Author: blass uri
Date: 23:33:56 12/29/99
Go up one level in this thread
On December 29, 1999 at 14:08:29, Albert Silver wrote: >On December 29, 1999 at 10:19:25, Bertil Eklund wrote: > >>On December 29, 1999 at 06:56:23, Graham Laight wrote: >> >>>Hi Bertil! >>> >>>Speaking as someone with insider knowledge of the SSDF ratings, do you think >>>that, at the top end, the ratings might have become too high? >>> >>>In the past, did you used find that, once the SSDF list had been calibrated >>>correctly, the ratings tended to stay at about the correct level by themselves, >>>or, when you tested against human players, did you then have to make significant >>>corrections to the scale? >>> >>>At the moment, there seem to be quite a few people about who strongly feel that >>>the top end of the SSDF list has gone too high (in relation to the FIDE Elo >>>rating list) - and that maybe the top computer programs should be rated more in >>>the region of 2450 than 2696. >>> >>>What are your feelings on the subject? >>> >>>Thanks for your time and thoughts on this. >>> >>>-g >>Hi! >> >>Yes I think the list is inflated but only 20-30 elo. Enrique checked some >>programs from the past vs today and the average decrease in elo was 18 points. I >>have did the same and found something between 20-30 points. >> >>However we have in this forum a couple of high-reputated (not without reasons in >>other cases) persons that over and over tries to compare the programs >>performance in single-game matches with increment time controls and sometime >>double increment time controls. I guess that humans can perform 50-100 elo >>better than under normal tournament conditions. As I sometimes play in >>tournaments I know the performance in round 9 is much worse than in round 1. You >>are tired, maybe you have dined and wined a little bit to much the day before >>and so on. I only know off one comp-human tournament this year in south-america >>where Rebel10, CM6000 and Hiarcs6 performed 2603 on P2-400 (average), someone >>say less motivated players but you can´t be fully motivated in 10-11 rounds. Yes >>the players rating was only 2080-2450 or so but the computers didn´t loss a >>single game. If you play in a tournament against 3-4 comps and 7-8 humans it´s >>impossible to prepare as you can do against a computer in this case Rebel that >>is a static program and always play the same moves in contrast to Nimzo or Comet >>that never play the same game twice. >> >>Conclusion: >> >>The list could be slightly inflated. >>It´s based on TOURNAMENT games, not single-game matches with Internet delays. >>If players practise a lot against computers and especially against a single >>program (s)he can learn the style and perform better, and that´s the main reason >>for the differnt results. >> >>IMO >> >>Regards Bertil SSDF > >I'm not going to get dragged into this again, but I will say that if a program >is said to be playing at 2600, then test it against 2600 players and see how it >does. Showing it can beat up players rated 200-400 points less means very >little, even if it can be entertaining. Just my 2 cents. I'm aware that can be >incredibly difficult to arrange, but I see no other way around this, if the >truth is what is being sought. Being 2600 means more than beating 2400 players, >it means holding your own against 2600 players, no? I think this upcoming >Israeli leagure in which both Junior 6 (Deep Junior?) and Rebel Century will be >playing should be quite instructive as the human opponents, due to the nature of >the event, will be very motivated. > > Albert Silver I think that the performance of Rebel,Deep Junior and Fritz in the Israeli league is going to be lower than the real rating of them because of the fact that the opponent teams can choose the human to play against the computer. 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.