Dragon vs GM Nakamura Analog Handicap Match

[lkaufman]

What I've consistently observed is that humans always do better taking handicaps from top engines than similarly rated engines do. Maybe it's because they have a better understanding of simplifying when ahead, or just that they know to respect the opponent and avoid anything unclear. I have been searching for a computer opponent that can truly simulate a human GM in these handicap matches, but I haven't really found one yet. It's easy for me to run such simulations, but they don't predict well how the human will do.

So you want more brains and less calculation. Here is what I came up with to mimic a result. Lc0-CPU (J94-40) and set the nodes per second option = 4. Lc0 is searching to only 3 or 4 ply. But does much better then the A/B engines with almost now search. Is the style more GM like? But it is fun to see what it takes to win this match for the human player.

Code: Select all

DESKTOP-CORSAIR, Blitz 5.0min+1.0sec  0

                                         1234
1   Lc0 v0.26.3 - CPU               +89   1½½½   2.5/4
2   Dragon by Komodo Chess 64-bit   -89  0½½½    1.5/4

[pgn][Event "DESKTOP-CORSAIR, Blitz 5.0min+1.0sec"]
[Site "DESKTOP-CORSAIR"]
[Date "2020.11.17"]
[Round "1"]
[White "Dragon by Komodo Chess 64-bit"]
[Black "Lc0 v0.26.3 - CPU"]
[Result "0-1"]
[Annotator "-2.82;-3.08"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/P1PPP1PP/RNBQKBNR w KQkq - 0 1"]
[PlyCount "200"]
[TimeControl "300+1"]

{AMD Ryzen Threadripper 2950X 16-Core Processor 3493 MHz W=25.4 plies; 24,
292kN/s; 3,302,788 TBAs B=3.0 plies; 0kN/s; 1 TBAs} 1. e3 {[%eval -282,25]
[%emt 0:00:10]} g6 {[%eval -308,2] [%emt 0:00:07]} 2. Nf3 {[%eval -303,26]
[%emt 0:00:14]} Bg7 {[%eval -338,3] [%emt 0:00:07]} 3. d4 {[%eval -308,26]
[%emt 0:00:08]} c5 {[%eval -368,3] [%emt 0:00:07] (Nf6)} 4. Nc3 {[%eval -298,
24] [%emt 0:00:04]} Nf6 {[%eval -347,3] [%emt 0:00:07] (Nc6)} 5. Rb1 {[%eval
-316,26] [%emt 0:00:18]} O-O {[%eval -329,3] [%emt 0:00:07]} 6. Bd3 {[%eval
-312,25] [%emt 0:00:07]} d5 {[%eval -308,3] [%emt 0:00:07] (b6)} 7. O-O {
[%eval -318,25] [%emt 0:00:12]} c4 {[%eval -305,4] [%emt 0:00:07] (b6)} 8. Be2
{[%eval -278,23] [%emt 0:00:05]} Nc6 {[%eval -348,3] [%emt 0:00:07]} 9. a4 {
[%eval -309,25] [%emt 0:00:10] (Ne5)} Qc7 {[%eval -400,3] [%emt 0:00:07] (Rb8)}
10. Qe1 {[%eval -300,25] [%emt 0:00:17]} Bf5 {[%eval -483,3] [%emt 0:00:07]}
11. Bd1 {[%eval -303,25] [%emt 0:00:04]} Rad8 {[%eval -449,3] [%emt 0:00:07]
(a6)} 12. Nb5 {[%eval -309,24] [%emt 0:00:15] (h3)} Qc8 {[%eval -529,3] [%emt
0:00:07] (Qd7)} 13. h3 {[%eval -316,23] [%emt 0:00:03] (Nc3)} Ne4 {[%eval -486,
3] [%emt 0:00:07]} 14. Nh4 {[%eval -311,23] [%emt 0:00:03] (Nd2)} Be6 {[%eval
-575,3] [%emt 0:00:07]} 15. Nc3 {[%eval -322,27] [%emt 0:00:14] (Nf3)} f5 {
[%eval -579,3] [%emt 0:00:07] (b6)} 16. Nf3 {[%eval -320,24] [%emt 0:00:03]} b6
{[%eval -528,3] [%emt 0:00:07] (Qc7)} 17. Bb2 {[%eval -327,25] [%emt 0:00:22]
(Ne2)} Qc7 {[%eval -583,2] [%emt 0:00:07]} 18. Ne2 {[%eval -329,23] [%emt 0:00:
03]} Bf7 {[%eval -652,2] [%emt 0:00:07]} 19. Ba3 {[%eval -343,26] [%emt 0:00:
29] (c3)} Rfe8 {[%eval -583,3] [%emt 0:00:07]} 20. c3 {[%eval -338,25] [%emt 0:
00:05]} e5 {[%eval -559,3] [%emt 0:00:07]} 21. Bc1 {[%eval -350,23] [%emt 0:00:
04] (Bc2)} h6 {[%eval -631,3] [%emt 0:00:07]} 22. Bc2 {[%eval -346,25] [%emt 0:
00:07]} a6 {[%eval -536,3] [%emt 0:00:07] (exd4)} 23. Bd2 {[%eval -332,23]
[%emt 0:00:04] (Ba3)} Rb8 {[%eval -817,2] [%emt 0:00:07]} 24. Qd1 {[%eval -337,
24] [%emt 0:00:05] (Bc1)} Nxd2 {[%eval -769,3] [%emt 0:00:07] (exd4)} 25. Qxd2
{[%eval -381,22] [%emt 0:00:03]} e4 {[%eval -888,4] [%emt 0:00:07]} 26. Nh4 {
[%eval -405,24] [%emt 0:00:09]} b5 {[%eval -795,3] [%emt 0:00:07]} 27. axb5 {
[%eval -401,25] [%emt 0:00:04] (Qe1)} axb5 {[%eval -939,4] [%emt 0:00:07]} 28.
Bd1 {[%eval -413,26] [%emt 0:00:06] (Qe1)} Rf8 {[%eval -948,3] [%emt 0:00:07]}
29. Nf4 {[%eval -444,28] [%emt 0:00:14] (Qe1)} Ne7 {[%eval -1461,3] [%emt 0:00:
07]} 30. Ne2 {[%eval -451,26] [%emt 0:00:02] (g4)} Be8 {[%eval -1069,3] [%emt
0:00:07] (Qd6)} 31. g3 {[%eval -396,22] [%emt 0:00:02]} g5 {[%eval -995,3]
[%emt 0:00:07] (Nc6)} 32. Ng2 {[%eval -406,23] [%emt 0:00:03]} Bh5 {[%eval
-998,3] [%emt 0:00:07] (Qd6)} 33. Kh2 {[%eval -414,24] [%emt 0:00:04] (Nef4)}
Qd6 {[%eval -980,3] [%emt 0:00:07] (Bf3)} 34. Nc1 {[%eval -391,22] [%emt 0:00:
02] (Ng1)} Be8 {[%eval -1035,3] [%emt 0:00:07] (Bg6)} 35. Na2 {[%eval -404,23]
[%emt 0:00:05]} Nc6 {[%eval -1130,3] [%emt 0:00:07]} 36. Qe1 {[%eval -413,23]
[%emt 0:00:04] (Nb4)} Rf6 {[%eval -1168,2] [%emt 0:00:07] (Ra8)} 37. Rf2 {
[%eval -396,20] [%emt 0:00:02]} Bf8 {[%eval -1163,3] [%emt 0:00:06]} 38. Rfb2 {
[%eval -419,21] [%emt 0:00:02]} Ra8 {[%eval -1258,3] [%emt 0:00:07] (Qe6)} 39.
Nc1 {[%eval -401,22] [%emt 0:00:03]} Ne7 {[%eval -1201,3] [%emt 0:00:06]} 40.
Rf2 {[%eval -435,25] [%emt 0:00:14] (Ra2)} Qc7 {[%eval -1578,2] [%emt 0:00:06]}
41. Qf1 {[%eval -449,25] [%emt 0:00:05] (Ra2)} Ra3 {[%eval -1415,2] [%emt 0:00:
06] (Rfa6)} 42. Ne2 {[%eval -434,22] [%emt 0:00:04] (Na2)} Rfa6 {[%eval -1324,
3] [%emt 0:00:06] (Ra2)} 43. Ne1 {[%eval -451,24] [%emt 0:00:03]} Ra1 {[%eval
-1444,3] [%emt 0:00:06]} 44. Rxa1 {[%eval -487,25] [%emt 0:00:03]} Rxa1 {
[%eval -1408,4] [%emt 0:00:05]} 45. Nc2 {[%eval -494,25] [%emt 0:00:01]} Rb1 {
[%eval -1274,3] [%emt 0:00:06]} 46. Rg2 {[%eval -532,25] [%emt 0:00:06] (Na3)}
Bh5 {[%eval -1620,3] [%emt 0:00:05] (Qd6)} 47. Na3 {[%eval -545,25] [%emt 0:00:
01]} Rb2 {[%eval -1361,3] [%emt 0:00:05]} 48. Nf4 {[%eval -576,24] [%emt 0:00:
01] (Ng1)} Rxg2+ {[%eval -1915,4] [%emt 0:00:05]} 49. Nxg2 {[%eval -698,26]
[%emt 0:00:03]} Be8 {[%eval -1709,3] [%emt 0:00:05] (Bxd1)} 50. Nc2 {[%eval
-504,25] [%emt 0:00:02]} Qa5 {[%eval -1770,2] [%emt 0:00:04]} 51. Qe1 {[%eval
-527,26] [%emt 0:00:03]} Nc6 {[%eval -1925,2] [%emt 0:00:04]} 52. Kg1 {[%eval
-545,26] [%emt 0:00:03] (h4)} Bd6 {[%eval -1731,2] [%emt 0:00:04]} 53. h4 {
[%eval -566,25] [%emt 0:00:05] (g4)} Qc7 {[%eval -2020,2] [%emt 0:00:04] (Ne7)}
54. Kh2 {[%eval -574,26] [%emt 0:00:02] (Qf2)} Kg7 {[%eval -2389,2] [%emt 0:00:
03]} 55. Be2 {[%eval -617,26] [%emt 0:00:04] (Qf2)} Qb8 {[%eval -2431,2] [%emt
0:00:03] (Qa5)} 56. Qf2 {[%eval -611,24] [%emt 0:00:01]} Bd7 {[%eval -2330,3]
[%emt 0:00:03]} 57. Bd1 {[%eval -631,24] [%emt 0:00:02] (Bh5)} b4 {[%eval
-2224,2] [%emt 0:00:03] (Qc7)} 58. cxb4 {[%eval -562,25] [%emt 0:00:01] (Nxb4)}
Nxb4 {[%eval -2350,3] [%emt 0:00:02]} 59. Nxb4 {[%eval -563,24] [%emt 0:00:01]}
Qxb4 {[%eval -2459,3] [%emt 0:00:02]} 60. Qe2 {[%eval -618,24] [%emt 0:00:01]
(Qa2)} Qb8 {[%eval -2731,3] [%emt 0:00:02]} 61. Qf2 {[%eval -632,23] [%emt 0:
00:00] (Qe1)} c3 {[%eval -3315,3] [%emt 0:00:02] (Qb1)} 62. Bc2 {[%eval -674,
24] [%emt 0:00:03]} Qb2 {[%eval -2842,3] [%emt 0:00:02]} 63. Kg1 {[%eval -765,
23] [%emt 0:00:01] (Qe2)} Qc1+ {[%eval -3873,2] [%emt 0:00:02] (Kg6)} 64. Kh2 {
[%eval -815,20] [%emt 0:00:00]} Qd2 {[%eval -4612,3] [%emt 0:00:02] (Kf6)} 65.
Kg1 {[%eval -688,25] [%emt 0:00:01]} Bb5 {[%eval -3562,2] [%emt 0:00:02]
(Qxf2+)} 66. hxg5 {[%eval -900,26] [%emt 0:00:02] (Qxf5)} Qxf2+ {[%eval -3795,
3] [%emt 0:00:02]} 67. Kxf2 {[%eval -941,25] [%emt 0:00:01]} hxg5 {[%eval
-4034,3] [%emt 0:00:02]} 68. Bb3 {[%eval -1088,26] [%emt 0:00:03] (Ne1)} Kg6 {
[%eval -3727,2] [%emt 0:00:01] (Bc4)} 69. Ne1 {[%eval -949,25] [%emt 0:00:00]}
Bc4 {[%eval -4056,2] [%emt 0:00:01]} 70. Bc2 {[%eval -1089,25] [%emt 0:00:02]
(Ba4)} f4 {[%eval -4190,2] [%emt 0:00:01] (Kg7)} 71. g4 {[%eval -1185,22]
[%emt 0:00:01] (gxf4)} Kf6 {[%eval -5280,2] [%emt 0:00:01] (Kg7)} 72. Ng2 {
[%eval -1251,23] [%emt 0:00:01] (Bd1)} Bd3 {[%eval -5142,2] [%emt 0:00:01]
(Kg7)} 73. Bd1 {[%eval -25000,23] [%emt 0:00:00] (Ne1)} c2 {[%eval -6985,2]
[%emt 0:00:01]} 74. Bxc2 {[%eval -25000,31] [%emt 0:00:00]} Bxc2 {[%eval -7976,
2] [%emt 0:00:01]} 75. exf4 {[%eval -25000,33] [%emt 0:00:00] (Ke1)} gxf4 {
[%eval -9198,2] [%emt 0:00:01]} 76. Nh4 {[%eval -32737,21] [%emt 0:00:01] (Ne1)
} Kg5 {[%eval -9843,1] [%emt 0:00:01] (e3+)} 77. Nf5 {[%eval -25000,39] [%emt
0:00:00]} Bb4 {[%eval -9768,2] [%emt 0:00:01] (e3+)} 78. Ng7 {[%eval -32747,28]
[%emt 0:00:00] (Ne3)} e3+ {[%eval -10502,1] [%emt 0:00:01] (Bd3)} 79. Ke2 {
[%eval -32749,23] [%emt 0:00:00]} Kxg4 {[%eval -10005,2] [%emt 0:00:01]} 80.
Ne8 {[%eval -32751,25] [%emt 0:00:00] (Ne6)} Bd2 {[%eval -10620,1] [%emt 0:00:
01] (Be7)} 81. Nf6+ {[%eval -25000,16] [%emt 0:00:00]} Kf5 {[%eval -9621,2]
[%emt 0:00:01] (Kg5)} 82. Nxd5 {[%eval -25000,24] [%emt 0:00:00]} Be4 {[%eval
-11823,2] [%emt 0:00:01]} 83. Nxe3+ {[%eval -25000,99] [%emt 0:00:00] (Ne7+)}
Bxe3 {[%eval -32507,1] [%emt 0:00:00]} 84. d5 {[%eval -27999,0] [%emt 0:00:00]}
Bxd5 {[%eval -27999,0] [%emt 0:00:00]} 85. Kd3 {[%eval -27999,0] [%emt 0:00:00]
} f3 {[%eval -27999,1] [%emt 0:00:00]} 86. Kxe3 {[%eval -27996,1] [%emt 0:00:
00]} Ke5 {[%eval -27997,0] [%emt 0:00:00]} 87. Kf2 {[%eval -27997,0] [%emt 0:
00:00]} Kf4 {[%eval -27998,1] [%emt 0:00:00]} 88. Ke1 {[%eval -27998,0] [%emt
0:00:00]} Kg3 {[%eval -27999,1] [%emt 0:00:00]} 89. Kf1 {[%eval -27999,0]
[%emt 0:00:00]} f2 {[%eval -27998,1] [%emt 0:00:00]} 90. Ke2 {[%eval -27998,1]
[%emt 0:00:00]} Bc4+ {[%eval -27999,0] [%emt 0:00:00]} 91. Kd1 {[%eval -27999,
0] [%emt 0:00:00]} f1=Q+ {[%eval -27997,1] [%emt 0:00:00]} 92. Kc2 {[%eval
-27997,0] [%emt 0:00:00]} Qa1 {[%eval -27998,0] [%emt 0:00:00]} 93. Kd2 {
[%eval -27998,1] [%emt 0:00:00]} Bd3 {[%eval -27999,1] [%emt 0:00:00]} 94. Kxd3
{[%eval -27993,1] [%emt 0:00:00]} Qa4 {[%eval -27994,1] [%emt 0:00:00]} 95. Kc3
{[%eval -27994,1] [%emt 0:00:00]} Kf3 {[%eval -27995,0] [%emt 0:00:00]} 96. Kb2
{[%eval -27995,1] [%emt 0:00:00]} Qb4+ {[%eval -27996,0] [%emt 0:00:00]} 97.
Kc1 {[%eval -27996,0] [%emt 0:00:00]} Ke2 {[%eval -27997,0] [%emt 0:00:00]} 98.
Kc2 {[%eval -27997,1] [%emt 0:00:00]} Ke3 {[%eval -27998,1] [%emt 0:00:00]} 99.
Kc1 {[%eval -27998,1] [%emt 0:00:00]} Kd3 {[%eval -27999,1] [%emt 0:00:00]}
100. Kd1 {[%eval -27999,1] [%emt 0:00:00]} Qd2# {[%eval -32765,0] [%emt 0:00:
00]} 0-1

[/pgn]

[pgn][Event "DESKTOP-CORSAIR, Blitz 5.0min+1.0sec"]
[Site "DESKTOP-CORSAIR"]
[Date "2020.11.17"]
[Round "2"]
[White "Dragon by Komodo Chess 64-bit"]
[Black "Lc0 v0.26.3 - CPU"]
[Result "1/2-1/2"]
[Annotator "-2.82;-2.34"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PP1PP1PP/RNBQKBNR w KQkq - 0 1"]
[PlyCount "92"]
[TimeControl "300+1"]

{AMD Ryzen Threadripper 2950X 16-Core Processor 3493 MHz W=28.8 plies; 25,
703kN/s; 918,626 TBAs B=3.8 plies; 0kN/s} 1. Nc3 {[%eval -282,27] [%emt 0:00:
19]} d5 {[%eval -234,3] [%emt 0:00:07]} 2. d4 {[%eval -281,27] [%emt 0:00:07]}
Nf6 {[%eval -239,3] [%emt 0:00:07] (c6)} 3. Bf4 {[%eval -286,26] [%emt 0:00:04]
(Nf3)} c6 {[%eval -312,3] [%emt 0:00:07]} 4. Qc2 {[%eval -283,25] [%emt 0:00:
05] (Nf3)} g6 {[%eval -285,3] [%emt 0:00:07]} 5. e3 {[%eval -296,27] [%emt 0:
00:04]} Bf5 {[%eval -363,4] [%emt 0:00:07]} 6. Bd3 {[%eval -298,26] [%emt 0:00:
04]} Bxd3 {[%eval -369,4] [%emt 0:00:07] (e6)} 7. Qxd3 {[%eval -271,24] [%emt
0:00:05]} Bg7 {[%eval -416,3] [%emt 0:00:07]} 8. Nf3 {[%eval -281,25] [%emt 0:
00:05]} Nbd7 {[%eval -404,3] [%emt 0:00:07]} 9. O-O {[%eval -275,26] [%emt 0:
00:12]} O-O {[%eval -387,3] [%emt 0:00:07]} 10. h3 {[%eval -283,24] [%emt 0:00:
07]} Re8 {[%eval -387,3] [%emt 0:00:07] (Nb6)} 11. Bh2 {[%eval -288,24] [%emt
0:00:06] (Rae1)} a5 {[%eval -366,2] [%emt 0:00:07] (Nb6)} 12. Rac1 {[%eval
-265,24] [%emt 0:00:12] (Rf2)} a4 {[%eval -427,3] [%emt 0:00:07] (b5)} 13. Qc2
{[%eval -279,25] [%emt 0:00:11] (Kh1)} Qa5 {[%eval -394,3] [%emt 0:00:07] (Nb6)
} 14. a3 {[%eval -265,20] [%emt 0:00:04] (Kh1)} Nb6 {[%eval -406,2] [%emt 0:00:
07] (Qa6)} 15. Bc7 {[%eval -262,25] [%emt 0:00:03] (Ne5)} Rec8 {[%eval -354,3]
[%emt 0:00:07] (Rac8)} 16. Bxb6 {[%eval -258,26] [%emt 0:00:04]} Qxb6 {[%eval
-285,5] [%emt 0:00:06]} 17. Nxa4 {[%eval -252,27] [%emt 0:00:05]} Qc7 {[%eval
-298,4] [%emt 0:00:07] (Qa7)} 18. Nc5 {[%eval -246,25] [%emt 0:00:11]} b6 {
[%eval -349,3] [%emt 0:00:07]} 19. Nd3 {[%eval -240,25] [%emt 0:00:04]} Qd6 {
[%eval -435,3] [%emt 0:00:07] (Qb7)} 20. Nfe5 {[%eval -253,23] [%emt 0:00:03]}
c5 {[%eval -447,4] [%emt 0:00:06]} 21. Rf3 {[%eval -266,25] [%emt 0:00:04]
(Qf2)} Rc7 {[%eval -723,3] [%emt 0:00:07] (Ne4)} 22. Qe2 {[%eval -241,24]
[%emt 0:00:04] (Rcf1)} Rf8 {[%eval -373,3] [%emt 0:00:07] (Ra4)} 23. Rcf1 {
[%eval -245,25] [%emt 0:00:04]} Rcc8 {[%eval -394,3] [%emt 0:00:07] (Qd8)} 24.
Qd1 {[%eval -220,25] [%emt 0:00:24] (h4)} c4 {[%eval -542,3] [%emt 0:00:07]
(Qc7)} 25. Nb4 {[%eval -239,24] [%emt 0:00:04]} Qe6 {[%eval -251,3] [%emt 0:00:
07]} 26. Na2 {[%eval -257,26] [%emt 0:00:16] (Qa4)} Ne4 {[%eval -147,3] [%emt
0:00:07] (Nh5)} 27. Nxf7 {[%eval -158,24] [%emt 0:00:03]} Bf6 {[%eval -133,4]
[%emt 0:00:07] (h5)} 28. Nh6+ {[%eval -121,27] [%emt 0:00:06]} Kg7 {[%eval -93,
5] [%emt 0:00:06]} 29. Ng4 {[%eval -158,28] [%emt 0:00:05]} h5 {[%eval -93,4]
[%emt 0:00:07] (Bh4)} 30. Nxf6 {[%eval -115,25] [%emt 0:00:06]} exf6 {[%eval
-80,4] [%emt 0:00:06]} 31. Qc1 {[%eval -136,27] [%emt 0:00:21] (Nc3)} h4 {
[%eval -117,3] [%emt 0:00:07] (Qe7)} 32. Nc3 {[%eval -51,27] [%emt 0:00:18]}
Ng3 {[%eval -108,4] [%emt 0:00:06] (f5)} 33. Rxg3 {[%eval -42,28] [%emt 0:00:
07] (Re1)} hxg3 {[%eval -75,5] [%emt 0:00:06]} 34. Rf3 {[%eval -60,26] [%emt 0:
00:04]} Rce8 {[%eval -90,3] [%emt 0:00:07] (g5)} 35. Rxg3 {[%eval -51,28]
[%emt 0:00:04]} Qd6 {[%eval -95,3] [%emt 0:00:07] (Rh8)} 36. Ne2 {[%eval -22,
25] [%emt 0:00:04] (Rf3)} Re4 {[%eval -87,3] [%emt 0:00:06] (Kf7)} 37. Qf1 {
[%eval 0,29] [%emt 0:00:04] (Qe1)} Rfe8 {[%eval -41,3] [%emt 0:00:06] (Kh6)}
38. Qf2 {[%eval 0,32] [%emt 0:00:04]} Kf7 {[%eval -32,4] [%emt 0:00:07]} 39. a4
{[%eval 0,32] [%emt 0:00:03] (Nc3)} Qe7 {[%eval -183,4] [%emt 0:00:06] (Qe6)}
40. Nf4 {[%eval 0,36] [%emt 0:00:05]} Rxe3 {[%eval -32,5] [%emt 0:00:06]} 41.
Nxg6 {[%eval 0,39] [%emt 0:00:03] (Rxg6)} Re1+ {[%eval -41,4] [%emt 0:00:06]}
42. Kh2 {[%eval 0,42] [%emt 0:00:08]} Qd6 {[%eval -47,5] [%emt 0:00:06]} 43.
Nh8+ {[%eval 0,47] [%emt 0:00:04] (Qf5)} Kf8 {[%eval 8,4] [%emt 0:00:06]} 44.
Ng6+ {[%eval 0,47] [%emt 0:00:03]} Kf7 {[%eval -7,4] [%emt 0:00:05]} 45. Nh8+ {
[%eval 0,52] [%emt 0:00:03]} Kf8 {[%eval 8,3] [%emt 0:00:06]} 46. Ng6+ {
[%eval 0,52] [%emt 0:00:03]} Kf7 {[%eval 0,2] [%emt 0:00:05] Draw accepted}
1/2-1/2

[/pgn]

[pgn][Event "DESKTOP-CORSAIR, Blitz 5.0min+1.0sec"]
[Site "DESKTOP-CORSAIR"]
[Date "2020.11.17"]
[Round "3"]
[White "Dragon by Komodo Chess 64-bit"]
[Black "Lc0 v0.26.3 - CPU"]
[Result "1/2-1/2"]
[Annotator "-2.68;-2.70"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/P1PPPP1P/RNBQKBNR w KQkq - 0 1"]
[PlyCount "99"]
[TimeControl "300+1"]

{AMD Ryzen Threadripper 2950X 16-Core Processor 3493 MHz W=32.2 plies; 26,
516kN/s; 105,593,506 TBAs B=4.1 plies; 0kN/s; 4 TBAs} 1. c4 {[%eval -268,25]
[%emt 0:00:10]} g6 {[%eval -270,3] [%emt 0:00:07]} 2. Bb2 {[%eval -284,27]
[%emt 0:00:07] (Nc3)} Nf6 {[%eval -325,3] [%emt 0:00:06]} 3. e3 {[%eval -272,
27] [%emt 0:00:07] (h4)} Bg7 {[%eval -323,3] [%emt 0:00:07] (d6)} 4. Bg2 {
[%eval -281,26] [%emt 0:00:07]} d6 {[%eval -330,3] [%emt 0:00:07] (Nc6)} 5. Ne2
{[%eval -268,25] [%emt 0:00:08] (d4)} O-O {[%eval -334,3] [%emt 0:00:07]} 6.
O-O {[%eval -275,26] [%emt 0:00:14] (d4)} e5 {[%eval -301,3] [%emt 0:00:07]} 7.
d4 {[%eval -283,23] [%emt 0:00:05] (Nbc3)} Nc6 {[%eval -362,3] [%emt 0:00:07]
(Na6)} 8. dxe5 {[%eval -296,25] [%emt 0:00:14] (Nd2)} Nxe5 {[%eval -385,3]
[%emt 0:00:07]} 9. Nd2 {[%eval -279,24] [%emt 0:00:06]} Re8 {[%eval -387,3]
[%emt 0:00:07] (Bd7)} 10. Rb1 {[%eval -312,25] [%emt 0:00:26] (Qb3)} Rb8 {
[%eval -443,3] [%emt 0:00:07]} 11. Qc2 {[%eval -312,24] [%emt 0:00:06] (Bd4)}
b6 {[%eval -418,3] [%emt 0:00:07] (Nh5)} 12. Ng3 {[%eval -304,25] [%emt 0:00:
19]} Bb7 {[%eval -487,3] [%emt 0:00:07]} 13. e4 {[%eval -325,25] [%emt 0:00:11]
} Ned7 {[%eval -433,3] [%emt 0:00:07] (h5)} 14. f3 {[%eval -318,22] [%emt 0:00:
06] (Rbe1)} h5 {[%eval -509,3] [%emt 0:00:07] (Nc5)} 15. Ne2 {[%eval -284,23]
[%emt 0:00:03]} Nc5 {[%eval -481,3] [%emt 0:00:07]} 16. Kh1 {[%eval -316,25]
[%emt 0:00:14] (Nb3)} Nfd7 {[%eval -492,3] [%emt 0:00:07]} 17. Bxg7 {[%eval
-313,26] [%emt 0:00:05]} Kxg7 {[%eval -462,4] [%emt 0:00:06]} 18. f4 {[%eval
-332,26] [%emt 0:00:07]} Nf6 {[%eval -413,3] [%emt 0:00:07] (Qh4)} 19. Nc3 {
[%eval -328,22] [%emt 0:00:06]} Ng4 {[%eval -389,3] [%emt 0:00:07]} 20. Rbe1 {
[%eval -334,22] [%emt 0:00:04]} Qh4 {[%eval -309,4] [%emt 0:00:07] (f5)} 21.
Nf3 {[%eval -202,23] [%emt 0:00:03]} Qf6 {[%eval -280,4] [%emt 0:00:06]} 22.
Ng5 {[%eval -269,23] [%emt 0:00:06]} Qd4 {[%eval -266,3] [%emt 0:00:07] (Ne6)}
23. Nb5 {[%eval -231,24] [%emt 0:00:03]} Qd3 {[%eval -182,6] [%emt 0:00:06]}
24. Qb2+ {[%eval -213,22] [%emt 0:00:03]} f6 {[%eval -138,6] [%emt 0:00:07]}
25. Nxc7 {[%eval -259,23] [%emt 0:00:03]} Re7 {[%eval -122,5] [%emt 0:00:06]}
26. Nd5 {[%eval -321,25] [%emt 0:00:08]} Bxd5 {[%eval -108,6] [%emt 0:00:06]}
27. cxd5 {[%eval -340,25] [%emt 0:00:05]} Ne3 {[%eval -172,4] [%emt 0:00:08]}
28. f5 {[%eval -381,25] [%emt 0:00:05] (Rf3)} Qc2 {[%eval -185,4] [%emt 0:00:
07] (Re5)} 29. Qxc2 {[%eval -135,26] [%emt 0:00:04] (Ne6+)} Nxc2 {[%eval -43,6]
[%emt 0:00:06]} 30. Re2 {[%eval -151,26] [%emt 0:00:03]} Nd4 {[%eval -27,6]
[%emt 0:00:06]} 31. Rd2 {[%eval -176,26] [%emt 0:00:05]} gxf5 {[%eval -114,6]
[%emt 0:00:07]} 32. Nh3 {[%eval -149,27] [%emt 0:00:04]} Nb5 {[%eval -98,5]
[%emt 0:00:06]} 33. Rxf5 {[%eval -153,28] [%emt 0:00:08] (exf5)} Re5 {[%eval
-116,4] [%emt 0:00:07] (Nxe4)} 34. Nf4 {[%eval -93,25] [%emt 0:00:05]} Rxf5 {
[%eval -78,4] [%emt 0:00:07] (Nc3)} 35. exf5 {[%eval -23,26] [%emt 0:00:06]}
Kh6 {[%eval -69,4] [%emt 0:00:06] (Rg8)} 36. Bf1 {[%eval 0,34] [%emt 0:00:04]}
Nc3 {[%eval -25,5] [%emt 0:00:07]} 37. Rg2 {[%eval 0,34] [%emt 0:00:04]} N5e4 {
[%eval -12,4] [%emt 0:00:06] (Re8)} 38. Bd3 {[%eval 0,36] [%emt 0:00:03] (Rg6+)
} Re8 {[%eval -22,3] [%emt 0:00:07]} 39. Rg6+ {[%eval 0,40] [%emt 0:00:04]} Kh7
{[%eval -10,1] [%emt 0:00:00]} 40. Nxh5 {[%eval 0,44] [%emt 0:00:05] (Bxe4)}
Nxd5 {[%eval -6,4] [%emt 0:00:13]} 41. Bxe4 {[%eval 0,50] [%emt 0:00:05]} Rxe4
{[%eval -10,5] [%emt 0:00:06]} 42. Nxf6+ {[%eval 0,54] [%emt 0:00:05]} Nxf6 {
[%eval 0,4] [%emt 0:00:06]} 43. Rxf6 {[%eval 0,58] [%emt 0:00:05]} Ra4 {
[%eval -1,3] [%emt 0:00:07]} 44. Rxd6 {[%eval 0,57] [%emt 0:00:04]} Rxa2 {
[%eval -1,3] [%emt 0:00:06]} 45. Rd7+ {[%eval 0,53] [%emt 0:00:03] (h4)} Kh6 {
[%eval -1,3] [%emt 0:00:05]} 46. f6 {[%eval 0,58] [%emt 0:00:03] (h4)} a5 {
[%eval -3,3] [%emt 0:00:05] (Kg6)} 47. Kg1 {[%eval 0,56] [%emt 0:00:04] (Rb7)}
a4 {[%eval -6,3] [%emt 0:00:05]} 48. h4 {[%eval 0,55] [%emt 0:00:04]} Rc2 {
[%eval -2,3] [%emt 0:00:05] (Re2)} 49. h5 {[%eval 0,56] [%emt 0:00:05] (Ra7)}
a3 {[%eval -7,2] [%emt 0:00:05] (Kxh5)} 50. f7 {[%eval 0,59] [%emt 0:00:04]
Draw accepted} 1/2-1/2

[/pgn]
[pgn][Event "DESKTOP-CORSAIR, Blitz 5.0min+1.0sec"]
[Site "DESKTOP-CORSAIR"]
[Date "2020.11.17"]
[Round "4"]
[White "Dragon by Komodo Chess 64-bit"]
[Black "Lc0 v0.26.3 - CPU"]
[Result "1/2-1/2"]
[Annotator "-2.82;-2.31"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PP1PPP1P/RNBQKBNR w KQkq - 0 1"]
[PlyCount "132"]
[TimeControl "300+1"]

{AMD Ryzen Threadripper 2950X 16-Core Processor 3493 MHz W=34.6 plies; 28,
589kN/s; 102,285,792 TBAs B=4.0 plies; 0kN/s; 3 TBAs} 1. d4 {[%eval -282,26]
[%emt 0:00:07]} d5 {[%eval -231,3] [%emt 0:00:07]} 2. Nc3 {[%eval -286,26]
[%emt 0:00:12]} Nf6 {[%eval -266,3] [%emt 0:00:07] (c6)} 3. Bg5 {[%eval -278,
26] [%emt 0:00:15] (Nf3)} c6 {[%eval -299,3] [%emt 0:00:07] (g6)} 4. e3 {
[%eval -273,23] [%emt 0:00:04]} Bf5 {[%eval -353,3] [%emt 0:00:07] (g6)} 5. Bd3
{[%eval -270,24] [%emt 0:00:05]} Bxd3 {[%eval -303,4] [%emt 0:00:07]} 6. Qxd3 {
[%eval -252,25] [%emt 0:00:06]} e6 {[%eval -291,3] [%emt 0:00:07]} 7. f4 {
[%eval -277,25] [%emt 0:00:11] (0-0-0)} Nbd7 {[%eval -303,3] [%emt 0:00:07]
(Bb4)} 8. O-O-O {[%eval -277,23] [%emt 0:00:04] (Nf3)} Bb4 {[%eval -361,3]
[%emt 0:00:07]} 9. Nge2 {[%eval -295,26] [%emt 0:00:05]} Qa5 {[%eval -316,3]
[%emt 0:00:07] (Be7)} 10. a3 {[%eval -259,23] [%emt 0:00:05] (f5)} Bxc3 {
[%eval -260,3] [%emt 0:00:07]} 11. Nxc3 {[%eval -249,25] [%emt 0:00:08]} g6 {
[%eval -240,3] [%emt 0:00:07] (Qa6)} 12. Kb1 {[%eval -247,23] [%emt 0:00:06]}
O-O {[%eval -225,3] [%emt 0:00:07] (Rb8)} 13. Rhg1 {[%eval -251,22] [%emt 0:00:
10]} Kg7 {[%eval -200,3] [%emt 0:00:07]} 14. Rg2 {[%eval -269,24] [%emt 0:00:
12] (h4)} Rg8 {[%eval -212,2] [%emt 0:00:07] (Rac8)} 15. Rdg1 {[%eval -257,22]
[%emt 0:00:11]} Rae8 {[%eval -230,3] [%emt 0:00:07] (Nh5)} 16. Bh4 {[%eval
-254,23] [%emt 0:00:08] (h4)} Kh8 {[%eval -303,3] [%emt 0:00:07]} 17. Rc2 {
[%eval -260,24] [%emt 0:00:14] (Bg5)} h6 {[%eval -361,3] [%emt 0:00:07] (Qc7)}
18. Rgc1 {[%eval -253,22] [%emt 0:00:09] (Ka2)} Kh7 {[%eval -408,3] [%emt 0:00:
07] (Rc8)} 19. Bxf6 {[%eval -273,25] [%emt 0:00:19] (Ka2)} Nxf6 {[%eval -728,3]
[%emt 0:00:06]} 20. Nd1 {[%eval -260,27] [%emt 0:00:06]} Ne4 {[%eval -649,3]
[%emt 0:00:07]} 21. Nf2 {[%eval -268,29] [%emt 0:00:07]} Nxf2 {[%eval -660,4]
[%emt 0:00:07] (Nd6)} 22. Rxf2 {[%eval -261,28] [%emt 0:00:03]} Qd8 {[%eval
-679,3] [%emt 0:00:07]} 23. Rfc2 {[%eval -260,29] [%emt 0:00:03] (Ka2)} Qf6 {
[%eval -669,3] [%emt 0:00:07] (Qh4)} 24. Ka2 {[%eval -264,30] [%emt 0:00:06]
(Qb3)} Rc8 {[%eval -702,3] [%emt 0:00:07] (Re7)} 25. Qb3 {[%eval -273,29]
[%emt 0:00:04] (Qd2)} Rc7 {[%eval -791,3] [%emt 0:00:07]} 26. Rc3 {[%eval -286,
29] [%emt 0:00:10] (Qa4)} Rgc8 {[%eval -775,3] [%emt 0:00:07] (Qh4)} 27. Qa4 {
[%eval -288,27] [%emt 0:00:03]} Ra8 {[%eval -855,3] [%emt 0:00:07]} 28. R1c2 {
[%eval -301,27] [%emt 0:00:08] (Qa5)} Qe7 {[%eval -839,2] [%emt 0:00:07] (Kg7)}
29. Ka1 {[%eval -288,27] [%emt 0:00:05] (Rc1)} a5 {[%eval -734,2] [%emt 0:00:
07] (Kg7)} 30. Rc1 {[%eval -297,26] [%emt 0:00:04] (Rc5)} b5 {[%eval -707,3]
[%emt 0:00:07] (Qh4)} 31. Qc2 {[%eval -335,25] [%emt 0:00:02]} b4 {[%eval -757,
4] [%emt 0:00:06]} 32. Rxc6 {[%eval -400,28] [%emt 0:00:05] (Rc5)} Rxc6 {
[%eval -287,4] [%emt 0:00:07] (Rb7)} 33. Qxc6 {[%eval -303,26] [%emt 0:00:02]}
Rb8 {[%eval -417,4] [%emt 0:00:07]} 34. Qc5 {[%eval -382,28] [%emt 0:00:08]}
Qb7 {[%eval -642,3] [%emt 0:00:07] (Qh4)} 35. a4 {[%eval -232,28] [%emt 0:00:
02]} b3 {[%eval -491,4] [%emt 0:00:06]} 36. Qxa5 {[%eval -301,29] [%emt 0:00:
02]} Ra8 {[%eval -130,4] [%emt 0:00:07] (Qa8)} 37. Rc7 {[%eval -174,29] [%emt
0:00:02]} Rxa5 {[%eval -44,6] [%emt 0:00:07]} 38. Rxb7 {[%eval -154,27] [%emt
0:00:02]} Rxa4+ {[%eval -53,7] [%emt 0:00:06]} 39. Kb1 {[%eval -147,5] [%emt 0:
00:00]} g5 {[%eval -68,7] [%emt 0:00:07] (Kg7)} 40. fxg5 {[%eval 0,43] [%emt 0:
00:06]} hxg5 {[%eval -67,7] [%emt 0:00:05]} 41. Rxf7+ {[%eval 0,48] [%emt 0:00:
08]} Kg6 {[%eval -64,7] [%emt 0:00:06]} 42. Rb7 {[%eval 0,51] [%emt 0:00:08]}
Ra8 {[%eval -55,7] [%emt 0:00:06] (Kf5)} 43. Rxb3 {[%eval 0,45] [%emt 0:00:04]}
Rh8 {[%eval -49,6] [%emt 0:00:06]} 44. Rb6 {[%eval 0,48] [%emt 0:00:05] (Kc2)}
Kf5 {[%eval -57,4] [%emt 0:00:07] (Rxh2)} 45. Kc2 {[%eval 0,48] [%emt 0:00:04]}
Rxh2+ {[%eval -19,6] [%emt 0:00:06]} 46. Kd3 {[%eval 0,50] [%emt 0:00:04]} g4 {
[%eval -16,4] [%emt 0:00:05] (Rh8)} 47. Rb8 {[%eval 0,52] [%emt 0:00:04]} Kg5 {
[%eval -16,3] [%emt 0:00:05] (g3)} 48. b4 {[%eval 0,50] [%emt 0:00:04]} g3 {
[%eval -19,4] [%emt 0:00:05]} 49. b5 {[%eval 0,51] [%emt 0:00:04]} Kg4 {
[%eval -16,3] [%emt 0:00:05] (Rb2)} 50. Rg8+ {[%eval 0,50] [%emt 0:00:03] (b6)}
Kh3 {[%eval -39,4] [%emt 0:00:04]} 51. Rg6 {[%eval 0,45] [%emt 0:00:04] (b6)}
Rb2 {[%eval -60,3] [%emt 0:00:04]} 52. Rh6+ {[%eval 0,50] [%emt 0:00:02]} Kg4 {
[%eval -42,4] [%emt 0:00:04] (Kg2)} 53. Rg6+ {[%eval 0,52] [%emt 0:00:03]} Kf3
{[%eval -38,5] [%emt 0:00:04] (Kh4)} 54. Rf6+ {[%eval 0,59] [%emt 0:00:03]} Kg2
{[%eval -38,5] [%emt 0:00:03]} 55. e4 {[%eval 0,60] [%emt 0:00:03] (Rxe6)} Rb3+
{[%eval -35,3] [%emt 0:00:03] (dxe4+)} 56. Ke2 {[%eval 0,61] [%emt 0:00:02]}
dxe4 {[%eval -9,3] [%emt 0:00:03]} 57. Rxe6 {[%eval 0,56] [%emt 0:00:03]} Rxb5
{[%eval 0,2] [%emt 0:00:03] (Kh3)} 58. Rxe4 {[%eval 0,99] [%emt 0:00:00]} Kh3 {
[%eval -28000,0] [%emt 0:00:00]} 59. Re3 {[%eval -28000,0] [%emt 0:00:00]} Rb2+
{[%eval -28000,0] [%emt 0:00:00]} 60. Kf3 {[%eval -28000,1] [%emt 0:00:00]}
Rf2+ {[%eval -28000,0] [%emt 0:00:00]} 61. Ke4 {[%eval -28000,1] [%emt 0:00:00]
} Kg4 {[%eval -28000,0] [%emt 0:00:00]} 62. Re1 {[%eval -28000,1] [%emt 0:00:
00]} Rf4+ {[%eval -28000,0] [%emt 0:00:00]} 63. Ke3 {[%eval -28000,0] [%emt 0:
00:00]} Rf3+ {[%eval -28000,0] [%emt 0:00:00]} 64. Ke4 {[%eval -28000,1] [%emt
0:00:00]} Rf4+ {[%eval -28000,0] [%emt 0:00:00]} 65. Ke3 {[%eval -28000,0]
[%emt 0:00:00]} Rf3+ {[%eval -28000,0] [%emt 0:00:00]} 66. Ke4 {[%eval -28000,
1] [%emt 0:00:00]} Rf4+ {[%eval -28000,0] [%emt 0:00:00] Draw accepted} 1/2-1/2

[/pgn]

To complete the experiment you should run the crippled Lc0 against the ancient SF in standard chess to see if they are really of roughly comparable strength. If they are then this might indeed be a good way to simulate a human.

[Nordlandia]

Time odds is also reliable alternative to somehow offset the pawn deficit. Last unofficial cutechess support it. Fritz interface allow more time for the white player, in engine matches.

[Laskos]

What I've consistently observed is that humans always do better taking handicaps from top engines than similarly rated engines do. Maybe it's because they have a better understanding of simplifying when ahead, or just that they know to respect the opponent and avoid anything unclear. I have been searching for a computer opponent that can truly simulate a human GM in these handicap matches, but I haven't really found one yet. It's easy for me to run such simulations, but they don't predict well how the human will do.

Have you tried Lc0 with an appropriate net to play handicaps, to mimic the human?
I also have a question about Dragon --- does it have the sophisticated contempt the plain Komodo possesses? It does matter quite a lot.

Yes, I've tried Lc0 as a human substitute, it doesn't do that much better than similarly rated regular engines. The problem is probably that the nets haven't been trained on piece down positions.
Yes, Dragon has the same Contempt as plain Komodo, we will make use of it vs. Nakamura.
At 4 pm Eastern time today (2 hours from now), we plan to play some test games of Dragon vs. my son IM Raymond Kaufman at knight odds (same tc as Hikaru match) on chess.com to make sure everything's working right and to check settings, mostly to compare regular mode with MCTS mode. Look for games of playkomodo vs mimsychess. Ray is about 2400 blitz on chess.com, a long way from Hikaru, but knight odds is much larger than two pawns, so maybe the relative prospects of the human vs. Dragon are fairly similar?

Yes, relative matches seem spaced well, at rapid TC I estimate the difference between Knight and 2 pawns handicaps at some 400 human Elo points.

Using what I have on my hand, I simulated as best as I could the Nakamura match using Dragon (Contempt=100) as White and Lc0 J92-330 net (Drawscore=100 ---> opposite sign from Contempt) as Black (human). The result was:

3 Dragon wins and 5 draws.

That there will be some, even many draws is to be expected, but it seems Naka also will have trouble winning games, if the simulation is worth a penny.

[lkaufman]

What I've consistently observed is that humans always do better taking handicaps from top engines than similarly rated engines do. Maybe it's because they have a better understanding of simplifying when ahead, or just that they know to respect the opponent and avoid anything unclear. I have been searching for a computer opponent that can truly simulate a human GM in these handicap matches, but I haven't really found one yet. It's easy for me to run such simulations, but they don't predict well how the human will do.

Have you tried Lc0 with an appropriate net to play handicaps, to mimic the human?
I also have a question about Dragon --- does it have the sophisticated contempt the plain Komodo possesses? It does matter quite a lot.

Yes, I've tried Lc0 as a human substitute, it doesn't do that much better than similarly rated regular engines. The problem is probably that the nets haven't been trained on piece down positions.
Yes, Dragon has the same Contempt as plain Komodo, we will make use of it vs. Nakamura.
At 4 pm Eastern time today (2 hours from now), we plan to play some test games of Dragon vs. my son IM Raymond Kaufman at knight odds (same tc as Hikaru match) on chess.com to make sure everything's working right and to check settings, mostly to compare regular mode with MCTS mode. Look for games of playkomodo vs mimsychess. Ray is about 2400 blitz on chess.com, a long way from Hikaru, but knight odds is much larger than two pawns, so maybe the relative prospects of the human vs. Dragon are fairly similar?

Yes, relative matches seem spaced well, at rapid TC I estimate the difference between Knight and 2 pawns handicaps at some 400 human Elo points.

Using what I have on my hand, I simulated as best as I could the Nakamura match using Dragon (Contempt=100) as White and Lc0 J92-330 net (Drawscore=100 ---> opposite sign from Contempt) as Black (human). The result was:

3 Dragon wins and 5 draws.

That there will be some, even many draws is to be expected, but it seems Naka also will have trouble winning games, if the simulation is worth a penny.

Ray is playing Dragon now at knight odds on chess.com (15+10). Do you have a way to simulate that pairing? His chess.com blitz rating is 2417, but FIDE rating is around 2260 or so I think.

[mwyoung]

What I've consistently observed is that humans always do better taking handicaps from top engines than similarly rated engines do. Maybe it's because they have a better understanding of simplifying when ahead, or just that they know to respect the opponent and avoid anything unclear. I have been searching for a computer opponent that can truly simulate a human GM in these handicap matches, but I haven't really found one yet. It's easy for me to run such simulations, but they don't predict well how the human will do.

Have you tried Lc0 with an appropriate net to play handicaps, to mimic the human?
I also have a question about Dragon --- does it have the sophisticated contempt the plain Komodo possesses? It does matter quite a lot.

Yes, I've tried Lc0 as a human substitute, it doesn't do that much better than similarly rated regular engines. The problem is probably that the nets haven't been trained on piece down positions.
Yes, Dragon has the same Contempt as plain Komodo, we will make use of it vs. Nakamura.
At 4 pm Eastern time today (2 hours from now), we plan to play some test games of Dragon vs. my son IM Raymond Kaufman at knight odds (same tc as Hikaru match) on chess.com to make sure everything's working right and to check settings, mostly to compare regular mode with MCTS mode. Look for games of playkomodo vs mimsychess. Ray is about 2400 blitz on chess.com, a long way from Hikaru, but knight odds is much larger than two pawns, so maybe the relative prospects of the human vs. Dragon are fairly similar?

Yes, relative matches seem spaced well, at rapid TC I estimate the difference between Knight and 2 pawns handicaps at some 400 human Elo points.

Using what I have on my hand, I simulated as best as I could the Nakamura match using Dragon (Contempt=100) as White and Lc0 J92-330 net (Drawscore=100 ---> opposite sign from Contempt) as Black (human). The result was:

3 Dragon wins and 5 draws.

That there will be some, even many draws is to be expected, but it seems Naka also will have trouble winning games, if the simulation is worth a penny.

Ray is playing Dragon now at knight odds on chess.com (15+10). Do you have a way to simulate that pairing? His chess.com blitz rating is 2417, but FIDE rating is around 2260 or so I think.

With my simulation to mimic that rating. You would use Lc0-CPU (J94-40) and set the nodes per second to about 1.

[Laskos]

What I've consistently observed is that humans always do better taking handicaps from top engines than similarly rated engines do. Maybe it's because they have a better understanding of simplifying when ahead, or just that they know to respect the opponent and avoid anything unclear. I have been searching for a computer opponent that can truly simulate a human GM in these handicap matches, but I haven't really found one yet. It's easy for me to run such simulations, but they don't predict well how the human will do.

Have you tried Lc0 with an appropriate net to play handicaps, to mimic the human?
I also have a question about Dragon --- does it have the sophisticated contempt the plain Komodo possesses? It does matter quite a lot.

Yes, I've tried Lc0 as a human substitute, it doesn't do that much better than similarly rated regular engines. The problem is probably that the nets haven't been trained on piece down positions.
Yes, Dragon has the same Contempt as plain Komodo, we will make use of it vs. Nakamura.
At 4 pm Eastern time today (2 hours from now), we plan to play some test games of Dragon vs. my son IM Raymond Kaufman at knight odds (same tc as Hikaru match) on chess.com to make sure everything's working right and to check settings, mostly to compare regular mode with MCTS mode. Look for games of playkomodo vs mimsychess. Ray is about 2400 blitz on chess.com, a long way from Hikaru, but knight odds is much larger than two pawns, so maybe the relative prospects of the human vs. Dragon are fairly similar?

Yes, relative matches seem spaced well, at rapid TC I estimate the difference between Knight and 2 pawns handicaps at some 400 human Elo points.

Using what I have on my hand, I simulated as best as I could the Nakamura match using Dragon (Contempt=100) as White and Lc0 J92-330 net (Drawscore=100 ---> opposite sign from Contempt) as Black (human). The result was:

3 Dragon wins and 5 draws.

That there will be some, even many draws is to be expected, but it seems Naka also will have trouble winning games, if the simulation is worth a penny.

Ray is playing Dragon now at knight odds on chess.com (15+10). Do you have a way to simulate that pairing? His chess.com blitz rating is 2417, but FIDE rating is around 2260 or so I think.

I would need almost 2 hours.

[lkaufman]

Ray won the first game. Dragon played well enough to recover two pawns by the time the endgame was reached, but the endgame with two pawns for the bishop was still easily won for Ray. Knight odds is a lot. Game two underway.

[Laskos]

Ray won the first game. Dragon played well enough to recover two pawns by the time the endgame was reached, but the endgame with two pawns for the bishop was still easily won for Ray. Knight odds is a lot. Game two underway.

Is Ray allowed to play the best openings? I remember now that Knight odds was tricky for the engine if it plays systematically an unlucky opening.

[mjlef]

What I've consistently observed is that humans always do better taking handicaps from top engines than similarly rated engines do. Maybe it's because they have a better understanding of simplifying when ahead, or just that they know to respect the opponent and avoid anything unclear. I have been searching for a computer opponent that can truly simulate a human GM in these handicap matches, but I haven't really found one yet. It's easy for me to run such simulations, but they don't predict well how the human will do.

Have you tried Lc0 with an appropriate net to play handicaps, to mimic the human?
I also have a question about Dragon --- does it have the sophisticated contempt the plain Komodo possesses? It does matter quite a lot.

Dragon does have Contempt, and does the most important things (controls if the program should seek a draw or avoid it, and not trade down pieces if Contempt is positive). But it lacks some of the special Contempt heuristics. But it should have learned some about them because it was trained solely on Komodo games (and in later stages Dragon games).

Mark

[Marcus9]

Dragon games look so crazy.
Maybe I don't have the right understanding of chess