Author: Robert Hyatt
Date: 12:02:10 09/20/99
Go up one level in this thread
On September 20, 1999 at 09:11:28, Lanny DiBartolomeo wrote: >On September 20, 1999 at 05:48:28, Enrique Irazoqui wrote: > >>On September 20, 1999 at 03:02:58, Bertil Eklund wrote: >> >>>On September 19, 1999 at 09:37:36, Enrique Irazoqui wrote: >>> >>>>On September 18, 1999 at 17:44:11, Stephen A. Boak wrote: >>>> >>>>>By the way, the Rebel Century Performance Rating for the match today was: >>>>> >> 2553 << >>>>>which is certainly in the range of Grandmaster ratings. >>>> >>>>Rebel's TPR so far is roughly 2480, which is not GM level. Ten games are not >>>>enough to know much, but taking into consideration that Rebel is the program >>>>that scored the best in Aegon over the years, this tpr seems to indicate that >>>>programs do not reach GM level yet at slow time controls and against motivated >>>>GMs. One more thing: considering that Rebel played these games with a hardware >>>>much faster than a P200MMX, it seems clear that the SSDF list is quite inflated, >>>>maybe by some 150 point >>> >>>Are you the same Enrique that have played computerchess, for the last 20 years!? >> >>Hi Bertil, >> >>I already said that "Ten games are not enough to know much". Therefore I am not >>"so sure". After the games we have so far, I posted some thoughts, that's all. I >>did not say I had evidence. And it is not only a matter of quantity, but of >>looking into the games to see what has been going on. Any chess player could >>have done this, with or without 20 years experience in computer chess. >> >>>How can you be so sure after 10-12 games! If only one or two of the games went >>>another way should it have proven that Rebel was of GM-strength. >> >>Maybe, but they didn't go the other way. And you are only talking quantities, >>while I am also trying to include the quality of the games, as seen by me. It is >>the combination of both that make me believe that so far programs are probably >>and in my opinion not GMs. >> >>>We can take the games for Fritz5 and use them and we have 2600+ something or >>>CM6000 >>>about 2600 or maybe ShredderX one game 2600 or maybe Hiarcs4 on P200MMX about >>>2500 or Hiarcs6 about 2600 in south-america. Rebel8 played two GM´s in Sweden >>>One win one draw, shall we use this for a guess?. >> >>I think it is best to count current games. Human players know programs better >>than before and in these games they are motivated, something that has not been >>always the case. My opinion, of course. >> >>>Is this Rebel-challenge the >>>only valid rating-basis? The match favor the GM´s also, they can prepare for the >>>opponent and it´s never any problem with the time-controls for the GM´s, as is >>>the case in a face to face match. Once I played Rebel9 against Gandalf3 after 7 >>>games Rebel9 was up 7-0. 8 games later the score was 7,5-7,5. >>>Mr Szub is also sure that Junior4.6 is better then Junior5. He knows that after >>>studying its style and a few games. I guess that he can be wrong this time. >>> >>>The list should be correlated to the reality but we need a better base then >>>feelings and guesses. I guess that the base of the list is to high but is it 50 >>>or maybe 250 points, it´s always fun to speculate! >> >>Yes, Bertil, I agree. I don't pretend the SSDF to drop 150 points just because I >>guess you should. In the post you are answering to I say your list "seems" >>inflated by some 150 points, not that "it is" with all evidence. >> >>Now for more speculation (why not? It may be interesting sometimes). Rebel 9 has >>a rating of 2525 in your list. Rebel 10-5 has been playing with a hardware that >>is roughly 3 times faster as an average than your P200MMX, which translates into >>some 100 points. I don't believe that any program is a top GM of 2625, and about >>this I am convinced, even without evidence to show in court. We both "guess" >>that the list is inflated. The 150 points excess sounds right to me. That's all. >> >>Enrique >> >>>Regards Bertil SSDF >>> >>> >>>>I think that someone has been saying all this for years. Hi Bob! :) >>>> >>>>Looking at the few 40/2 games played so far by programs against strong human >>>>opponents, I wonder if results wouldn't be similar if played against 2300 >>>>people. The positional superiority of a 2300 player is still immense, and for >>>>them it might be a matter of avoiding tactics, as wise IMs and GMs do when >>>>playing computer programs. Maybe the Elo system works differently for programs? >>>> >>>>Aside form this, I don't think it makes sense to use the same opening book in >>>>comp-comp and in human-comp games. It is quite absurd to play openings that lead >>>>to positional games, where programs are quite dumb, and this is happening too >>>>often. Is it not a better idea to build a gambit-like book that tries to open >>>>the game and play tactics? Same for playing style. A program can afford to play >>>>the Orthodox against another program, because neither will understand a thing, >>>>but against a strong human player it's a mistake. Look, for instance, at Rebel-2 >>>>yesterday. >>>> >>>>Enrique > > Do you think preparing a month or more in advanced for a chessprogram could >really come near seeing a reflection of true strength? I think for a program to >have this high a rating (2480) after 10-12 games is amazing how many people here >feel they would have the same rating if everyone got to study and prepare >against you very minimum your dropping at least 100pts ever go into a lower >rated players pet line?? This is what it amounts to. (in my opinion) There is an interesting counter-argument. Several people claim that when the 'automatic' programs play on ICC, they have a big advantage, even if the human plays with a reasonable increment. Agree or disagree? If you agree, what about the fact that a GM can play dozens or hundreds of games, can watch the computer play others, can study the openings it tends to play, the mistakes it tends to make, etc. IE wouldn't you also agree that by playing an unlimited number of games, in a public place, that keeping your electronic player's rating at 3100 is _very_ difficult? One of the two issues is not quite correct... Most overlook the difficulties of an automatic program playing all comers as many games as they want... and many play lots. IE I have watched Crafty vs Kamsky today go way over 24 games non-stop. I'd bet Crafty would clean his clock in a surprise meeting where he had not played it in 6 months. It is amazing that it can _still_ clean his clock even though he has played hundreds of practice games against it. So far, after 31 games, crafty has lost 2 and drawn 4. Rest you can guess. :)
This page took 0.01 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.