing a match between Dragon 3 at full strength playing 1 minute for the whole game and Dragon 3 Elo 3000 (which should simulate a 2800 Elo player playing classical). After 354 games Dragon 3 full strength leads by 494 Elo (+317 =35 -2). I'm using one thread on my i5.

I decided to see if I could confirm your result, by running the same test except against Dragon 3 Elo 3400 instead of 3000. After about 6700 games, the result for normal Dragon taking one minute was +94 Elo, in effect EXACTLY matching your +494 elo result vs Elo 3000!! Obviously this matching performance was just lucky, as there is a substantial margin of error around your result based on 354 games and of course the Dragon Elos are not exactly right, but it is indeed a remarkable coincidence! The 3400 elo version plays somewhat faster than the game/1 minute time limit (roughly game in 36 seconds on average) and uses its time less efficiently than standard Dragon, so the +94 elo result for normal Dragon is plausible.

These two results show why we place so much emphasis on material handicaps for human vs engine matches. If we wanted to make for an even match between Dragon on one thread and a human playing standard chess, let's consider how we could reduce the nearly 500 elo gap between Dragon on one thread playing bullet without ponder and human playing Classical time limit. First, human gets White every game (40 elo). Second, no opening book for Dragon (except for a couple moves of randomly choosing between good moves to avoid repeat games) (est. 50 elo). Also, no tablebases (maybe ten elo). So we're down to about 400 elo now. Then, give the human double the normal classical time limit (making total game time normal for classical play), and halve the time for the engine so it is playing hyper-bullet (game in 30 sec), maybe close to another 100 elo combined, still at least 300 elo gap. Next, choose Magnus C. as the opponent, bringing the gap down to about 240 elo due to his rating being 2864 rather than 2800. Finally, the big one, games are Armageddon, Black must win, so Magnus wins all draws, maybe 160 elo or so I estimate. So Dragon remains an estimated 80 elo favorite, but at least that's close enough to be interesting. So if anyone wants to sponsor a match between Dragon on one thread playing hyper-bullet with Black every game and no opening book and no ponder and no TBs and losing all drawn games against Magnus Carlsen playing White every game with 3 hours plus one minute increment per move and draw odds, I'll still bet on Dragon, but I might lose my bet. It would be very interesting, but it's a lot cheaper to play a knight odds match with a random 2500 rated GM!

And giving Carlsen 4 hours + 1 minute increment (which is double the normal classical time limit od 2 hours + 30 seconds) and using a slow hardware would decrease the difference a bit further.