The Real #2 Chess Engine

[Mark Tang]

I happened to come across a ranking list that I had never seen before, which is called ICCRL (https://www.iccrl.org/rankings). And on the rank, I noticed an engine called Gillespie is #2, stronger than Torch and is only 1 elo behind Stockfish 17.1. Checking its official website(https://gillespie.alexdukas.dev/), The author Alexander Dukas wrote Gillespie employs a Mixture of Experts (MoE) NNUE architecture with five specialized neural networks, each trained on billions of positions from specific game phases,(seems like Cerberus on SPCC)which interests me a lot. Can anyone get this chess engine from the author and send it to me please？

[AndrewGrant]

I happened to come across a ranking list that I had never seen before, which is called ICCRL (https://www.iccrl.org/rankings). And on the rank, I noticed an engine called Gillespie is #2, stronger than Torch and is only 1 elo behind Stockfish 17.1. Checking its official website(https://gillespie.alexdukas.dev/), The author Alexander Dukas wrote Gillespie employs a Mixture of Experts (MoE) NNUE architecture with five specialized neural networks, each trained on billions of positions from specific game phases,(seems like Cerberus on SPCC)which interests me a lot. Can anyone get this chess engine from the author and send it to me please？

I imagine the list is fake, because I have not provided Torch to any such list.

So it probably only exists to promote "Gillespie". Which itself sounds like its not an actual original work, just running 5 engines at once, and at that point you're basically just running Stockfish with extra steps.

A shame though, the site has a nice layout to it.

[AndrewGrant]

Yeah so I'm going to go out on a limb here and say that the "author" of the ICCRL rating list is the same as the "author" of the Gillespie chess engine, that the Gillespie chess engine does not actually exist, the ICCRL games are fabricated, and that the poster in this thread (his first post) is the same individual running both. And ask that the mods delete this.

The basis:

This ICCRL "rating list", which no one has ever heard of, has a sudden new engine called "Gillespie".
If you go to the website for that (https://gillespie.alexdukas.dev/about), there are some PGN drop downs.
One of the PGNs there is this one:

Code: Select all

[Event "CCRL Blitz"]
[Site "CCRL"]
[Date "2023.12.25"]
[Round "211.10.997"]
[White "Gillespie v1.0.0 64-bit"]
[Black "Torch v1 64-bit"]
[Result "1/2-1/2"]
[ECO "D10"]
[Opening "QGD Slav defence"]
[PlyCount "279"]
[WhiteElo "3656"]
[BlackElo "3759"]

1. d4 d5 2. c4 c6 3. Nc3 Nf6 4. Bg5 Nbd7 5. cxd5 cxd5 6. Qc2 e6 7. Nf3 h6 8.
Bh4 a6 9. e3 b5 10. Bd3 Bd6 11. a4 b4 12. Ne2 a5 13. Bxf6 gxf6 14. O-O Qb6 15.
Rfc1 Ke7 16. g3 Ba6 17. Bxa6 Qxa6 18. Qd1 Rhc8 19. Rxc8 Rxc8 20. Rc1 Rxc1 21.
Nxc1 Qc6 22. b3 Qc3 23. Ne2 Qc8 24. Ne1 f5 25. Qc2 Qb7 26. Kg2 Nf6 27. Qc1 Ne4
28. Nd3 Qc7 29. Qxc7+ Bxc7 30. Ng1 Nd2 31. Nc1 f6 32. Nf3 Nxf3 33. Kxf3 e5 34.
Ne2 Ke6 35. Kg2 Bb6 36. f4 Bc7 37. Ng1 e4 38. Ne2 h5 39. Kf1 Ke7 40. Kg1 Bd8
41. Nc1 Kf7 42. h3 Bc7 43. Kg2 Ke6 44. Kf2 Bb6 45. Kf1 Bd8 46. Ke1 Kd7 47. Kd1
Be7 48. Ne2 Bd8 49. Ke1 Ke7 50. Kf1 Bc7 51. Ng1 Bd8 52. Ke2 Kd7 53. Kf2 Bb6 54.
Ne2 Bc7 55. Kg2 Ke6 56. Kh1 Bd8 57. Kh2 Kf7 58. Nc1 Ke6 59. Kg1 Kd7 60. Kf2 Bb6
61. Kf1 Bd8 62. Ne2 Ke7 63. Kg2 Ke6 64. Kf2 Bb6 65. Ke1 Kd7 66. Kd2 Ke7 67. Kd1
Bd8 68. Ng1 Ke6 69. Kc1 Kd7 70. Kb1 Bb6 71. Ne2 Kc8 72. Nc1 Kc7 73. Kb2 Kd8 74.
Ne2 Kd7 75. Ng1 Ke7 76. Kc1 Bc7 77. Kd1 Bd8 78. Kd2 Kd7 79. Ne2 Bb6 80. Nc1 Bc7
81. Na2 Ke6 82. Ke1 Bd8 83. Kf2 Kd7 84. Kg2 Bb6 85. Kg1 Ke7 86. Kf2 Bd8 87. Kg2
Kd7 88. Nc1 Ke6 89. Kh1 Ke7 90. h4 Bb6 91. Ne2 Ke6 92. Kg1 Bd8 93. Kf2 Kf7 94.
Ng1 Ke7 95. Ke1 Bb6 96. Ne2 Bc7 97. Kf2 Bb6 98. Kg2 Bd8 99. Kf1 Kd7 100. Ke1
Ke6 101. Ng1 Kd7 102. Kd2 Bb6 103. Kd1 Ke7 104. Ke1 Ke6 105. Nh3 Kf7 106. Kd2
Ke7 107. Nf2 Kf7 108. Nh1 Kg6 109. Ke1 Ba7 110. Kf2 Kg7 111. Ke2 Kf7 112. Nf2
Ke7 113. Kd2 Bb6 114. Nh1 Ke6 115. Ke2 Kd7 116. Nf2 Ke7 117. Ke1 Ke6 118. Nh1
Ba7 119. Kf1 Kf7 120. Ke2 Kg6 121. Kd1 Bb6 122. Kd2 Bd8 123. Ke2 Bc7 124. Kf1
Bd8 125. Kf2 Bb6 126. Kg2 Kg7 127. Nf2 Ba7 128. Nh3 Kh7 129. Kf2 Bb6 130. Ke2
Bd8 131. Kd1 Bb6 132. Kd2 Kg6 133. Ke2 Kh7 134. Ng1 Kh8 135. Ke1 Kh7 136. Kf1
Bc7 137. Kg2 Bb6 138. Kh2 Kh6 139. Kh3 Bd8 140. Ne2 1/2-1/2

This PGN, interestingly, states that it comes from CCRL. Not ICCRL, but CCRL.
But CCRL has never tested "Gillespie". Furthermore, how would this "Gillespie" author have access to Torch-v1, a private engine?
Well if we download CCRL's archive for Torch-v1, we very quickly find this PGN, which is identical, but with the name Obsidian instead.

Code: Select all

[Event "CCRL Blitz"]
[Site "CCRL"]
[Date "2023.12.25"]
[Round "211.10.997"]
[White "Obsidian 9.0 64-bit"]
[Black "Torch v1 64-bit"]
[Result "1/2-1/2"]
[ECO "D10"]
[Opening "QGD Slav defence"]
[PlyCount "279"]
[WhiteElo "3653"]
[BlackElo "3756"]

1. d4 d5 2. c4 c6 3. Nc3 Nf6 4. Bg5 Nbd7 5. cxd5 cxd5 6. Qc2 e6 7. Nf3 h6 8.
Bh4 a6 9. e3 b5 10. Bd3 Bd6 11. a4 b4 12. Ne2 a5 13. Bxf6 gxf6 14. O-O Qb6 15.
Rfc1 Ke7 16. g3 Ba6 17. Bxa6 Qxa6 18. Qd1 Rhc8 19. Rxc8 Rxc8 20. Rc1 Rxc1 21.
Nxc1 Qc6 22. b3 Qc3 23. Ne2 Qc8 24. Ne1 f5 25. Qc2 Qb7 26. Kg2 Nf6 27. Qc1 Ne4
28. Nd3 Qc7 29. Qxc7+ Bxc7 30. Ng1 Nd2 31. Nc1 f6 32. Nf3 Nxf3 33. Kxf3 e5 34.
Ne2 Ke6 35. Kg2 Bb6 36. f4 Bc7 37. Ng1 e4 38. Ne2 h5 39. Kf1 Ke7 40. Kg1 Bd8
41. Nc1 Kf7 42. h3 Bc7 43. Kg2 Ke6 44. Kf2 Bb6 45. Kf1 Bd8 46. Ke1 Kd7 47. Kd1
Be7 48. Ne2 Bd8 49. Ke1 Ke7 50. Kf1 Bc7 51. Ng1 Bd8 52. Ke2 Kd7 53. Kf2 Bb6 54.
Ne2 Bc7 55. Kg2 Ke6 56. Kh1 Bd8 57. Kh2 Kf7 58. Nc1 Ke6 59. Kg1 Kd7 60. Kf2 Bb6
61. Kf1 Bd8 62. Ne2 Ke7 63. Kg2 Ke6 64. Kf2 Bb6 65. Ke1 Kd7 66. Kd2 Ke7 67. Kd1
Bd8 68. Ng1 Ke6 69. Kc1 Kd7 70. Kb1 Bb6 71. Ne2 Kc8 72. Nc1 Kc7 73. Kb2 Kd8 74.
Ne2 Kd7 75. Ng1 Ke7 76. Kc1 Bc7 77. Kd1 Bd8 78. Kd2 Kd7 79. Ne2 Bb6 80. Nc1 Bc7
81. Na2 Ke6 82. Ke1 Bd8 83. Kf2 Kd7 84. Kg2 Bb6 85. Kg1 Ke7 86. Kf2 Bd8 87. Kg2
Kd7 88. Nc1 Ke6 89. Kh1 Ke7 90. h4 Bb6 91. Ne2 Ke6 92. Kg1 Bd8 93. Kf2 Kf7 94.
Ng1 Ke7 95. Ke1 Bb6 96. Ne2 Bc7 97. Kf2 Bb6 98. Kg2 Bd8 99. Kf1 Kd7 100. Ke1
Ke6 101. Ng1 Kd7 102. Kd2 Bb6 103. Kd1 Ke7 104. Ke1 Ke6 105. Nh3 Kf7 106. Kd2
Ke7 107. Nf2 Kf7 108. Nh1 Kg6 109. Ke1 Ba7 110. Kf2 Kg7 111. Ke2 Kf7 112. Nf2
Ke7 113. Kd2 Bb6 114. Nh1 Ke6 115. Ke2 Kd7 116. Nf2 Ke7 117. Ke1 Ke6 118. Nh1
Ba7 119. Kf1 Kf7 120. Ke2 Kg6 121. Kd1 Bb6 122. Kd2 Bd8 123. Ke2 Bc7 124. Kf1
Bd8 125. Kf2 Bb6 126. Kg2 Kg7 127. Nf2 Ba7 128. Nh3 Kh7 129. Kf2 Bb6 130. Ke2
Bd8 131. Kd1 Bb6 132. Kd2 Kg6 133. Ke2 Kh7 134. Ng1 Kh8 135. Ke1 Kh7 136. Kf1
Bc7 137. Kg2 Bb6 138. Kh2 Kh6 139. Kh3 Bd8 140. Ne2 1/2-1/2

So its established that at least this "Gillespie" individual is willing to lie about things, and slap his own name onto some other source's PGN.

[Mark Tang]

Yes, Andrew. Your evidence is pretty convincing.But via this, I still believe that the algorithm of multiple specialized networks for different parts of game is an innovative idea.

[reflectionofpower]

I happened to come across a ranking list that I had never seen before, which is called ICCRL (https://www.iccrl.org/rankings). And on the rank, I noticed an engine called Gillespie is #2, stronger than Torch and is only 1 elo behind Stockfish 17.1. Checking its official website(https://gillespie.alexdukas.dev/), The author Alexander Dukas wrote Gillespie employs a Mixture of Experts (MoE) NNUE architecture with five specialized neural networks, each trained on billions of positions from specific game phases,(seems like Cerberus on SPCC)which interests me a lot. Can anyone get this chess engine from the author and send it to me please？

I imagine the list is fake, because I have not provided Torch to any such list.

So it probably only exists to promote "Gillespie". Which itself sounds like its not an actual original work, just running 5 engines at once, and at that point you're basically just running Stockfish with extra steps.

A shame though, the site has a nice layout to it.

It must be real. It has the word "International" in it. Note to that the name is very similar to CCRL chess rating site, which is genuinely legitimate.

[Mark Tang]

But looking back the past competition on that ranking site, you will find Torch V4 in a 2014 competition, which is absolutely absurd!!!

[Rebel]

Good catch Andy, did a double game check, gillespie-games-dataset.pgn vs Obsidian_9_0_64-bit.pgn from CCRL, both 2051 games and 2051 doubles. A fraud + no executable.

The things that people do.

[chrisw]

Yeah so I'm going to go out on a limb here and say that the "author" of the ICCRL rating list is the same as the "author" of the Gillespie chess engine, that the Gillespie chess engine does not actually exist, the ICCRL games are fabricated, and that the poster in this thread (his first post) is the same individual running both. And ask that the mods delete this.

The basis:

This ICCRL "rating list", which no one has ever heard of, has a sudden new engine called "Gillespie".
If you go to the website for that (https://gillespie.alexdukas.dev/about), there are some PGN drop downs.
One of the PGNs there is this one:

Code: Select all

[Event "CCRL Blitz"]
[Site "CCRL"]
[Date "2023.12.25"]
[Round "211.10.997"]
[White "Gillespie v1.0.0 64-bit"]
[Black "Torch v1 64-bit"]
[Result "1/2-1/2"]
[ECO "D10"]
[Opening "QGD Slav defence"]
[PlyCount "279"]
[WhiteElo "3656"]
[BlackElo "3759"]

1. d4 d5 2. c4 c6 3. Nc3 Nf6 4. Bg5 Nbd7 5. cxd5 cxd5 6. Qc2 e6 7. Nf3 h6 8.
Bh4 a6 9. e3 b5 10. Bd3 Bd6 11. a4 b4 12. Ne2 a5 13. Bxf6 gxf6 14. O-O Qb6 15.
Rfc1 Ke7 16. g3 Ba6 17. Bxa6 Qxa6 18. Qd1 Rhc8 19. Rxc8 Rxc8 20. Rc1 Rxc1 21.
Nxc1 Qc6 22. b3 Qc3 23. Ne2 Qc8 24. Ne1 f5 25. Qc2 Qb7 26. Kg2 Nf6 27. Qc1 Ne4
28. Nd3 Qc7 29. Qxc7+ Bxc7 30. Ng1 Nd2 31. Nc1 f6 32. Nf3 Nxf3 33. Kxf3 e5 34.
Ne2 Ke6 35. Kg2 Bb6 36. f4 Bc7 37. Ng1 e4 38. Ne2 h5 39. Kf1 Ke7 40. Kg1 Bd8
41. Nc1 Kf7 42. h3 Bc7 43. Kg2 Ke6 44. Kf2 Bb6 45. Kf1 Bd8 46. Ke1 Kd7 47. Kd1
Be7 48. Ne2 Bd8 49. Ke1 Ke7 50. Kf1 Bc7 51. Ng1 Bd8 52. Ke2 Kd7 53. Kf2 Bb6 54.
Ne2 Bc7 55. Kg2 Ke6 56. Kh1 Bd8 57. Kh2 Kf7 58. Nc1 Ke6 59. Kg1 Kd7 60. Kf2 Bb6
61. Kf1 Bd8 62. Ne2 Ke7 63. Kg2 Ke6 64. Kf2 Bb6 65. Ke1 Kd7 66. Kd2 Ke7 67. Kd1
Bd8 68. Ng1 Ke6 69. Kc1 Kd7 70. Kb1 Bb6 71. Ne2 Kc8 72. Nc1 Kc7 73. Kb2 Kd8 74.
Ne2 Kd7 75. Ng1 Ke7 76. Kc1 Bc7 77. Kd1 Bd8 78. Kd2 Kd7 79. Ne2 Bb6 80. Nc1 Bc7
81. Na2 Ke6 82. Ke1 Bd8 83. Kf2 Kd7 84. Kg2 Bb6 85. Kg1 Ke7 86. Kf2 Bd8 87. Kg2
Kd7 88. Nc1 Ke6 89. Kh1 Ke7 90. h4 Bb6 91. Ne2 Ke6 92. Kg1 Bd8 93. Kf2 Kf7 94.
Ng1 Ke7 95. Ke1 Bb6 96. Ne2 Bc7 97. Kf2 Bb6 98. Kg2 Bd8 99. Kf1 Kd7 100. Ke1
Ke6 101. Ng1 Kd7 102. Kd2 Bb6 103. Kd1 Ke7 104. Ke1 Ke6 105. Nh3 Kf7 106. Kd2
Ke7 107. Nf2 Kf7 108. Nh1 Kg6 109. Ke1 Ba7 110. Kf2 Kg7 111. Ke2 Kf7 112. Nf2
Ke7 113. Kd2 Bb6 114. Nh1 Ke6 115. Ke2 Kd7 116. Nf2 Ke7 117. Ke1 Ke6 118. Nh1
Ba7 119. Kf1 Kf7 120. Ke2 Kg6 121. Kd1 Bb6 122. Kd2 Bd8 123. Ke2 Bc7 124. Kf1
Bd8 125. Kf2 Bb6 126. Kg2 Kg7 127. Nf2 Ba7 128. Nh3 Kh7 129. Kf2 Bb6 130. Ke2
Bd8 131. Kd1 Bb6 132. Kd2 Kg6 133. Ke2 Kh7 134. Ng1 Kh8 135. Ke1 Kh7 136. Kf1
Bc7 137. Kg2 Bb6 138. Kh2 Kh6 139. Kh3 Bd8 140. Ne2 1/2-1/2

This PGN, interestingly, states that it comes from CCRL. Not ICCRL, but CCRL.
But CCRL has never tested "Gillespie". Furthermore, how would this "Gillespie" author have access to Torch-v1, a private engine?
Well if we download CCRL's archive for Torch-v1, we very quickly find this PGN, which is identical, but with the name Obsidian instead.

Code: Select all

[Event "CCRL Blitz"]
[Site "CCRL"]
[Date "2023.12.25"]
[Round "211.10.997"]
[White "Obsidian 9.0 64-bit"]
[Black "Torch v1 64-bit"]
[Result "1/2-1/2"]
[ECO "D10"]
[Opening "QGD Slav defence"]
[PlyCount "279"]
[WhiteElo "3653"]
[BlackElo "3756"]

1. d4 d5 2. c4 c6 3. Nc3 Nf6 4. Bg5 Nbd7 5. cxd5 cxd5 6. Qc2 e6 7. Nf3 h6 8.
Bh4 a6 9. e3 b5 10. Bd3 Bd6 11. a4 b4 12. Ne2 a5 13. Bxf6 gxf6 14. O-O Qb6 15.
Rfc1 Ke7 16. g3 Ba6 17. Bxa6 Qxa6 18. Qd1 Rhc8 19. Rxc8 Rxc8 20. Rc1 Rxc1 21.
Nxc1 Qc6 22. b3 Qc3 23. Ne2 Qc8 24. Ne1 f5 25. Qc2 Qb7 26. Kg2 Nf6 27. Qc1 Ne4
28. Nd3 Qc7 29. Qxc7+ Bxc7 30. Ng1 Nd2 31. Nc1 f6 32. Nf3 Nxf3 33. Kxf3 e5 34.
Ne2 Ke6 35. Kg2 Bb6 36. f4 Bc7 37. Ng1 e4 38. Ne2 h5 39. Kf1 Ke7 40. Kg1 Bd8
41. Nc1 Kf7 42. h3 Bc7 43. Kg2 Ke6 44. Kf2 Bb6 45. Kf1 Bd8 46. Ke1 Kd7 47. Kd1
Be7 48. Ne2 Bd8 49. Ke1 Ke7 50. Kf1 Bc7 51. Ng1 Bd8 52. Ke2 Kd7 53. Kf2 Bb6 54.
Ne2 Bc7 55. Kg2 Ke6 56. Kh1 Bd8 57. Kh2 Kf7 58. Nc1 Ke6 59. Kg1 Kd7 60. Kf2 Bb6
61. Kf1 Bd8 62. Ne2 Ke7 63. Kg2 Ke6 64. Kf2 Bb6 65. Ke1 Kd7 66. Kd2 Ke7 67. Kd1
Bd8 68. Ng1 Ke6 69. Kc1 Kd7 70. Kb1 Bb6 71. Ne2 Kc8 72. Nc1 Kc7 73. Kb2 Kd8 74.
Ne2 Kd7 75. Ng1 Ke7 76. Kc1 Bc7 77. Kd1 Bd8 78. Kd2 Kd7 79. Ne2 Bb6 80. Nc1 Bc7
81. Na2 Ke6 82. Ke1 Bd8 83. Kf2 Kd7 84. Kg2 Bb6 85. Kg1 Ke7 86. Kf2 Bd8 87. Kg2
Kd7 88. Nc1 Ke6 89. Kh1 Ke7 90. h4 Bb6 91. Ne2 Ke6 92. Kg1 Bd8 93. Kf2 Kf7 94.
Ng1 Ke7 95. Ke1 Bb6 96. Ne2 Bc7 97. Kf2 Bb6 98. Kg2 Bd8 99. Kf1 Kd7 100. Ke1
Ke6 101. Ng1 Kd7 102. Kd2 Bb6 103. Kd1 Ke7 104. Ke1 Ke6 105. Nh3 Kf7 106. Kd2
Ke7 107. Nf2 Kf7 108. Nh1 Kg6 109. Ke1 Ba7 110. Kf2 Kg7 111. Ke2 Kf7 112. Nf2
Ke7 113. Kd2 Bb6 114. Nh1 Ke6 115. Ke2 Kd7 116. Nf2 Ke7 117. Ke1 Ke6 118. Nh1
Ba7 119. Kf1 Kf7 120. Ke2 Kg6 121. Kd1 Bb6 122. Kd2 Bd8 123. Ke2 Bc7 124. Kf1
Bd8 125. Kf2 Bb6 126. Kg2 Kg7 127. Nf2 Ba7 128. Nh3 Kh7 129. Kf2 Bb6 130. Ke2
Bd8 131. Kd1 Bb6 132. Kd2 Kg6 133. Ke2 Kh7 134. Ng1 Kh8 135. Ke1 Kh7 136. Kf1
Bc7 137. Kg2 Bb6 138. Kh2 Kh6 139. Kh3 Bd8 140. Ne2 1/2-1/2

So its established that at least this "Gillespie" individual is willing to lie about things, and slap his own name onto some other source's PGN.

I would not delete it, but leave it up with a big red line through it

[Sylwy]

I happened to come across a ranking list that I had never seen before, which is called ICCRL (https://www.iccrl.org/rankings). And on the rank, I noticed an engine called Gillespie is #2, stronger than Torch and is only 1 elo behind Stockfish 17.1. Checking its official website(https://gillespie.alexdukas.dev/), The author Alexander Dukas wrote Gillespie employs a Mixture of Experts (MoE) NNUE architecture with five specialized neural networks, each trained on billions of positions from specific game phases,(seems like Cerberus on SPCC)which interests me a lot. Can anyone get this chess engine from the author and send it to me please？

Nothingness generates geniuses....and geniuses write chess engines that are the envy of the AI ​​world......Do you remember the famous site (it's been a few years since then) that promoted a never-before-seen commercial engine.....in the end a poor clone appeared.....The original site was deleted but....traces remained.....the world is full of scammers and frustrated people.....

https://www.facebook.com/rick48engine/

https://www.chess.com/forum/view/genera ... tockfish-9

Rick48-v4 is a Winboard chess engine employing a revolutionary Active Space Algorithm (ASA).

[Mark Tang]

Good catch Andy, did a double game check, gillespie-games-dataset.pgn vs Obsidian_9_0_64-bit.pgn from CCRL, both 2051 games and 2051 doubles. A fraud + no executable.

The things that people do.

Yes,Rebel. I also realize that Obsidian is not the ICCRL ranking list, maybe the author only changes the name of Obsidian and writes the elo by himself(because Obsidian doesn’t have such strength to be #2)