Author: Vincent Diepeveen
Date: 06:47:07 06/26/01
Go up one level in this thread
On June 25, 2001 at 11:12:09, Robert Hyatt wrote: >On June 25, 2001 at 03:32:24, Uri Blass wrote: > >>On June 25, 2001 at 00:52:23, Sandro Necchi wrote: >> >>>On June 24, 2001 at 15:15:21, Uri Blass wrote: >>> >>>>On June 24, 2001 at 14:27:12, Sandro Necchi wrote: >>>> >>>>>On June 24, 2001 at 03:25:13, Uri Blass wrote: >>>>> >>>>>>On June 24, 2001 at 00:33:34, Sandro Necchi wrote: >>>>>> >>>>>>>On June 22, 2001 at 10:58:35, Robert Hyatt wrote: >>>>>>> >>>>>>>>On June 22, 2001 at 09:36:17, Vincent Diepeveen wrote: >>>>>>>> >>>>>>>>> >>>>>>>>>chessbase could get good PR out of this by focussing on the fact that >>>>>>>>>deepfritz will search way deeper as deep blue did! >>>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>>Not a single chance in hell this would happen. >>>>>>> >>>>>>>Hi, >>>>>>> >>>>>>>yes, you are, as always, 100% correct. >>>>>>> >>>>>>>I really do not understand how people can think a chess program which can play >>>>>>>about 2200 in close positions and 2600 in open positions (extimated ratings) >>>>>>>against one of the strongest player of the World can have any chance expecially >>>>>>>since the human player will know exactly what the opponent will play. This would >>>>>>>bring down the strenght of the program of at least another 200 points! >>>>>> >>>>>>Fritz is better than 2600 in open positions and 2200 in closed position. >>>>>> >>>>>>2900 in open positions and 2500 in closed position seems to be more realistic. >>>>> >>>>>No if talking about human Elo. I do not believe it. >>>>>> >>>>>>A computer with rating of 2600 in open positions and 2200 in closed position >>>>>>could not get performance of more than 2400 against humans even if the humans do >>>>>>no special preperation and old Fritz got prerformance of more than 2400 in the >>>>>>israeli league on hardware that is clearly slower than the hardware against >>>>>>kramnik even after part of the humans did special preperation. >>>>> >>>>>Ok, but the computers are very good against chess players up to 2500 Elo, so >>>>>this may give the impression they play about 2600 as an average. >>>>>Against stronger opponents (2650 and above) they would score less. >>>> >>>>The opposite >>>>Against strong humans they score more points >>>>The best performance of computers against humans was not against weak 2400 >>>>players but against players with average rating of 2702. >>> >>>At blitz and fast games not at slow games. >> >>No >>I am talking about the best performance at tournament time control(if you do not >>include deeper blue) >> >>Deep Junior got this best performance of 2702 at 2 hours/40 moves games in >>dortmund. >> >>Uri > > >So? Cray Blitz was the first computer to win an "open section tournament", back >in 1981. Yet it should have lost to a 1700 player in that event, as he had it >down to -3 at one point, but a single mistake let the machine escape with a >win. > >Computers can put up a "better than real life" result from time to time. It >doesn't mean they are super-GM players however. After the 1981 event, people >started taking CB (and computers in general) more seriously and things got much >more difficult after that. I agree here with Bob. If you play a zillion tournaments with a program against 2000 players then the 2200 players in that tournament will nail you as they don't give away material. the 2000s just get now and then a draw easily. programs have a few drawbacks. If i know from a person that he's bad in tactics, then i open position. the only luck is that most humans do not know the difference between fritz and tiger, but like coming dutch open championship where i join, if i ask a random participant of the open championship group whether a certain Israeli player (where i have never heart of) where he's good at, then i hear directly that he sucks for example in tactics but that his openings are more than ok. So against humans they focus on some points where they are weak, but against computers they do not. The reason is that except in the GM - rebel challenge, that no GM gets paid to beat a computer. Even kramnik isn't paid to beat it with a big number. In fact a 4-4 would be nicer for kramnik to get (from money viewpoint) as a 8-0 score. The problem is of course that the organizer of such a match is also financially involved in not letting the program look too bad, so in short the only info we have is from the Rebel - GM challenge where they get only paid if they draw or beat it. Here you suddenly see very exciting games out of boring openings, after which the GM even from positions where computer says +1.0 for itself still loses. Programs simply have a few weak spots where they are so weak that a 2000 player can profit. At the same time, the 2000 player usually blunders tactical too much, whereas a 2285 rated player like me is already able to play a game aversus a computer without blundering bigtime. Other proof is from yearly matches as organized by Jan Louwman. There even top programs of today can't win from a 2200 player. the other boards 2000- rated players are sitting there, and usually they all lose, sometimes 1 draw there. Usually the score is like 6.5- 1.5 or 7-1. If you consider that a 500Mhz laptop isn't that slow with shredder at it, or genius or whatever commercial program, then the TPR of this 2200 rated player is *amazing* high seemingly. Winning versus 2200+ is a major problem in short if he has experience against programs, not to mention if he knows the difference between a program X and program Y. On the other hand there are also players which are good in openings, sometime even 2600 rated, who completely suck technical and in middlegame. Basically they win games because of their superior opening. Now this is nowadays quite risky against computers. Though some games they win annihilating their opponent, usually you need a *bit* of tactics to finish off your opponent. Such players will always have and remain having a score near to 0% against programs. Best regards, Vincent
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.