Author: Uri Blass
Date: 14:14:05 03/15/02
Go up one level in this thread
On March 15, 2002 at 16:51:28, Joshua Lee wrote: >I like the idea of longer time controls but i think their should be more games. >4 games in rebel's instance is rediculous. Anything can happen in 4 games, Also >the more games the more statistically accurate the performance rating. > >Also I am not impressed with Chess Tiger beating a couple of people with an >average rating UNDER 2450!!! Just because it got 2700+ doesn't mean it would = >a GM norm as the average elo was nowhere near as high as it needs to be. Maybe it is not a GM norm ny the rules(I am not sure) but I am sure that if you add to the result of tiger 2 more losses against 2700 players the performace is going to continue to be more than 2600 and the result is going to be considered as a GM norm. I am also sure that tiger cannot do worse than 0 out of 2 against 2 2700 players so for me the result is clearly a GM norm. > >Junior got a norm atleast if FIDE would acknowledge it. However 9 games is not >enough either to tell anything. For instance IF the SSDF rating list is accurate >Junior should've been higher and counting the difference 450Mhz to 4130Mhz it >should be 180 elo higher than the rating they gave Junior 6 on a 450, but it is >not. So someone's ratings are wrong it's either FIDE, SSDF, or whoever said that >for every trippling of clockspeed you get 90elo. I think that the assumption that you get 90 elo for trippling the clockspeed is wrong because I beleive in diminishing returns. Another point is that the guess that the rating of top programs is at least 2600 is based on games against humans and not on comp-comp games. > >Another thing to note is that Junior and Rebel lost to 2600+ rated players, that >means in those games it played like a 2200+ And back to Chess Tiger it's >opponents statistically would hardly ever win a game against someone 2700+ >And this rating would NOT stay the same if 20+ games were played. > >I don't know how many games shrink the margin of error to where we can as >accurately as possible say yep this program will always be 2700 or 2500, but i >tell you what 4 games is not the correct answer. Rebel did do something >impressive ,it did play more than enough games but the clockspeed's were >different. If someone could point out the most games on the same system that >would be appreciated. It did score over 2400 but as i said before this rating >suffered because not all opponents had the fastest computer opponent at the time >some were on 300Mhz some 500 etc. > >I know you believe Deep Blue to be weaker than Fritz, What about it's >performance in 6 games 2800+ If that rating is accurate Fritz would be >underrated. You had said if the two were on equal hardware, well what would be >good enough for Fritz to Search as needed to be at the level of DB? I believe that humans have today advantage that they know the computers that they play against them when kasparov did not know deeper blue and could not buy it so the comparison is not fair. I believe that Fritz and even Junior with 4000Mhz are better than Deeper blue and Junior already used more than 4000Mhz in durtmond. Uri
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.