Author: Peter Kasinski
Date: 10:41:04 08/27/99
Go up one level in this thread
On August 27, 1999 at 11:16:22, Robert Hyatt wrote: >On August 27, 1999 at 04:02:50, Bernhard Bauer wrote: > >>On August 27, 1999 at 00:09:35, Robert Hyatt wrote: >> >>>On August 26, 1999 at 21:34:13, odell hall wrote: >>> >>>>On August 26, 1999 at 14:01:59, Robert Hyatt wrote: >>>> >>>>>On August 26, 1999 at 12:55:01, odell hall wrote: >>>>> >>>>>>On August 26, 1999 at 10:36:56, Robert Hyatt wrote: >>>>>> >>>>>>>On August 26, 1999 at 08:41:37, odell hall wrote: >>>>>>> >>>>>>>>On August 26, 1999 at 07:32:06, Claudio A. Amorim wrote: >>>>>>>> >>>>>>>>>On August 26, 1999 at 02:21:39, odell hall wrote: >>>>>>>>> >>>>>>>>>>Hi >>>>>>>>>> >>>>>>>>>> If anyone is interested in how crafty would do against Grandmaster Atopkian do >>>>>>>>>>a Search Crafty Vagr on icc. After these Games were played I asked Akopian What >>>>>>>>>>happened surprised that he lost. He said he was experimenting with some opening. >>>>>>>>>>However he admitted that he could not beat crafty and claimed this fact as the >>>>>>>>>>reason he played it so few games. Ofcourse these were all blitz games. Akopian >>>>>>>>>>said that playing the computers at 40/2 would not be interesting for him because >>>>>>>>>>they would be no challenge. >>>>>>>>> >>>>>>>>>I think Akopian is plain wrong on this matter. Playing the best programs at 40/2 >>>>>>>>>is already a challenge even to the very best human players in the world, and >>>>>>>>>Akopian is hardly in that league (Kasparov, Anand, Kramminik, etc., etc.). FIDE >>>>>>>>>Championship, these days, is a display of physical will and blitz wizardy. It >>>>>>>>>has little to do with top level chess. >>>>>>>>>I'd love to see a tournament involving the greatest human and the better >>>>>>>>>computers, round robin, $500.000 to 1st place. Matters should be more clear, >>>>>>>>>them. >>>>>>>>> >>>>>>>>>Regards, >>>>>>>>> >>>>>>>>>Cláudio. >>>>>>>> >>>>>>>> >>>>>>>> While I do believe programs are low GM Strength I don't think any program out >>>>>>>>there could defeat a 2600 Grandmaster like akopian at 40/2 in a match. I don't >>>>>>>>think one has to be garry kasparov , or annand to beat the best micros at long >>>>>>>>time controls. >>>>>>> >>>>>>> >>>>>>>You had better be careful... you are beginning to sound a lot like me. >>>>>>> >>>>>>>:) >>>>>> >>>>>> No Bob I have never maintained that Computers are 2600 strength. I do believe >>>>>>they are between 2500-2540, ofcourse this is a long way from where you stand, I >>>>>>believe you said they are around the low 2400's. In view of all the recent 40/2 >>>>>>games I doubt you still have such a low opionion of programs. >>>>> >>>>> >>>>>I have been fairly consistent saying 2450 is what I would peg as the upper bound >>>>>of today's programs... >>>>> >>>>>lets see, estimated TPR so far would be roughly 2200+2600+2600 (one loss, >>>>>rating-400, two draws, generously giving the opponents 2600 ratings). That >>>>>turns out to be 7400/3 which is 2466. Right in line with my speculation, >>>>>wouldn't you say? >>>> >>>> >>>> >>>> This May sound good, but there is only one major problem, You are >>>>conviently ignoring the results at the WCCC99 Where Fritz5 beat Sokolov along >>>>with the draws against 2600+ players, I bet if you factor in these number that >>>>2466 will disappear. Even if you were correct how can you make a elo claim >>>>based on three games, I am sure you know this is not accurate. COme on Bob Let >>>>go of the pride and admitt that computers are much better than you originally >>>>thought. You yourself admitted that Rebel's results so far are very good. >>> >>> >>>(1) you can't cherry pick. IE you can't pick a tournament where fritz does >>>well and use that, and ignore one where it gets torn up. I am using Rebel as >>>a reference point, because I am just taking _every_ game in Ed's GM challenge, >>>and not counting others. >>> >> >>We all? know that computers usually have *no* Elo rating. May be with the >>exeption of Junior. So all that kind of discussion is about TPR ratings. >>For a real Elo rating you have to consider all rated games. >> >>>Count them if you want, of course... but if you pick the right events, you >>>can prove anything you want... It's been done before... >>> >> >>Yes, if you pick the human event "MK Cafe Cup, Koszalin, Poland" you will see >>the american GM A. Shabalov rated at 2816 from 10 games and >>the german GM H. Teske rated at 2237. >>Data is from TWIC 249. >>The Elo rating of Shabalov is 2566 and Teske has 2515 acording to TWIC. >>So Shabalov has performed 250 points above his Elo rating, whereas Teske >>performed 278 points under his rating. >> >>>(2) you can use 3 games to produce a 'performance rating'. That is what it is >>>all about. And yes, I think 2466 is a very good result for Rebel. It is a bit >>>better than I expected... and is no disgrace at all. >> >>Those numbers doesn't say too much. Current programs play in some positions like >>GMs in other positions they play perfect, but sometimes they have no idea about >>the position and play like club players or they fail terribly due to holes. >> >>Kind regards >>Bernhard > > >Here is my 'position', for the record again. If you give a computer a real >position, it can do _very_ well. IE in the FIDE game last night, Crafty >predicted nearly _every_ move... and in the cases where it missed, the GM >clearly played a bad move (the game was likely drawn at one point, but black >folded his tent and lost pretty easily). So in that game, yes, Crafty looked >just like any other GM player. And that is surprising in a way, knowing what >it knows about chess. But a reasonable eval, and a fast search, when combined, >produce some pretty remarkable chess at times. So from this, you might think >that I am changing to say that Crafty is a GM? > >Wrong. Just because it plays like a GM in GM-like positions, doesn't mean that >it plays like a GM in _all_ positions. And _that_ is a serious challenge for >all of us... to play as well in anti-computer positions as we do in normal >chess positions. IE when one GM is attacking another, the computer will often >find better defensive moves than the GM does. But the problem is, that the >computer won't respond to the attack until it is too late, while the human GM >immediately notices what is happening and starts defending. So given a position >reached by a GM, the computer is stronger, quite often. But if it has to reach >such positions by itself, it doesn't have a chance to do so. > >That is the difference between computers and the GM-level, IMHO. They can win >many won GM positions. But they can't _reach_ those positions by themselves... > >_yet_. Good point. Being able to finish things off from a tactical position doesn’t guarantee arriving at that position in the first place. Same for finding key defensive moves when real challenge may be in recognizing that attacking is no longer a good idea. One example and one quote: 1. Still looking for a program that would play …Re6 in Reshevsky-Petrosjan from Zurich '53. Giving up exchange for blockade because the long-term outlook is grim. 2. I recall a quote by R.Spielman where he talked about A.Alekhine: “I can comprehend his combinations easily enough, but where he finds his attacking ideas and how he manages to infuse so much life into every opening, that is beyond me”. IMHO most micros would say the same thing if given a chance. PK
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.