Komodo Dragon Vs. GM Perelshteyn Where is the Link ?

[Chessqueen]

I took another game that GM Perelshteyn won and replace it with "Gaviota-win32-0.84 playing with contempt =65 . Then I decided to run it 100 times with contempt = 65 with the same opening and another 100 times with contempt = 125 the more aggressive contempt = 65 won by 10% over contempt =100

[pgn][Event "knight odds match"]
[Site "Chess.com"]
[Date "2021.11.06"]
[Round "100"]
[White "Dragon-2.5-64bit-avx2"]
[Black "Gaviota-win32-0.84"]
[Result "1-0"]
[BlackElo "2500"]
[WhiteElo "2500"]
[TimeControl "300+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "355"]
[WhiteType "program"]
[BlackType "program"]

1. f4 {[%clk 0:15:10]} d5 {[%clk 0:15:08.5]} 2. Nf3 {[%clk 0:15:19.9]} g6
{[%clk 0:15:14.7]} 3. e3 {[%clk 0:15:29.8]} Bg7 {[%clk 0:15:23.4]} 4. Be2
{[%clk 0:15:39.7]} c5 {[%clk 0:15:26.3]} 5. O-O {[%clk 0:15:45.5]} Nc6
{[%clk 0:15:34.5]} 6. Qe1 {[%clk 0:15:48.4]} Bg4 {[%clk 0:15:35.7]} 7. d3
Qa5 8. Bd2 Qb5 9. h3 Bxf3 10. Bxf3 Qxb2 11. Rb1 Qxc2 12. Rxb7 Nd8 13. Rb3
Nf6 14. Qc1 Qxc1 15. Rxc1 Ne6 16. g4 c4 17. dxc4 Nc5 18. Rb5 Nfe4 19. Rc2
a6 20. Rxc5 Nxc5 21. cxd5 Nd7 22. Rc7 Rd8 23. d6 exd6 24. Bc6 Ke7 25. Ba5
g5 26. Rb7 Bh6 27. f5 Bg7 28. Kg2 Be5 29. Ra7 Bb2 30. Kf3 Ba3 31. Bxd8+
Rxd8 32. Ke4 Bc5 33. Rb7 f6 34. Kd5 Bxe3 35. Rc7 Bd2 36. Rb7 Bc3 37. Rc7
Be1 38. Ra7 Bf2 39. Rb7 Bg3 40. Ra7 Bf2 41. Rb7 Bg3 42. Ra7 Be1 43. Rb7 Bc3
44. Ra7 Bb4 45. Rb7 Bd2 46. Rc7 Be3 47. Rb7 Bf4 48. Rc7 Be5 49. Rb7 Bh2 50.
Ba4 Be5 51. Bc6 Bh2 52. Ba4 h6 53. Bc6 Bf4 54. Rc7 Bd2 55. Ra7 Be3 56. Rc7
Bf2 57. Rb7 Bg3 58. Ra7 Bf2 59. Rb7 Be3 60. Rc7 Bf4 61. Rb7 Bg3 62. Ra7 Bh2
63. Rb7 Bf4 64. Rc7 Bd2 65. Rb7 Bc3 66. Rc7 Be1 67. Rb7 Bc3 68. Rc7 Bb4 69.
Rb7 Ba5 70. Ra7 Be1 71. Rc7 Ba5 72. Rb7 Bd2 73. Ra7 Bb4 74. Rc7 Ba5 75. Rb7
Be1 76. Rc7 Bb4 77. Ra7 Bc5 78. Rb7 Bf2 79. Rc7 Bg1 80. Rb7 Bh2 81. Ba4 Bf4
82. Rc7 Be5 83. Rb7 Bc3 84. Rc7 Be5 85. Rb7 Bf4 86. Rc7 a5 87. Rb7 Be3 88.
Rc7 Bc5 89. Bb5 Bb4 90. Bc6 Bd2 91. Ba4 Bb4 92. Bc6 Bd2 93. Ba4 Be3 94. Bb5
Bb6 95. Rb7 Bc5 96. Rc7 Be3 97. Ba4 Bb6 98. Rb7 Bc5 99. Rc7 Bb6 100. Rb7
Bc5 101. Rc7 Ba3 102. Rb7 Bb4 103. Rc7 Be1 104. Bc6 Bg3 105. Ba4 Be5 106.
Bb5 Bf4 107. Ba4 Be5 108. Bb5 Bf4 109. Ba4 Bd2 110. Bb5 Be1 111. Ba4 Bf2
112. Bb5 Bb6 113. Rb7 Bf2 114. Rc7 Be1 115. Bc6 Bg3 116. Ba4 Bf4 117. Rb7
Be5 118. Rc7 Ba1 119. Bb5 Be5 120. Ba4 Bh2 121. Bb5 Bg1 122. Ba4 Be3 123.
Bb5 Bf2 124. Ba4 Bg3 125. Bc6 a4 126. Bxa4 Be1 127. Ra7 Bd2 128. Bc6 Be3
129. Rb7 Bf4 130. Ra7 Be3 131. Rb7 Bc1 132. Ba4 Bf4 133. Bc6 Be5 134. a4
Bc3 135. Ra7 Bd2 136. a5 Bxa5 137. Rxa5 Nb6+ 138. Kd4 Rc8 139. Ra7+ Kd8
140. Bd5 Rc2 141. Be6 Rh2 142. Rb7 Na4 143. Rf7 Rxh3 144. Rxf6 Nc5 145. Bc4
Ke7 146. Rf7+ Ke8 147. Ra7 Rg3 148. Kd5 Rxg4 149. Bb5+ Kf8 150. Kxd6 Rf4
151. Ke5 Ne4 152. Bc4 Rf2 153. Bd3 Nc5 154. Bb5 Rb2 155. Ra8+ Kg7 156. f6+
Kh7 157. Ra7+ Kh8 158. Kf5 Ne4 159. Kxe4 Rxb5 160. Ra8+ Kh7 161. f7 Rb4+
162. Ke5 Rb5+ 163. Ke6 Rb6+ 164. Ke7 Rb7+ 165. Ke8 Rxf7 166. Kxf7 h5 167.
Kf6 h4 168. Kxg5 Kg7 169. Ra7+ Kf8 170. Kxh4 Ke8 171. Kg5 Kd8 172. Kf6 Kc8
173. Ke6 Kb8 174. Rh7 Ka8 175. Kd6 Kb8 176. Kc6 Ka8 177. Kb6 Kb8 178. Rh8#[/pgn]

This game shows why human GMs do better with odds than most engines. In this game, when the fifty move rule approached, Gaviota just gave away a vital pawn to start the count over; a human would see that this could only help the opponent. Probably only a weakened Komodo/Dragon with a highly negative Contempt setting would settle for the draw rather than play this silly giveaway.

In your opinion when will Komodo Dragon xxx.x be able to beat GM Pereshteyn in a match of 15'+10" without counting draw as a win for Komodo Dragon? Probably you should only play 2 games out of 6 with Knight b1 missing. In 6 months can a GM with similar rating as GM Pereshteyn from India play against Komodo Dragon? The sale of Komodo Dragon xxx.x in India with an estimated chess playing fan of over 30 Million can be a great advertising opportunity.

Indian Players from this: list

34 Harsha Bharathakoti GM 2509 2000
35 Iniyan, P GM 2506 2002
36 Grover, Sahaj GM 2504 1995
37 Arjun Kalyan GM 2503 2002

[lkaufman]

I took another game that GM Perelshteyn won and replace it with "Gaviota-win32-0.84 playing with contempt =65 . Then I decided to run it 100 times with contempt = 65 with the same opening and another 100 times with contempt = 125 the more aggressive contempt = 65 won by 10% over contempt =100

[pgn][Event "knight odds match"]
[Site "Chess.com"]
[Date "2021.11.06"]
[Round "100"]
[White "Dragon-2.5-64bit-avx2"]
[Black "Gaviota-win32-0.84"]
[Result "1-0"]
[BlackElo "2500"]
[WhiteElo "2500"]
[TimeControl "300+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "355"]
[WhiteType "program"]
[BlackType "program"]

1. f4 {[%clk 0:15:10]} d5 {[%clk 0:15:08.5]} 2. Nf3 {[%clk 0:15:19.9]} g6
{[%clk 0:15:14.7]} 3. e3 {[%clk 0:15:29.8]} Bg7 {[%clk 0:15:23.4]} 4. Be2
{[%clk 0:15:39.7]} c5 {[%clk 0:15:26.3]} 5. O-O {[%clk 0:15:45.5]} Nc6
{[%clk 0:15:34.5]} 6. Qe1 {[%clk 0:15:48.4]} Bg4 {[%clk 0:15:35.7]} 7. d3
Qa5 8. Bd2 Qb5 9. h3 Bxf3 10. Bxf3 Qxb2 11. Rb1 Qxc2 12. Rxb7 Nd8 13. Rb3
Nf6 14. Qc1 Qxc1 15. Rxc1 Ne6 16. g4 c4 17. dxc4 Nc5 18. Rb5 Nfe4 19. Rc2
a6 20. Rxc5 Nxc5 21. cxd5 Nd7 22. Rc7 Rd8 23. d6 exd6 24. Bc6 Ke7 25. Ba5
g5 26. Rb7 Bh6 27. f5 Bg7 28. Kg2 Be5 29. Ra7 Bb2 30. Kf3 Ba3 31. Bxd8+
Rxd8 32. Ke4 Bc5 33. Rb7 f6 34. Kd5 Bxe3 35. Rc7 Bd2 36. Rb7 Bc3 37. Rc7
Be1 38. Ra7 Bf2 39. Rb7 Bg3 40. Ra7 Bf2 41. Rb7 Bg3 42. Ra7 Be1 43. Rb7 Bc3
44. Ra7 Bb4 45. Rb7 Bd2 46. Rc7 Be3 47. Rb7 Bf4 48. Rc7 Be5 49. Rb7 Bh2 50.
Ba4 Be5 51. Bc6 Bh2 52. Ba4 h6 53. Bc6 Bf4 54. Rc7 Bd2 55. Ra7 Be3 56. Rc7
Bf2 57. Rb7 Bg3 58. Ra7 Bf2 59. Rb7 Be3 60. Rc7 Bf4 61. Rb7 Bg3 62. Ra7 Bh2
63. Rb7 Bf4 64. Rc7 Bd2 65. Rb7 Bc3 66. Rc7 Be1 67. Rb7 Bc3 68. Rc7 Bb4 69.
Rb7 Ba5 70. Ra7 Be1 71. Rc7 Ba5 72. Rb7 Bd2 73. Ra7 Bb4 74. Rc7 Ba5 75. Rb7
Be1 76. Rc7 Bb4 77. Ra7 Bc5 78. Rb7 Bf2 79. Rc7 Bg1 80. Rb7 Bh2 81. Ba4 Bf4
82. Rc7 Be5 83. Rb7 Bc3 84. Rc7 Be5 85. Rb7 Bf4 86. Rc7 a5 87. Rb7 Be3 88.
Rc7 Bc5 89. Bb5 Bb4 90. Bc6 Bd2 91. Ba4 Bb4 92. Bc6 Bd2 93. Ba4 Be3 94. Bb5
Bb6 95. Rb7 Bc5 96. Rc7 Be3 97. Ba4 Bb6 98. Rb7 Bc5 99. Rc7 Bb6 100. Rb7
Bc5 101. Rc7 Ba3 102. Rb7 Bb4 103. Rc7 Be1 104. Bc6 Bg3 105. Ba4 Be5 106.
Bb5 Bf4 107. Ba4 Be5 108. Bb5 Bf4 109. Ba4 Bd2 110. Bb5 Be1 111. Ba4 Bf2
112. Bb5 Bb6 113. Rb7 Bf2 114. Rc7 Be1 115. Bc6 Bg3 116. Ba4 Bf4 117. Rb7
Be5 118. Rc7 Ba1 119. Bb5 Be5 120. Ba4 Bh2 121. Bb5 Bg1 122. Ba4 Be3 123.
Bb5 Bf2 124. Ba4 Bg3 125. Bc6 a4 126. Bxa4 Be1 127. Ra7 Bd2 128. Bc6 Be3
129. Rb7 Bf4 130. Ra7 Be3 131. Rb7 Bc1 132. Ba4 Bf4 133. Bc6 Be5 134. a4
Bc3 135. Ra7 Bd2 136. a5 Bxa5 137. Rxa5 Nb6+ 138. Kd4 Rc8 139. Ra7+ Kd8
140. Bd5 Rc2 141. Be6 Rh2 142. Rb7 Na4 143. Rf7 Rxh3 144. Rxf6 Nc5 145. Bc4
Ke7 146. Rf7+ Ke8 147. Ra7 Rg3 148. Kd5 Rxg4 149. Bb5+ Kf8 150. Kxd6 Rf4
151. Ke5 Ne4 152. Bc4 Rf2 153. Bd3 Nc5 154. Bb5 Rb2 155. Ra8+ Kg7 156. f6+
Kh7 157. Ra7+ Kh8 158. Kf5 Ne4 159. Kxe4 Rxb5 160. Ra8+ Kh7 161. f7 Rb4+
162. Ke5 Rb5+ 163. Ke6 Rb6+ 164. Ke7 Rb7+ 165. Ke8 Rxf7 166. Kxf7 h5 167.
Kf6 h4 168. Kxg5 Kg7 169. Ra7+ Kf8 170. Kxh4 Ke8 171. Kg5 Kd8 172. Kf6 Kc8
173. Ke6 Kb8 174. Rh7 Ka8 175. Kd6 Kb8 176. Kc6 Ka8 177. Kb6 Kb8 178. Rh8#[/pgn]

This game shows why human GMs do better with odds than most engines. In this game, when the fifty move rule approached, Gaviota just gave away a vital pawn to start the count over; a human would see that this could only help the opponent. Probably only a weakened Komodo/Dragon with a highly negative Contempt setting would settle for the draw rather than play this silly giveaway.

In your opinion when will Komodo Dragon xxx.x be able to beat GM Pereshteyn in a match of 15'+10" without counting draw as a win for Komodo Dragon? Probably you should only play 2 games out of 6 with Knight b1 missing. In 6 months can a GM with similar rating as GM Pereshteyn from India play against Komodo Dragon? The sale of Komodo Dragon xxx.x in India with an estimated chess playing fan of over 30 Million can be a great advertising opportunity.

Indian Players from this: list

34 Harsha Bharathakoti GM 2509 2000
35 Iniyan, P GM 2506 2002
36 Grover, Sahaj GM 2504 1995
37 Arjun Kalyan GM 2503 2002

I think six months is a reasonable estimate for winning such a match; it could happen sooner if we made some breakthru in terms of posing problems for human players. I have no preference regarding the nationality of the opponent; my only requirement is that the player has a good reputation for honesty and congeniality; I wouldn't want to choose an unknown opponent who doesn't play in online tournaments or have other reasons to be trusted not to cheat. We have never faced an opponent in all these matchesm who I even suspected of cheating, and I want to keep it that way. That's the main reason I tend to pick trusted opponents as opposed to unknown ones. If a GM plays regularly in chess.com events and hasn't been credibly accused of cheating, that would be good enough for me, or if he a streamer with a reputation to uphold.

[Chessqueen]

I found all the games that Komodo Dragon 2.5.1 Lost at Knight Odds with b1 missing vs Human GMs in Chess.com and replayed them after the 7 moves at contempt = 65 it won all of except 2 games that it drew vs Gaviota-win32-0.84 . I know that it is a small sample but it is a good indication that against engines it prefer to attack more than to be in the defensive side of the board.

[lkaufman]

I found all the games that Komodo Dragon 2.5.1 Lost at Knight Odds with b1 missing vs Human GMs in Chess.com and replayed them after the 7 moves at contempt = 65 it won all of except 2 games that it drew vs Gaviota-win32-0.84 . I know that it is a small sample but it is a good indication that against engines it prefer to attack more than to be in the defensive side of the board.

You know how medicines are tested with a control group? You need to run the same games vs. the Contempt 125 that was used before you can draw this conclusion. Maybe all you are showing is that Gaviota .84 is not as good at exploiting knight odds as a human GM. What time limit did you use? Also, when you say that contempt 65 is more "attacking", it is also more willing to trade, which decreases attack possibilities, so it's not at all clear which value really will attack more.

[Chessqueen]

I found all the games that Komodo Dragon 2.5.1 Lost at Knight Odds with b1 missing vs Human GMs in Chess.com and replayed them after the 7 moves at contempt = 65 it won all of except 2 games that it drew vs Gaviota-win32-0.84 . I know that it is a small sample but it is a good indication that against engines it prefer to attack more than to be in the defensive side of the board.

You know how medicines are tested with a control group? You need to run the same games vs. the Contempt 125 that was used before you can draw this conclusion. Maybe all you are showing is that Gaviota .84 is not as good at exploiting knight odds as a human GM. What time limit did you use? Also, when you say that contempt 65 is more "attacking", it is also more willing to trade, which decreases attack possibilities, so it's not at all clear which value really will attack more.

I did run the same games and it lost the 2 games that contempt = 65 drew, plus it also lost 3 more games total 5 games were lost. It is a small sample but contempt = 65 performed better than contempt = 125. The T/C was 3'+2" for both engines.

[lkaufman]

I found all the games that Komodo Dragon 2.5.1 Lost at Knight Odds with b1 missing vs Human GMs in Chess.com and replayed them after the 7 moves at contempt = 65 it won all of except 2 games that it drew vs Gaviota-win32-0.84 . I know that it is a small sample but it is a good indication that against engines it prefer to attack more than to be in the defensive side of the board.

You know how medicines are tested with a control group? You need to run the same games vs. the Contempt 125 that was used before you can draw this conclusion. Maybe all you are showing is that Gaviota .84 is not as good at exploiting knight odds as a human GM. What time limit did you use? Also, when you say that contempt 65 is more "attacking", it is also more willing to trade, which decreases attack possibilities, so it's not at all clear which value really will attack more.

I did run the same games and it lost the 2 games that contempt = 65 drew, plus it also lost 3 more games total 5 games were lost. It is a small sample but contempt = 65 performed better than contempt = 125. The T/C was 3'+2" for both engines.

I did try 1000 game runs with Contempt 65 and 125 vs Dragon 2.5.1 Skill 27 Contempt -125 with knight odds (ChrisW book), and 125 did better, but only by 12 elo, not a dramatic difference. It may be that for unknown reasons lower Contempt works better against Gaviota and higher works better against Dragon Skill levels. Perhaps I'll test that later. Even if I confirm your result vs. Gaviota, it won't tell us which is better vs. human GM unless the result is dramatic.

[Chessqueen]

I found all the games that Komodo Dragon 2.5.1 Lost at Knight Odds with b1 missing vs Human GMs in Chess.com and replayed them after the 7 moves at contempt = 65 it won all of except 2 games that it drew vs Gaviota-win32-0.84 . I know that it is a small sample but it is a good indication that against engines it prefer to attack more than to be in the defensive side of the board.

You know how medicines are tested with a control group? You need to run the same games vs. the Contempt 125 that was used before you can draw this conclusion. Maybe all you are showing is that Gaviota .84 is not as good at exploiting knight odds as a human GM. What time limit did you use? Also, when you say that contempt 65 is more "attacking", it is also more willing to trade, which decreases attack possibilities, so it's not at all clear which value really will attack more.

I did run the same games and it lost the 2 games that contempt = 65 drew, plus it also lost 3 more games total 5 games were lost. It is a small sample but contempt = 65 performed better than contempt = 125. The T/C was 3'+2" for both engines.

I did try 1000 game runs with Contempt 65 and 125 vs Dragon 2.5.1 Skill 27 Contempt -125 with knight odds (ChrisW book), and 125 did better, but only by 12 elo, not a dramatic difference. It may be that for unknown reasons lower Contempt works better against Gaviota and higher works better against Dragon Skill levels. Perhaps I'll test that later. Even if I confirm your result vs. Gaviota, it won't tell us which is better vs. human GM unless the result is dramatic.

That is true Komodo Dragon vs Gaviota might yield better result than against Dragon Skill levels, but Humans GMs are too cautious when they see possible attacking chances 7 or 8 moves ahead and immediately switch gear knowing that bringing their extra knight to be traded will be a tough choice for Dragon to defeat them if they are extra careful playing. What seems to be obvious is that Human GMs knows better when to trade or force a trade as soon as they see an opportunity whereas probably Gaviota only see that Dragon is coming after her and try to keep the extra Knight at all cost.

Note: Since the Elo difference is only 12 based on your test, it would not hurt for the next human GM vs Komodo Dragon to have Dragon playing half of the games at Contempt = 65 and the other half at contempt = 125, and also only play 2 games with b1 Knight missing and 4 games with g1 Knight missing. I noticed that GM Perelshteyn was not too comfortable in the last game and he even said that playing an attacking Engine like Komodo is very unpleasant. Engines are not affected by the way their opponents play weather aggressive or defensive, but us human have to deal with the fact that specially when a higher rated player weather human or engine start attacking you it create a negative Psychological effect on your mind and immediately we human get nervous and the chances of blundering is much higher than when the engine play defensive moves.

[lkaufman]

I found all the games that Komodo Dragon 2.5.1 Lost at Knight Odds with b1 missing vs Human GMs in Chess.com and replayed them after the 7 moves at contempt = 65 it won all of except 2 games that it drew vs Gaviota-win32-0.84 . I know that it is a small sample but it is a good indication that against engines it prefer to attack more than to be in the defensive side of the board.

You know how medicines are tested with a control group? You need to run the same games vs. the Contempt 125 that was used before you can draw this conclusion. Maybe all you are showing is that Gaviota .84 is not as good at exploiting knight odds as a human GM. What time limit did you use? Also, when you say that contempt 65 is more "attacking", it is also more willing to trade, which decreases attack possibilities, so it's not at all clear which value really will attack more.

I did run the same games and it lost the 2 games that contempt = 65 drew, plus it also lost 3 more games total 5 games were lost. It is a small sample but contempt = 65 performed better than contempt = 125. The T/C was 3'+2" for both engines.

I did try 1000 game runs with Contempt 65 and 125 vs Dragon 2.5.1 Skill 27 Contempt -125 with knight odds (ChrisW book), and 125 did better, but only by 12 elo, not a dramatic difference. It may be that for unknown reasons lower Contempt works better against Gaviota and higher works better against Dragon Skill levels. Perhaps I'll test that later. Even if I confirm your result vs. Gaviota, it won't tell us which is better vs. human GM unless the result is dramatic.

That is true Komodo Dragon vs Gaviota might yield better result than against Dragon Skill levels, but Humans GMs are too cautious when they see possible attacking chances 7 or 8 moves ahead and immediately switch gear knowing that bringing their extra knight to be traded will be a tough choice for Dragon to defeat them if they are extra careful playing. What seems to be obvious is that Human GMs knows better when to trade or force a trade as soon as they see an opportunity whereas probably Gaviota only see that Dragon is coming after her and try to keep the extra Knight at all cost.

Note: Since the Elo difference is only 12 based on your test, it would not hurt for the next human GM vs Komodo Dragon to have Dragon playing half of the games at Contempt = 65 and the other half at contempt = 125, and also only play 2 games with b1 Knight missing and 4 games with g1 Knight missing. I noticed that GM Perelshteyn was not too comfortable in the last game and he even said that playing an attacking Engine like Komodo is very unpleasant. Engines are not affected by the way their opponents play weather aggressive or defensive, but us human have to deal with the fact that specially when a higher rated player weather human or engine start attacking you it create a negative Psychological effect on your mind and immediately we human get nervous and the chances of blundering is much higher than when the engine play defensive moves.

Dragon really likes to attack when giving knight odds, regardless of Contempt setting (within reason), but it comes down to whether the opening allows for attacking chances. With 1e4 at g1 odds this is usually possible; I thought it would also be so with 1.f4 at b1 odds but it seems that against proper play by Black that just doesn't happen. Maybe there's some opening at b1 odds that sets more problems for Black, but I haven't found it yet. Maybe by the next event we'll discover something that makes a real difference in these matches.

[lkaufman]

I found all the games that Komodo Dragon 2.5.1 Lost at Knight Odds with b1 missing vs Human GMs in Chess.com and replayed them after the 7 moves at contempt = 65 it won all of except 2 games that it drew vs Gaviota-win32-0.84 . I know that it is a small sample but it is a good indication that against engines it prefer to attack more than to be in the defensive side of the board.

You know how medicines are tested with a control group? You need to run the same games vs. the Contempt 125 that was used before you can draw this conclusion. Maybe all you are showing is that Gaviota .84 is not as good at exploiting knight odds as a human GM. What time limit did you use? Also, when you say that contempt 65 is more "attacking", it is also more willing to trade, which decreases attack possibilities, so it's not at all clear which value really will attack more.

I did run the same games and it lost the 2 games that contempt = 65 drew, plus it also lost 3 more games total 5 games were lost. It is a small sample but contempt = 65 performed better than contempt = 125. The T/C was 3'+2" for both engines.

I did try 1000 game runs with Contempt 65 and 125 vs Dragon 2.5.1 Skill 27 Contempt -125 with knight odds (ChrisW book), and 125 did better, but only by 12 elo, not a dramatic difference. It may be that for unknown reasons lower Contempt works better against Gaviota and higher works better against Dragon Skill levels. Perhaps I'll test that later. Even if I confirm your result vs. Gaviota, it won't tell us which is better vs. human GM unless the result is dramatic.

That is true Komodo Dragon vs Gaviota might yield better result than against Dragon Skill levels, but Humans GMs are too cautious when they see possible attacking chances 7 or 8 moves ahead and immediately switch gear knowing that bringing their extra knight to be traded will be a tough choice for Dragon to defeat them if they are extra careful playing. What seems to be obvious is that Human GMs knows better when to trade or force a trade as soon as they see an opportunity whereas probably Gaviota only see that Dragon is coming after her and try to keep the extra Knight at all cost.

Note: Since the Elo difference is only 12 based on your test, it would not hurt for the next human GM vs Komodo Dragon to have Dragon playing half of the games at Contempt = 65 and the other half at contempt = 125, and also only play 2 games with b1 Knight missing and 4 games with g1 Knight missing. I noticed that GM Perelshteyn was not too comfortable in the last game and he even said that playing an attacking Engine like Komodo is very unpleasant. Engines are not affected by the way their opponents play weather aggressive or defensive, but us human have to deal with the fact that specially when a higher rated player weather human or engine start attacking you it create a negative Psychological effect on your mind and immediately we human get nervous and the chances of blundering is much higher than when the engine play defensive moves.

In Blitz (2' + 1", CCRL blitz tc), Gaviota 0.84 32 bit loses rather badly with knight odds to Dragon in my testing, regardless of Contempt setting, well over a 2 to 1 ratio. But if I move up to Gaviota 0.86 64 bit, it is coming out clearly ahead (62 elo) so this might be a suitable simulation of a human GM playing Rapid at knight odds. Gaviot 0.86 64 bit happens to have almost identical CCRL ratings (rapid and blitz) to Fritz 10, which is the engine that defeated Kramnik by 4 to 2 in classical chess in 2006. It ran on a 2 core machine at 3 GHz, which is probably roughly comparable to one core (thread) on the CCRL reference i7 hardware. Kramnik was 2750 FIDE at the time, and standards have risen, so maybe Fritz 10 on one thread would be about 2800 today in classical chess, somewhat higher in Rapid. But playing at 2' + 1" instead of 15' + 10" should bring its level down to perhaps 2600 FIDE Rapid level. There's still a hundred elo between the 2500 GMs and this elo, but that is perhaps due to the human knowing the handicap situation. Anyway Gaviota 0.86 64 bit playing blitz seems a good simulation of the 2500 human GMs playing Rapid at knight odds.

[Chessqueen]

I replayed this game from GM Perelshteyn Vs Komodo Dragon using the same Opening up to move 7 the first game was using Contempt = 65 T/C 2'+2"


[pgn][Event "knight odds match"]
[Site "Chess.com"]
[Date "2021.11.06"]
[Round "1"]
[White "Dragon-2.5-64bit-avx2"]
[Black "Gaviota-win32-0.84"]
[Result "1-0"]
[BlackElo "2500"]
[WhiteElo "2500"]
[TimeControl "120+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "129"]
[WhiteType "program"]
[BlackType "program"]

1. f4 {[%clk 0:15:10]} d5 {[%clk 0:15:08.5]} 2. Nf3 {[%clk 0:15:19.9]} g6
{[%clk 0:15:14.7]} 3. e3 {[%clk 0:15:29.8]} Bg7 {[%clk 0:15:23.4]} 4. Be2
{[%clk 0:15:39.7]} c5 {[%clk 0:15:26.3]} 5. O-O {[%clk 0:15:45.5]} Nc6
{[%clk 0:15:34.5]} 6. Qe1 {[%clk 0:15:48.4]} Bg4 {[%clk 0:15:35.7]} 7. Rb1
Nf6 8. a3 a5 9. d3 O-O 10. b3 Qc7 11. Qh4 Bd7 12. a4 d4 13. e4 Nb4 14. Bd1
Na2 15. Bd2 Nc3 16. Rc1 Ng4 17. Ng5 h5 18. Bxg4 Bxg4 19. h3 Bd7 20. Rce1
Rae8 21. f5 Be5 22. Nf3 Bf6 23. Bg5 Kg7 24. Rf2 b6 25. e5 Bxg5 26. Qxg5
Bxf5 27. Nh4 e6 28. Rxf5 Nd5 29. Rf6 Nxf6 30. exf6+ Kh7 31. Re5 Qd8 32. Qf4
Rg8 33. Nf3 Rg7 34. Ng5+ Kg8 35. fxg7 Qe7 36. Re2 Kxg7 37. Qe5+ Qf6 38.
Nxf7 Re7 39. Qxf6+ Kxf6 40. Rf2+ Kg7 41. Ng5 e5 42. h4 Rd7 43. Ne6+ Kh8 44.
Rf8+ Kh7 45. Kf2 Rb7 46. Kf3 b5 47. axb5 Rxb5 48. Ke4 a4 49. bxa4 Ra5 50.
Kxe5 Rxa4 51. Kf6 Ra6 52. Rf7+ Kh8 53. Re7 c4 54. Kxg6 Ra8 55. dxc4 d3 56.
cxd3 Rb8 57. Rh7+ Kg8 58. Rd7 Rc8 59. c5 Re8 60. Nf4 Ra8 61. Nxh5 Ra6+ 62.
Nf6+ Rxf6+ 63. Kxf6 Kh8 64. Kg6 Kg8 65. Rd8# *[/pgn]