Author: Robert Hyatt
Date: 06:48:08 08/03/98
Go up one level in this thread
On August 03, 1998 at 00:43:52, Jouni Uski wrote: >I fully respect Hyatt in his knowledge about computer chess. But when he >says PCs are still not at GM level in long play I totally disagree! > >Here in Finland we have 3 GMs and they are almost helpless aginst any TOP >PC program even in P90. And don't forget, that Mchess Pro 6 won GM Nesis >4 - 2 last year in match (it was at 90 + 90 minutes games = quite near 40/2). > >And Fritz 5 current SSDF rating 2573 speaks himself - even if we substract 70 >points for safety it's still over 2500 = GM level. > >Jouni A couple of notes: 1. if you have 3 GM's that are "almost helpless" against any TOP PC program, even on a P90, you have some very weak GM's there. Note I am not talking about blitz. A month ago I would have guessed that a top GM player would still be superior to programs in blitz, but I took the time to analyze several thousand crafty vs GM games and discovered that *no* GM playing on either ICC or chess.net can even score 25% against Crafty, which is surprising. But it seems to be real. But at 40/2, I can't imagine any GM being "helpless" against any program, because that would suggest that the programs are 300-400 points *better*. And that I know is not true. 2. the 2573 from fritz means *zero*. IE subtract 500 points from everyone on the SSDF list, and the ratings are *still* correct, for that pool of players. Or add 500 for that matter. It is the *spread* between the numbers that is useful, not the absolute values. And since the SSDF numbers are not normalized to any standard rating base (FIDE, USCF, BCF, etc) there is no way to take a number like 2573 and say that equates to GM in FIDE... It would seem to me that Fritz vs Anand (total wipe-out) and Rebel vs Anand (Rebel had winning chances) shows that the SSDF numbers are a little out of whack when you normalize to human opponents rather than computer vs computer results. And when you factor in that in rebel/fritz vs anand, the result was 3-1, that is +200 for anand. With too few games to see how accurate that is, of course. But, opinions count. You have yours. I have mine. If we see enough games, one will prove true, of course...
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.