I toast to LeelaqueenOdds and its creators. I raise my wine glass high and shout: Long live the queen. We have a queen!!

[lkaufman]

Is there a way to download that version of Leela to play on your own hardware? It really is a PIA to try and get games and from reading the latest posts now, it seems that if you do get to play QueenOdds BOT, it is not playing at full power?? Would be nice to have the BOT to play on your own hardware (like a 4090 Ti).

I did notice a few times the BOT plays one move blunders ... sometimes right out of the opening ... and is quickly lost (rare but happens). I am guessing that is because it is sharing hardware resources. I definitely feel it is not playing consistently strong ... sometimes it is crazy strong ... sometimes it feels not so strong.

The queen odds net is available, check on Leela Discord to see how to get it. The knight/rook odds net is private. The hardware sharing resources was more of a problem a couple weeks ago, when demand was six to 7 k games per day, but now it is not much of an issue at 3k or less. Color is a major issue; the net was only trained for White, not Black. When it seems to be playing "not so strong", is that with Leela playing White or playing Black? Blunders are likely due to the net limiting opponent's nodes to 40, the estimated level of a "fair" opponent for it at queen odds. To win at queen odds, it has to assume a rather low level of play by the opponent.

[lkaufman]

Is there a way to download that version of Leela to play on your own hardware? It really is a PIA to try and get games and from reading the latest posts now, it seems that if you do get to play QueenOdds BOT, it is not playing at full power?? Would be nice to have the BOT to play on your own hardware (like a 4090 Ti).

I did notice a few times the BOT plays one move blunders ... sometimes right out of the opening ... and is quickly lost (rare but happens). I am guessing that is because it is sharing hardware resources. I definitely feel it is not playing consistently strong ... sometimes it is crazy strong ... sometimes it feels not so strong.

A statistical survey of the LQO games in March and April at 3'2" was posted, which showed that the break-even LiChess Blitz rating at that time control was 2767 when Leela plays White but only 2607 when Leela plays Black, a 160 elo difference.

[Father]

Is there a way to download that version of Leela to play on your own hardware? It really is a PIA to try and get games and from reading the latest posts now, it seems that if you do get to play QueenOdds BOT, it is not playing at full power?? Would be nice to have the BOT to play on your own hardware (like a 4090 Ti).

I did notice a few times the BOT plays one move blunders ... sometimes right out of the opening ... and is quickly lost (rare but happens). I am guessing that is because it is sharing hardware resources. I definitely feel it is not playing consistently strong ... sometimes it is crazy strong ... sometimes it feels not so strong.

A statistical survey of the LQO games in March and April at 3'2" was posted, which showed that the break-even LiChess Blitz rating at that time control was 2767 when Leela plays White but only 2607 when Leela plays Black, a 160 elo difference.

Mr. Larry Kaufman, good morning. This player looks very strong in bullet. He has an Elo rating of over 3,200!! Today, 3,208!! and is already ranked 22nd in the leaderboard, with only 86 bullet games. He's undoubtedly a good dancer who puts the dancing queen's skills to the test.

[Father]

Is there a way to download that version of Leela to play on your own hardware? It really is a PIA to try and get games and from reading the latest posts now, it seems that if you do get to play QueenOdds BOT, it is not playing at full power?? Would be nice to have the BOT to play on your own hardware (like a 4090 Ti).

I did notice a few times the BOT plays one move blunders ... sometimes right out of the opening ... and is quickly lost (rare but happens). I am guessing that is because it is sharing hardware resources. I definitely feel it is not playing consistently strong ... sometimes it is crazy strong ... sometimes it feels not so strong.

A statistical survey of the LQO games in March and April at 3'2" was posted, which showed that the break-even LiChess Blitz rating at that time control was 2767 when Leela plays White but only 2607 when Leela plays Black, a 160 elo difference.

Mr. Larry Kaufman, good morning. This player looks very strong in bullet. He has an Elo rating of over 3,200!! Today, 3,208!! and is already ranked 22nd in the leaderboard, with only 86 bullet games. He's undoubtedly a good dancer who puts the dancing queen's skills to the test.

verystablegenius69

[lkaufman]

Is there a way to download that version of Leela to play on your own hardware? It really is a PIA to try and get games and from reading the latest posts now, it seems that if you do get to play QueenOdds BOT, it is not playing at full power?? Would be nice to have the BOT to play on your own hardware (like a 4090 Ti).

I did notice a few times the BOT plays one move blunders ... sometimes right out of the opening ... and is quickly lost (rare but happens). I am guessing that is because it is sharing hardware resources. I definitely feel it is not playing consistently strong ... sometimes it is crazy strong ... sometimes it feels not so strong.

A statistical survey of the LQO games in March and April at 3'2" was posted, which showed that the break-even LiChess Blitz rating at that time control was 2767 when Leela plays White but only 2607 when Leela plays Black, a 160 elo difference.

Mr. Larry Kaufman, good morning. This player looks very strong in bullet. He has an Elo rating of over 3,200!! Today, 3,208!! and is already ranked 22nd in the leaderboard, with only 86 bullet games. He's undoubtedly a good dancer who puts the dancing queen's skills to the test.

verystablegenius69

He might be the highest rated player who has played a lot of bullet games with LQO (Hikaru played only 5), probably a strong GM with a blitz rating over 2800, but his score of 20% is not near the top. The record (Leela White, ten game minimum) appears to be IM "Wateenellende" at almost 35%. Note that top level LiChess bullet ratings are about 70 elo inflated relative to chess.com (the opposite is the case with blitz!), so he might be about 3138 on chess.com bullet, compared to Hikaru usually near 3300 (the top other than players who just play hyperbullet). So perhaps Hikaru would score about 40% as Black based on this. His actual score of 50% taking White 3 out of 5 times is consistent with this.

[Father]

Is there a way to download that version of Leela to play on your own hardware? It really is a PIA to try and get games and from reading the latest posts now, it seems that if you do get to play QueenOdds BOT, it is not playing at full power?? Would be nice to have the BOT to play on your own hardware (like a 4090 Ti).

I did notice a few times the BOT plays one move blunders ... sometimes right out of the opening ... and is quickly lost (rare but happens). I am guessing that is because it is sharing hardware resources. I definitely feel it is not playing consistently strong ... sometimes it is crazy strong ... sometimes it feels not so strong.

A statistical survey of the LQO games in March and April at 3'2" was posted, which showed that the break-even LiChess Blitz rating at that time control was 2767 when Leela plays White but only 2607 when Leela plays Black, a 160 elo difference.

Mr. Larry Kaufman, good morning. This player looks very strong in bullet. He has an Elo rating of over 3,200!! Today, 3,208!! and is already ranked 22nd in the leaderboard, with only 86 bullet games. He's undoubtedly a good dancer who puts the dancing queen's skills to the test.

verystablegenius69

He might be the highest rated player who has played a lot of bullet games with LQO (Hikaru played only 5), probably a strong GM with a blitz rating over 2800, but his score of 20% is not near the top. The record (Leela White, ten game minimum) appears to be IM "Wateenellende" at almost 35%. Note that top level LiChess bullet ratings are about 70 elo inflated relative to chess.com (the opposite is the case with blitz!), so he might be about 3138 on chess.com bullet, compared to Hikaru usually near 3300 (the top other than players who just play hyperbullet). So perhaps Hikaru would score about 40% as Black based on this. His actual score of 50% taking White 3 out of 5 times is consistent with this.

Meditations on LeelaQueenOdds and Elo, “A robot proffesor.!!”
Just a few months ago, the prodigious "robot" had a total of fewer than 10,000 games against humans, while "right now LeelaQueenOdds has close to 273,197 games against humans." Furthermore, we can see that there is a leaderboard tendence of the top 100 humans. It turns out that simply appearing on the table constitutes another qualitative privilege for humans in itself: The human rating on the table starts at 2916 with GM and three-time US champion Joel Benjamin, and the table closes with a human rating (LeelaQueenOdds table) of 2222 (ignoring the probable conversion to Fide Ell)!! "
LeelaQueenOdds human's elo ( except for the "human exceptions" )
LeelaQueenOdds is already passing "collection bill" to her opponents: A player in the range 2900 to 3000 GM Joel Benjamin !!; A human in the range 2800; One in 2700; Six in 2600 !!; 72% higher than 2300 elo; and 100% higher than 2200 elo "LeelaQueenOdds table." That is to say, everything would indicate a tendency in the direction of "the Mastery of the robot" and why not? Perhaps to the same doctorate.

[Father]

Is there a way to download that version of Leela to play on your own hardware? It really is a PIA to try and get games and from reading the latest posts now, it seems that if you do get to play QueenOdds BOT, it is not playing at full power?? Would be nice to have the BOT to play on your own hardware (like a 4090 Ti).

I did notice a few times the BOT plays one move blunders ... sometimes right out of the opening ... and is quickly lost (rare but happens). I am guessing that is because it is sharing hardware resources. I definitely feel it is not playing consistently strong ... sometimes it is crazy strong ... sometimes it feels not so strong.

A statistical survey of the LQO games in March and April at 3'2" was posted, which showed that the break-even LiChess Blitz rating at that time control was 2767 when Leela plays White but only 2607 when Leela plays Black, a 160 elo difference.

Mr. Larry Kaufman, good morning. This player looks very strong in bullet. He has an Elo rating of over 3,200!! Today, 3,208!! and is already ranked 22nd in the leaderboard, with only 86 bullet games. He's undoubtedly a good dancer who puts the dancing queen's skills to the test.

verystablegenius69

He might be the highest rated player who has played a lot of bullet games with LQO (Hikaru played only 5), probably a strong GM with a blitz rating over 2800, but his score of 20% is not near the top. The record (Leela White, ten game minimum) appears to be IM "Wateenellende" at almost 35%. Note that top level LiChess bullet ratings are about 70 elo inflated relative to chess.com (the opposite is the case with blitz!), so he might be about 3138 on chess.com bullet, compared to Hikaru usually near 3300 (the top other than players who just play hyperbullet). So perhaps Hikaru would score about 40% as Black based on this. His actual score of 50% taking White 3 out of 5 times is consistent with this.

Meditations on LeelaQueenOdds and Elo, “A robot proffesor.!!”
Just a few months ago, the prodigious "robot" had a total of fewer than 10,000 games against humans, while "right now LeelaQueenOdds has close to 273,197 games against humans." Furthermore, we can see that there is a leaderboard tendence of the top 100 humans. It turns out that simply appearing on the table constitutes another qualitative privilege for humans in itself: The human rating on the table starts at 2916 with GM and three-time US champion Joel Benjamin, and the table closes with a human rating (LeelaQueenOdds table) of 2222 (ignoring the probable conversion to Fide Ell)!! "
LeelaQueenOdds human's elo ( except for the "human exceptions" )
LeelaQueenOdds is already passing "collection bill" to her opponents: A player in the range 2900 to 3000 GM Joel Benjamin !!; A human in the range 2800; One in 2700; Six in 2600 !!; 72% higher than 2300 elo; and 100% higher than 2200 elo "LeelaQueenOdds table." That is to say, everything would indicate a tendency in the direction of "the Mastery of the robot" and why not? Perhaps to the same doctorate.

Easter Meditations.

I've been thinking about the games I play with a one-minute time control... many of these games it's clear to me that I have absolutely winning positions... specifically, "if I had enough time" I could implement winning techniques... but "unfortunately I don't have enough time for it." So it seems I can only run and run to escape the carnivorous tiger that wants to devour me, until I tire the feline out. My mind understands that everything is set to win, but I don't have the time to carry out the plan. So, my pragmatic path is to resign myself to achieving a simple draw. Even if the simple draw occurs after 182 moves in just one minute. My arms, my hands, and my fingers are slower than my thinking. If I were only pure spirit, I firmly believe I would defeat this cybernetic feline, but for now I am more than a spirit, I am also a spirit imprisoned in my body of flesh.

[pgn][Event "Casual bullet game"]
[Site "https://lichess.org/DQ9njf6I"]
[Date "2025.04.22"]
[White "Catecan"]
[Black "LeelaQueenOdds"]
[Result "1/2-1/2"]
[GameId "DQ9njf6I"]
[UTCDate "2025.04.22"]
[UTCTime "03:06:12"]
[WhiteElo "2030"]
[BlackElo "2000"]
[BlackTitle "BOT"]
[Variant "From Position"]
[TimeControl "60+0"]
[ECO "?"]
[Opening "?"]
[Termination "Normal"]
[FEN "rnb1kbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1"]
[SetUp "1"]
[Annotator "lichess.org"]

1. d4 e6 2. f4 f5 3. e3 g5 4. Nf3 g4 5. Nh4 b6 6. g3 Nc6 7. Bg2 Ba6 8. Bf1 Bb7 9. Bg2 Ba6 10. Bf1 Bb7 11. Bg2 O-O-O 12. O-O Ba6 13. Bd2 Be7 14. Be1 Nf6 15. Rf2 Rhg8 16. c3 Ne4 17. Rc2 Bxh4 18. gxh4 g3 19. Bxe4 gxh2+ 20. Kxh2 fxe4 21. Bg3 Rg7 22. Rg2 Rdg8 23. Qg1 Ne7 24. Bf2 Rxg2+ 25. Qxg2 Rxg2+ 26. Kxg2 Bd3 27. Nd2 a5 28. a3 d5 29. Kh2 Nf5 30. Rg1 Kd7 31. Rg5 a4 32. Rxf5 exf5 33. Kg3 Kc6 34. Kh2 Kb5 35. Kg1 Be2 36. Kh2 Kc6 37. Kg3 Bb5 38. Kh2 Be2 39. Kg3 Ba6 40. Kh2 h6 41. Kg3 Kd6 42. Kh2 Kc6 43. Kg3 Kd7 44. Kh2 h5 45. Kg3 Kc6 46. Kh2 Kb7 47. Kg3 Kc6 48. Kh2 Kd7 49. Kg3 Kd6 50. Kh2 Ke7 51. Kg3 Be2 52. Kh2 Kd6 53. Kg3 Kc6 54. Kh2 Kd6 55. Kg3 Ke7 56. Kh2 Kd7 57. Kg3 Kc6 58. Kh2 Kb7 59. Kg3 Ka7 60. Kh2 Ka6 61. Kg3 Kb7 62. Kh2 Ka6 63. Kg3 Ka7 64. Kh2 Kb8 65. Kg3 Kc8 66. Kh2 Kd7 67. Bg3 Kd8 68. Kg1 Kd7 69. Kf2 Bb5 70. Ke1 Kd8 71. Kd1 Kd7 72. Kc1 Kd6 73. Bh2 Kc6 74. Bg3 Kd7 75. Bf2 Kd6 76. Bg3 Kc6 77. Bf2 Kd6 78. Bg3 Kd7 79. Bf2 Ba6 80. Bg3 Kc6 81. Bf2 Kd7 82. Bg3 Bd3 83. Bf2 Kd6 84. Bg3 Bb5 85. Bf2 Kd7 86. Bg3 Ke6 87. Bf2 Be2 88. Bg3 Kd6 89. Bf2 Kc6 90. Bg3 Kb7 91. Bf2 Ka6 92. Bg3 c6 93. Bf2 Ka5 94. Bg3 Ka6 95. Bf2 Ka5 96. Bg3 Kb5 97. Bf2 Bd3 98. Bg3 Be2 99. Bf2 Ka6 100. Bg3 Ka5 101. Bf2 Ka6 102. Bg3 Bd3 103. Bf2 Kb5 104. Bg3 Ka5 105. Bf2 Ka6 106. Bg3 Ka7 107. Bf2 Ka6 108. Bg3 Kb7 109. Bf2 Be2 110. Bg3 Kc7 111. Bf2 Kd6 112. Bg3 Bd3 113. Bf2 Be2 114. Bg3 Kc7 115. Bf2 Kb7 116. Bg3 Bb5 117. Bf2 Kc7 118. Bg3 Kb7 119. Bf2 Kc7 120. Bg3 Kc8 121. Bf2 Kd8 122. Bg1 Kd7 123. Bf2 Kd8 124. Bg1 Ba6 125. Bf2 Kd7 126. Bg1 Kd8 127. Bf2 Kc8 128. Bg1 Kc7 129. Bf2 Kb7 130. Bg1 Kc8 131. Bf2 Kd8 132. Bg1 Ke7 133. Bf2 Kd7 134. Bg1 Kd6 135. Bf2 Ke6 136. Bg1 Kd6 137. Bf2 c5 138. Bg1 Be2 139. Bf2 Ba6 140. Bg1 Be2 141. Bf2 Kc6 142. Bg1 Kc7 143. Bf2 Kd6 144. Bg1 Kc6 145. Bf2 Kd6 146. Bg1 Kc7 147. Bf2 Kc6 148. Bg1 Kd7 149. Bf2 Kc7 150. Bg1 Kd7 151. Bf2 Ke6 152. Bg1 Ke7 153. Bf2 Ke6 154. Bg1 Ke7 155. Bf2 Kd7 156. Bg1 Ke6 157. Bf2 Bd3 158. Bg1 Be2 159. Bf2 Bd3 160. Bg1 Kd7 161. Bf2 Kd6 162. Bg1 Kd7 163. Bf2 Kd6 164. Bg1 Kc6 165. Bf2 Kb5 166. Bg1 Be2 167. Bf2 Bd3 168. Bg1 Kc6 169. Bf2 Kb7 170. Bg1 Ka7 171. Bf2 Be2 172. Bg1 Ka6 173. Bf2 Ka5 174. Bg1 Ka6 175. Bf2 Kb7 176. Bg1 Ka7 177. Bf2 Ka6 178. Bg1 Ka7 179. Bf2 Ka6 180. Bg1 Kb7 181. Bf2 Ka7 182. Bg1 Ka6 { The game is a draw. } 1/2-1/2[pgn][/pgn][/pgn]

[Father]

Easter Meditations.

I've been thinking about the games I play with a one-minute time control... many of these games it's clear to me that I have absolutely winning positions... specifically, "if I had enough time" I could implement winning techniques... but "unfortunately I don't have enough time for it." So it seems I can only run and run to escape the carnivorous tiger that wants to devour me, until I tire the feline out. My mind understands that everything is set to win, but I don't have the time to carry out the plan. So, my pragmatic path is to resign myself to achieving a simple draw. Even if the simple draw occurs after 182 moves in just one minute. My arms, my hands, and my fingers are slower than my thinking. If I were only pure spirit, I firmly believe I would defeat this cybernetic feline, but for now I am more than a spirit, I am also a spirit imprisoned in my body of flesh.

[pgn][Event "Casual bullet game"]
[Site "https://lichess.org/DQ9njf6I"]
[Date "2025.04.22"]
[White "Catecan"]
[Black "LeelaQueenOdds"]
[Result "1/2-1/2"]
[GameId "DQ9njf6I"]
[UTCDate "2025.04.22"]
[UTCTime "03:06:12"]
[WhiteElo "2030"]
[BlackElo "2000"]
[BlackTitle "BOT"]
[Variant "From Position"]
[TimeControl "60+0"]
[ECO "?"]
[Opening "?"]
[Termination "Normal"]
[FEN "rnb1kbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1"]
[SetUp "1"]
[Annotator "lichess.org"]

1. d4 e6 2. f4 f5 3. e3 g5 4. Nf3 g4 5. Nh4 b6 6. g3 Nc6 7. Bg2 Ba6 8. Bf1 Bb7 9. Bg2 Ba6 10. Bf1 Bb7 11. Bg2 O-O-O 12. O-O Ba6 13. Bd2 Be7 14. Be1 Nf6 15. Rf2 Rhg8 16. c3 Ne4 17. Rc2 Bxh4 18. gxh4 g3 19. Bxe4 gxh2+ 20. Kxh2 fxe4 21. Bg3 Rg7 22. Rg2 Rdg8 23. Qg1 Ne7 24. Bf2 Rxg2+ 25. Qxg2 Rxg2+ 26. Kxg2 Bd3 27. Nd2 a5 28. a3 d5 29. Kh2 Nf5 30. Rg1 Kd7 31. Rg5 a4 32. Rxf5 exf5 33. Kg3 Kc6 34. Kh2 Kb5 35. Kg1 Be2 36. Kh2 Kc6 37. Kg3 Bb5 38. Kh2 Be2 39. Kg3 Ba6 40. Kh2 h6 41. Kg3 Kd6 42. Kh2 Kc6 43. Kg3 Kd7 44. Kh2 h5 45. Kg3 Kc6 46. Kh2 Kb7 47. Kg3 Kc6 48. Kh2 Kd7 49. Kg3 Kd6 50. Kh2 Ke7 51. Kg3 Be2 52. Kh2 Kd6 53. Kg3 Kc6 54. Kh2 Kd6 55. Kg3 Ke7 56. Kh2 Kd7 57. Kg3 Kc6 58. Kh2 Kb7 59. Kg3 Ka7 60. Kh2 Ka6 61. Kg3 Kb7 62. Kh2 Ka6 63. Kg3 Ka7 64. Kh2 Kb8 65. Kg3 Kc8 66. Kh2 Kd7 67. Bg3 Kd8 68. Kg1 Kd7 69. Kf2 Bb5 70. Ke1 Kd8 71. Kd1 Kd7 72. Kc1 Kd6 73. Bh2 Kc6 74. Bg3 Kd7 75. Bf2 Kd6 76. Bg3 Kc6 77. Bf2 Kd6 78. Bg3 Kd7 79. Bf2 Ba6 80. Bg3 Kc6 81. Bf2 Kd7 82. Bg3 Bd3 83. Bf2 Kd6 84. Bg3 Bb5 85. Bf2 Kd7 86. Bg3 Ke6 87. Bf2 Be2 88. Bg3 Kd6 89. Bf2 Kc6 90. Bg3 Kb7 91. Bf2 Ka6 92. Bg3 c6 93. Bf2 Ka5 94. Bg3 Ka6 95. Bf2 Ka5 96. Bg3 Kb5 97. Bf2 Bd3 98. Bg3 Be2 99. Bf2 Ka6 100. Bg3 Ka5 101. Bf2 Ka6 102. Bg3 Bd3 103. Bf2 Kb5 104. Bg3 Ka5 105. Bf2 Ka6 106. Bg3 Ka7 107. Bf2 Ka6 108. Bg3 Kb7 109. Bf2 Be2 110. Bg3 Kc7 111. Bf2 Kd6 112. Bg3 Bd3 113. Bf2 Be2 114. Bg3 Kc7 115. Bf2 Kb7 116. Bg3 Bb5 117. Bf2 Kc7 118. Bg3 Kb7 119. Bf2 Kc7 120. Bg3 Kc8 121. Bf2 Kd8 122. Bg1 Kd7 123. Bf2 Kd8 124. Bg1 Ba6 125. Bf2 Kd7 126. Bg1 Kd8 127. Bf2 Kc8 128. Bg1 Kc7 129. Bf2 Kb7 130. Bg1 Kc8 131. Bf2 Kd8 132. Bg1 Ke7 133. Bf2 Kd7 134. Bg1 Kd6 135. Bf2 Ke6 136. Bg1 Kd6 137. Bf2 c5 138. Bg1 Be2 139. Bf2 Ba6 140. Bg1 Be2 141. Bf2 Kc6 142. Bg1 Kc7 143. Bf2 Kd6 144. Bg1 Kc6 145. Bf2 Kd6 146. Bg1 Kc7 147. Bf2 Kc6 148. Bg1 Kd7 149. Bf2 Kc7 150. Bg1 Kd7 151. Bf2 Ke6 152. Bg1 Ke7 153. Bf2 Ke6 154. Bg1 Ke7 155. Bf2 Kd7 156. Bg1 Ke6 157. Bf2 Bd3 158. Bg1 Be2 159. Bf2 Bd3 160. Bg1 Kd7 161. Bf2 Kd6 162. Bg1 Kd7 163. Bf2 Kd6 164. Bg1 Kc6 165. Bf2 Kb5 166. Bg1 Be2 167. Bf2 Bd3 168. Bg1 Kc6 169. Bf2 Kb7 170. Bg1 Ka7 171. Bf2 Be2 172. Bg1 Ka6 173. Bf2 Ka5 174. Bg1 Ka6 175. Bf2 Kb7 176. Bg1 Ka7 177. Bf2 Ka6 178. Bg1 Ka7 179. Bf2 Ka6 180. Bg1 Kb7 181. Bf2 Ka7 182. Bg1 Ka6 { The game is a draw. } 1/2-1/2[/pgn]

[OneTrickPony]

Blunders are likely due to the net limiting opponent's nodes to 40, the estimated level of a "fair" opponent for it at queen odds. To win at queen odds, it has to assume a rather low level of play by the opponent.

Hi Larry, I would like to ask if this limitation about opponent's strength was used in training or is used during play in search (by modifying the searching algorithm) or maybe both?

[lkaufman]

Blunders are likely due to the net limiting opponent's nodes to 40, the estimated level of a "fair" opponent for it at queen odds. To win at queen odds, it has to assume a rather low level of play by the opponent.

Hi Larry, I would like to ask if this limitation about opponent's strength was used in training or is used during play in search (by modifying the searching algorithm) or maybe both?

It is used during actual bot play online. The training of the queen odds net was done before the invention of "search Contempt" (the limiting of opponent's nodes). Recent tests suggest that it isn't useful for training, only for actual play vs humans (or simulated humans).