lkaufman wrote: ↑Thu Jan 20, 2022 4:24 am
It occurred to me that thirty years ago, the program with the best results against human grandmasters in Rapid was getting performance ratings of roughly the same 2426 figure as I calculated for Dragon. Only difference: thirty years ago the games were played on equal terms, now they are knight handicap! Further curiosity: the engine was Socrates, which just like the original Komodo was a collaboration between Don Dailey and myself, and then as now it couldn't quite match the best program of its day in direct play, but did better against humans!
I just wonder if your old Socrates can defeat Komodo Dragon 2.6.1 at Knight Odds
Of course Socrates 30 years ago was running on 30 year old hardware, early Pentiums I think. I'm sure that Socrates running on an I7 would get a much higher rating than 2426 in human Rapid, maybe something like 2700 or so I imagine. Even so, I think it would come out behind at knight odds vs. Dragon. Even if they understand the principle of trading pieces when winning, engines don't understand the principle of avoiding tactical complications and unclear variations when winning. At least not normal A/B engines. I did find that it was very hard for Dragon to give knight odds to weakened versions Lc0 that looked at very few nodes; presumably they know that a simple position a piece up is better than a messy one.
Probably if you remember which version of Socrates that was somebody will have a version and test them out, another thing is that a human rated 2426 back in 1993 is only about a 2350 nowadays in comparison, so that might make up for the hardware difference
Even if the drop in elo levels is 100, as I believe from the data I've seen, the hardware advance over the last 30 years is far more than that. I don't remember the specific version of Socrates, but if anyone has "Kasparov's Gambit", which was based on Socrates around that time, that should be close enough.
There is a way that I know of making these two engines even, I have an intel i3 which I can use with Kasparov's Gambit engine in which the 100 Elo lost due to Fide rated players back in 1993 were at least 100 Elo weaker than what they are nowadays, and since the intel i3 will give Kasparov's Gambit an Elo increase of at least 100 compared to a Pentium 4F-series ,it will bring it back to 2426. I also have an intel i7 to use it with Komodo Dragon 2.6.1. The only disadvantage would be that since it is a DOS engine I do Not like to input each move for Kasparov's' Gambit https://www.myabandonware.com/game/kasparovs-gambit-1zl
I managed to play one game at Knight g1 Odds Versus Stockfish 14.1, Stockfish had White and Kasparov's Gambit Black this is the output of Kasparov's Gambit. Both Stockfish 14.1 and Komodo Dragon 2.6.1 are too strong for Kaspavov's Gambit at Knight Odds. The T/C 15 Minutes 10 Sec Bonus, I used DOSBOX Version 0.74-3 here is how to set it up
Chessqueen wrote: ↑Fri Jan 21, 2022 10:02 pm
I managed to play one game at Knight g1 Odds Versus Stockfish 14.1, Stockfish had White and Kasparov's Gambit Black this is the output of Kasparov's Gambit. Both Stockfish 14.1 and Komodo Dragon 2.6.1 are too strong for Kaspavov's Gambit at Knight Odds. The T/C 15 Minutes 10 Sec Bonus, I used DOSBOX Version 0.74-3 here is how to set it up
So Stockfish gives away a pawn on move two, then goes out of its way to trade queens down a knight and pawn, and somehow still wins? Pretty bizarre. I guess the strength gap is just so huge that even starting in this ridiculous way isn't enough for Black. I think that a normal human player of say 2200 Elo would win this position after the queen trade vs. Stockfish. I believe I know why Stockfish plays so poorly when giving piece odds, but I'll leave it to their team to figure it out.
Chessqueen wrote: ↑Fri Jan 21, 2022 10:02 pm
I managed to play one game at Knight g1 Odds Versus Stockfish 14.1, Stockfish had White and Kasparov's Gambit Black this is the output of Kasparov's Gambit. Both Stockfish 14.1 and Komodo Dragon 2.6.1 are too strong for Kaspavov's Gambit at Knight Odds. The T/C 15 Minutes 10 Sec Bonus, I used DOSBOX Version 0.74-3 here is how to set it up
So Stockfish gives away a pawn on move two, then goes out of its way to trade queens down a knight and pawn, and somehow still wins? Pretty bizarre. I guess the strength gap is just so huge that even starting in this ridiculous way isn't enough for Black. I think that a normal human player of say 2200 Elo would win this position after the queen trade vs. Stockfish. I believe I know why Stockfish plays so poorly when giving piece odds, but I'll leave it to their team to figure it out.
Stockfish prefers giving a different pawn after some search time
[fen]rnbqkb1r/pppppppp/5n2/8/3P4/8/PPP1PPPP/RNBQKB1R w KQkq - 1 2[/fen]
Here is the 2nd game in Arena Format PGN, Stockfish was using my Google Pixel 6 Pro, and Kasparov's Gambit was using my Intel i7 T/C 15 Minutes 10 Sec Bonus. In both games Stockfish 14.1 NNUE managed to trade Queens and took away Gambit opportunity to Castle, just before Gambit resigned, Stockfish announced mate in 8 on my Google Pixel 6 Pro Cell Phone which use a Tensor CPU which is NOT the fastest CPU for Cell phone, the iphone 13 has the fastest CPU at the moment, the Google Pixel6 Pro is like the 4th or 5th fastest Cell Phone
Just to give you an idea of how fast is the Tensor chip for the Google Pixel6 Pro in comparison to an intel i7, , Stockfish 14.1 NNUE was playing with my Google Pixel 6 Pro Cell Phone and Komodo Dragon 2.6.1 with my intel i7
A good Solid 2400 human player that play like Komodo Dragon when I set the UCI_Elo = 2400 should NOT lose a game against any strong engine like Stockfish 14,1 NNUE, but there are NOT too many solid strong Humans Fide players that play like Komodo Dragon 2.6.1 set up to Limit Playing strength = 2400
