When will Dragon give a Knight Odds in FRC to a GM ?

[Chessqueen]

With Certain positions where the King and Rooks are place in its original position, when will Dragon give a Knight Odds in FRC to a GM ? Do you believe that Dragon can finally give a GM rated around 2660 to 2690 a Knight Odds. ?

[Uri Blass]

At what time control?

Of course Dragon can win if the time control is fast enough.

[Chessqueen]

At what time control?

Of course Dragon can win if the time control is fast enough.

This is what Mr. Kaufman wrote.
GM Lenderman had enough trouble with Komodo in FRC Rapid at knight odds, and that was just before NNUE and Rook for knight is only about half of knight odds. I know that the Rapid rating difference between SF or Dragon and Lenderman is about the 940 elo or so I reported for SF self-play at rook for knight FRC, but when the odds-receiver is 2600 rather than 3500+ the handicap value will be much lower, though hard to say how much lower. Anyway I don't think that even Magnus Carlsen stands a chance in Rapid with FRC rook for knight odds against SF or Dragon, though he would surely get a reasonable number of draws. Interestingly, I tried a few randomly chosen positions from the FRC rook for knight file to see how recent SF evaluates the win probability in the initial position with a deep search, and in all cases tried it showed White just getting 1% draws with no wins, which works out to -920 elo, roughly confirming the actual playout results.

[lkaufman]

With Certain positions where the King and Rooks are place in its original position, when will Dragon give a Knight Odds in FRC to a GM ? Do you believe that Dragon can finally give a GM rated around 2660 to 2690 a Knight Odds. ?

Here are some results using the new FRC odds files. Note this is for FRC in general, not just the positions with rooks and kings on normal squares. I ran recent Stockfish dev against Critter 1.6a on one thread at 2' + 1". The CEGT blitz rating for that Critter is 3026 (I use CEGT rather than CCRL because CCRL uses BayesElo which is not mathematically consistent with adding Elo gains from matches calculated by actual elo formula). With rook for knight odds (White missing the rook), Stockfish won by 266 elo, suggesting that a CEGT blitz rating of 3292 would be an even match for this SF (which should be rated about 3650 or a tad higher on CEGT blitz) at Exchange odds FRC. So Exchange odds appears to be worth a bit over 350 elo under these conditions, which sounds plausible to me. But the same test at (White) knight odds led to Critter winning by 318 elo, suggesting that a CEGT blitz rating of 2708 would be an even match for SF under these conditions.

How to interpret these results for human opponents of a top engine? Well, CEGT ratings are much stricter (lower) than FIDE ratings, especially at fast time controls, so a 2708 CEGT blitz engine would crush a similarly rated human at 2' +1", and would probably even defeat 2900 rated Nakamura or Firouja, especially in FRC. I think it is roughly accurate (at top human level) to say that a CEGT engine playing blitz (2' + 1") is comparable to a similarly FIDE rated human playing Rapid (15' + 10"). In general, humans do much better than comparable engines when getting a handicap, but much worse in FRC, so perhaps these two factors roughly offset each other. If so, Then the engine rating at CEGT blitz should perform about the same as the human FIDE rating playing Rapid when receiving a handicap in FRC. If all this is correct, this means that recent SF should be an even match with a 2708 FIDE human in Rapid playing FRC and giving (White) knight odds. Komodo Dragon with the best net for odds play should do a bit better. Although there are a lot of assumptions here, this 2700+ figure seems about right to me, based on the old Lenderman games and the progress since then.

[Look]

With Certain positions where the King and Rooks are place in its original position, when will Dragon give a Knight Odds in FRC to a GM ? Do you believe that Dragon can finally give a GM rated around 2660 to 2690 a Knight Odds. ?

Here are some results using the new FRC odds files. Note this is for FRC in general, not just the positions with rooks and kings on normal squares. I ran recent Stockfish dev against Critter 1.6a on one thread at 2' + 1". The CEGT blitz rating for that Critter is 3026 (I use CEGT rather than CCRL because CCRL uses BayesElo which is not mathematically consistent with adding Elo gains from matches calculated by actual elo formula). With rook for knight odds (White missing the rook), Stockfish won by 266 elo, suggesting that a CEGT blitz rating of 3292 would be an even match for this SF (which should be rated about 3650 or a tad higher on CEGT blitz) at Exchange odds FRC. So Exchange odds appears to be worth a bit over 350 elo under these conditions, which sounds plausible to me. But the same test at (White) knight odds led to Critter winning by 318 elo, suggesting that a CEGT blitz rating of 2708 would be an even match for SF under these conditions.

How to interpret these results for human opponents of a top engine? Well, CEGT ratings are much stricter (lower) than FIDE ratings, especially at fast time controls, so a 2708 CEGT blitz engine would crush a similarly rated human at 2' +1", and would probably even defeat 2900 rated Nakamura or Firouja, especially in FRC. I think it is roughly accurate (at top human level) to say that a CEGT engine playing blitz (2' + 1") is comparable to a similarly FIDE rated human playing Rapid (15' + 10"). In general, humans do much better than comparable engines when getting a handicap, but much worse in FRC, so perhaps these two factors roughly offset each other. If so, Then the engine rating at CEGT blitz should perform about the same as the human FIDE rating playing Rapid when receiving a handicap in FRC. If all this is correct, this means that recent SF should be an even match with a 2708 FIDE human in Rapid playing FRC and giving (White) knight odds. Komodo Dragon with the best net for odds play should do a bit better. Although there are a lot of assumptions here, this 2700+ figure seems about right to me, based on the old Lenderman games and the progress since then.

Actually "Firouzja" is the right spelling.