LeelaKnightOdds

[lkaufman]

I managed to get a draw against it myself today at 5'3", so at least I could calculate a perf. rating for its games today. It played nine games at 5'2" or slower games against 2300+ players, scoring 8.5 out of 9 (plus winning all other games) against average blitz opposition of 2480, for a 2972 performance, roughly 2700 FIDE equivalent. 2/3 of the games were within the blitz range. So I think it is perhaps near 2700 FIDE in slow blitz, but probably not in Rapid, maybe 2600 FIDE or so at 10' + 5".

[Chessqueen]

I managed to get a draw against it myself today at 5'3", so at least I could calculate a perf. rating for its games today. It played nine games at 5'2" or slower games against 2300+ players, scoring 8.5 out of 9 (plus winning all other games) against average blitz opposition of 2480, for a 2972 performance, roughly 2700 FIDE equivalent. 2/3 of the games were within the blitz range. So I think it is perhaps near 2700 FIDE in slow blitz, but probably not in Rapid, maybe 2600 FIDE or so at 10' + 5".

Therefore, as far as I understand you, it will have a hard time beating GM Nakamura or Carlsen at KnightOdds at Blitz 5/3

[lkaufman]

I managed to get a draw against it myself today at 5'3", so at least I could calculate a perf. rating for its games today. It played nine games at 5'2" or slower games against 2300+ players, scoring 8.5 out of 9 (plus winning all other games) against average blitz opposition of 2480, for a 2972 performance, roughly 2700 FIDE equivalent. 2/3 of the games were within the blitz range. So I think it is perhaps near 2700 FIDE in slow blitz, but probably not in Rapid, maybe 2600 FIDE or so at 10' + 5".

Therefore, as far as I understand you, it will have a hard time beating GM Nakamura or Carlsen at KnightOdds at Blitz 5/3

I think that is accurate. Maybe 3'1" (like titled tuesday) might be balanced for those two.

[Chessqueen]

I managed to get a draw against it myself today at 5'3", so at least I could calculate a perf. rating for its games today. It played nine games at 5'2" or slower games against 2300+ players, scoring 8.5 out of 9 (plus winning all other games) against average blitz opposition of 2480, for a 2972 performance, roughly 2700 FIDE equivalent. 2/3 of the games were within the blitz range. So I think it is perhaps near 2700 FIDE in slow blitz, but probably not in Rapid, maybe 2600 FIDE or so at 10' + 5".

Therefore, as far as I understand you, it will have a hard time beating GM Nakamura or Carlsen at KnightOdds at Blitz 5/3

I think that is accurate. Maybe 3'1" (like titled tuesday) might be balanced for those two.

Is there a way to find out if LeelaKnightOdds plays KnightOdds better than your latest Komodo Dragon or MCTS, based on the data that you have so far ?

[lkaufman]

I managed to get a draw against it myself today at 5'3", so at least I could calculate a perf. rating for its games today. It played nine games at 5'2" or slower games against 2300+ players, scoring 8.5 out of 9 (plus winning all other games) against average blitz opposition of 2480, for a 2972 performance, roughly 2700 FIDE equivalent. 2/3 of the games were within the blitz range. So I think it is perhaps near 2700 FIDE in slow blitz, but probably not in Rapid, maybe 2600 FIDE or so at 10' + 5".

Therefore, as far as I understand you, it will have a hard time beating GM Nakamura or Carlsen at KnightOdds at Blitz 5/3

I think that is accurate. Maybe 3'1" (like titled tuesday) might be balanced for those two.

Is there a way to find out if LeelaKnightOdds plays KnightOdds better than your latest Komodo Dragon or MCTS, based on the data that you have so far ?

I believe the answer is yes, but it isn't totally certain yet, since Leela hasn't played any grandmasters yet except for me, and I only play about FM level now at age 76.

[Uri Blass]

today lc0 lost one 14+9 game against somebody(wild151) but I read that the account violated the lichess terms of service.

[Uri Blass]

If we do not count that loss then lc0 has 100% against humans today but most games against weak players.

I see 19 wins in rapid when losses at faster time control are only against bots.

[lkaufman]

If we do not count that loss then lc0 has 100% against humans today but most games against weak players.

I see 19 wins in rapid when losses at faster time control are only against bots.

Most of the losses were actually against latest SF at full strength at knight odds, which is ridiculous. In some of the games the operator deliberately blundered back the extra piece or so in the opening, with mixed results. But of course Leela with gigantic Contempt can't expect to play even with Stockfish from a roughly balanced position. In bullet (1'+1") it has played something like 200 games against players over 2500, several vs. 2900 or even near-3100, and excluding a disconnection or crash loss I believe it has only given up two draws!