This means that sf14 or Dragon2 are somewhere around 4000 FIDE blitz? I don't think it's too high unless someone denies me, I'm just realizing that today's engines are really close to the phantom 4000 mark and in the future they will also be in longer TC, Referring to the human scale, elo is relative after all.lkaufman wrote: ↑Fri Aug 27, 2021 5:35 amFinal result after 2000 games by Dragon 2 Skill 24 single thread at 2' + 1" (500 each vs Rybka 2.3.2a 64 bit, Wasp 2, Wasp 1.01, and Fruit 2.2.1, average CCRL blitz rating 2887.5) was a 2905 performance. Probably that's close to 3200 on the human FIDE blitz scale, meaning that Carlsen or Nakamura might score about 15% in blitz against it. It was averaging about 33 milliseconds per move, roughly 100 times faster than the normal speed at which a 2' + 1" game would be played. So this suggests that single thread Dragon could give 100 to one time odds (without pondering!) to the world's two top human blitz players and still score about 85%. Or it could give 100 to 1 time odds to Rybka 2.3.2a and score about 40%, which is something given that Rybka 2.3.2a was clearly the world's best engine in 2007, well above the level of any human, able to give pawn odds successfully at serious time limits to famous grandmasters. Incredible progress, no one could have imagined this.
GM Anand playing Vs Komodo ( Anand Bot) ?
Moderators: hgm, Rebel, chrisw
-
- Posts: 85
- Joined: Mon May 13, 2019 6:05 pm
- Full name: Marco Giorgio
Re: GM Anand playing Vs Komodo ( Anand Bot) ?
-
- Posts: 5960
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: GM Anand playing Vs Komodo ( Anand Bot) ?
I mean standard scoring, since we have already defeated a “par” GM (Wirig) with Armageddon scoring.jorose wrote: ↑Sun Aug 29, 2021 4:45 amWith Armageddon or standard scoring? Anthony Wirig would have won his match with standard scoring and I think the two categories are separate challenges. In one format you are trying to find the toughest defence and in another it is the toughest defence while keeping the ball in play.lkaufman wrote: ↑Sat Aug 28, 2021 4:41 pmI don't think Nakamura would want to replay that match, even with a little more time. Chess.com isn't interested in sponsoring events with long time controls like this, they prefer matches that fall within the "blitz" range. I think that we have convincingly graduated from the two White pawn handicap, although of course I wouldn't turn down a match with a strong GM if we had a sponsor. We could try the same two pawn handicaps but as Black rather than White against a strong GM at normal Rapid (15' + 10"), that might be about fair. But the prime focus for Dragon vs. human matches must be knight odds, with the time limit and the rating of the grandmaster as variables. Beating a "par" (FIDE 2500 or a bit more) GM in a Rapid match or a top GM in a blitz (not bullet) match at knight odds are the two big goals we can realistically hope for in the next couple years.Chessqueen wrote: ↑Sat Aug 28, 2021 4:56 amIs it hard to get GM Nkakamura to play versus Dragon 2 at TC 30'+10'" ( 2 pawns odds) or you have to offer a strong GM from India like GM Pentala Harikrishna ?lkaufman wrote: ↑Sat Aug 28, 2021 3:03 amDragon 2 does use NNUE in the skill levels, but we reduced the search depth (by one ply in the depth 20 to 23 range) to compensate as well as significantly increasing the degree of randomization. Despite these changes, the skill levels in Dragon 2 are somewhat stronger (at least in the top range, 20 and up) than pre-Dragon (Komodo 14.1), maybe about fifty elo on average.Marcus9 wrote: ↑Fri Aug 27, 2021 5:56 pm Not bad, 33ms is also the time between frames in a 30fps video, in fact the responses seem really instant!
Unfortunately I don't have dragon but only komodo 13, did Dragon use NNUE or it's just "normal" komodo at fixed depth? In the first case I expect the higher levels of Dragon to have had an increase in elo over the pre-Dragon version Otherwise it could use the NNUE net to reach the same force at lower depth and move faster, But I realize that it might be awkward to have versions of komodo where the strength of the levels varies between version and version.
Komodo rules!
-
- Posts: 5960
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: GM Anand playing Vs Komodo ( Anand Bot) ?
With sf14 on 8 threads at 3736 on Ccrl blitz list, and if we accept my estimate of adding 300 for human fide blitz then we’re already over 4000 fide blitz. But only add 100 to Ccrl rapid list, then move 20% towards 2800 to estimate fide human classical tc rating, so maybe 400 elo to go or so. It seems possible but several years from now.Marcus9 wrote: ↑Sun Aug 29, 2021 12:53 pmThis means that sf14 or Dragon2 are somewhere around 4000 FIDE blitz? I don't think it's too high unless someone denies me, I'm just realizing that today's engines are really close to the phantom 4000 mark and in the future they will also be in longer TC, Referring to the human scale, elo is relative after all.lkaufman wrote: ↑Fri Aug 27, 2021 5:35 amFinal result after 2000 games by Dragon 2 Skill 24 single thread at 2' + 1" (500 each vs Rybka 2.3.2a 64 bit, Wasp 2, Wasp 1.01, and Fruit 2.2.1, average CCRL blitz rating 2887.5) was a 2905 performance. Probably that's close to 3200 on the human FIDE blitz scale, meaning that Carlsen or Nakamura might score about 15% in blitz against it. It was averaging about 33 milliseconds per move, roughly 100 times faster than the normal speed at which a 2' + 1" game would be played. So this suggests that single thread Dragon could give 100 to one time odds (without pondering!) to the world's two top human blitz players and still score about 85%. Or it could give 100 to 1 time odds to Rybka 2.3.2a and score about 40%, which is something given that Rybka 2.3.2a was clearly the world's best engine in 2007, well above the level of any human, able to give pawn odds successfully at serious time limits to famous grandmasters. Incredible progress, no one could have imagined this.
Komodo rules!