Author: Robert Hyatt
Date: 20:19:20 05/22/99
Go up one level in this thread
On May 22, 1999 at 21:36:28, Mark Young wrote: >On May 22, 1999 at 20:58:54, Robert Hyatt wrote: > >>On May 22, 1999 at 19:51:15, Mark Young wrote: >> >>>On May 22, 1999 at 19:07:15, Rajen Gupta wrote: >>> >>>>seeing the way GM rohde crushed rebel 10 I see very little chances for H7.It >>>>will be running on a much slower machine, >>> >>>This is incorrect, a G233 is not a much slower machine then a PII 450. The G233 >>>is equal to about a PII 350 running Hiarcs7. >>> >>>it is not as strong (against humans) >>>>any way as rebel 10, >>> >>>This is a unknown, Hiarcs7 may be better... >>> >>>(supposed to be a relatively weak defender) >>> >>>This is inaccurate, Hiarcs7 in my test has shown itself to be a very strong and >>>cunning defender playing humans and other programs. >>> >>> The Yerminator >>>>would have had enough time to seek out the holes in H7 's defences >>> >>>This is very true, but this is what makes this match the most interesting over >>>the other matches played. >>> >>> and besides >>>>the Yerminator is a higher rated player than GM Rohde. >>> >>>True, and both have shown the know how to play against programs, but GM Rohde >>>has been more impressive in this regard. IMO >>> >>> H7 wil I predict bite >>>>the dust. >>> >>>Define bite the dust, 0-6 ???. I predict a 2.5 to 3.5 result in GM Yermo's >>>favor, but this would still be a GM result for Hiarcs7 and far from bitting the >>>dust. >>> >>>I think Bob is right, micros have a long way to go yet against top >>>>class humans. >>> >>>I think this has already been shown to be incorrect, micro have already shown >>>they can hold their own playing the best players even at long time controls. >>>Better no, but little doubt that they are in a GM class. IMO >>>> >>>>rajen gupta >> >> >>A couple of points. >> >>1. The Yermo/Hiarcs match will _not_ suggest Hiarcs is a GM, _unless_ Hiarcs >>wins. Once Yermo gets to 3.5 (ie suppose he wins the first 3 and then draws >>number 4) then he can resign the rest, save his 'brainpower' and still walk >>out a winner. This is the reason a _match_ can not be used to reach a GM >>Norm. And is the reason that the USCF never allowed "matches" to be rated when >>a computer was involved. > >I agree, if Yermo wins 3.5 points and walks though the rest of the games, but I >think we will be able to see this in the games and would take this into >consideration. I hope Yermo's pride prevents this. > >We are not that far off in are thinking, you think that programs are just under >the 2500 GM rating (2450 or 2400). I think the are a bit over 2500, but under >2550. I think this because I take IM results into account, player with ratings >2400 or better but under 2500. Here they have shown they are better even at long >time controls over IM's. I don't know it you consider this a long way to go? > Here is what worries me about saying 'computers are GM players'. (1) I am a programmer (chess programmer) and I know what _my_ program lacks in positional understanding would fill a book. And I know that anyone that has a program reasonably close to mine in speed is going to have an equal number of huge holes. And while I don't know a lot about hiarcs, I have seen it make the very same mistakes with respect to king safety. So knowing what programs lack, leaves me feeling uncertain about the GM idea. (2) programs are tactically wonderful. And they will find tactical busts against GMs from time to time. And this will (may) be enough to win (say) one game of every 4. But that leaves them 200 points below that player, roughly. The point is, can the tactical skill of a program overcome the positional ignorance it has, and allow it to compete at GM level? 5 years ago I thought no. Excepting for Blitz, as I had seen Cray Blitz shred GM after GM like cheese (at 5 0 using a real board, a real clock, and my manually operating the program, no fudging by making the GM play on the programs GUI board). And last year several of us tore the fool out of several GMs in a game/30 match... which was _totally_ surprising to me. But it happened, Bruce/I/Ed/others have reproduced this, so it is apparently the case that tactics plus a little positional understanding is enough to win at game/30. But now we get to 40/2hrs, and that looks different, because suddenly the GM players are actually tactically stronger in many ways. Yes, they will make mistakes. Yes they will find tactics that a program has no prayer of finding. And they do _not_ have all the positional holes that a program has. So that is what leaves me cold... we are doing well, but we are doing it with tactics. And I find it hard to believe that is enough. Else chess is not an intellectual game at all, just something for a fast calculator. >> >>2. I've still seen no serious evidence that suggests that PCs are GM-level >>at 40/2. At 5 0 there is no question. At 30 0 there is no question. But >>at longer time controls, nothing I have seen would suggest this. At least >>two GMs today made this comment... > >Again I do look at the IM games were computers have shown they can be much >better, but I can understand why you do not. I watch them also... but I use a bigger body of information, in that I have watched _hundreds_ of IM games... I have watched two in particular (Hartman and Borriss) totally decimate computers of all kinds. And I have watched non-IM players (Tim on ICC is one) that can decimate a computer yet has no chance against a GM or even IM. And we don't have a lot of 40/2hr games to look at, but even at faster games, the IMs can tell you the weaknesses of the programs. And they are better at anti-computer than GM players are, because the IMs want to win at all costs. IE try on Danny Kopec for a real thrill at anti-computer play... > > that at 40/2 the tactical wizardry of the >>computer becomes generally ineffective and then the huge evaluation holes are >>simply overwhelming. Both GMs recognize that at 40/2 a computer is going to >>win a game here and there on tactics. But it is going to lose more based on >>understanding (or misunderstanding.) > >This is why I am looking forward to the Yermo match, I think this match will be >telling.... If Hiarcs7 gets crushed I will be in full agreement with you. >Barring something strange by the operator of Hiarcs7 as shown in its games. The >operator of Hiarcs7 is a unknown to me and I hope he does not waste this match >by hurting Hiarcs7 play in some unknown way to him. > me too... there are not enough of these games to be able to fritter away any opportunity... we need data. The Rohde game was only one data point and isn't enough to draw any conclusion, although I know more after seeing him win than I would have had he lost. Because now we know computers are not head and shoulders above 2550 players. The only question is how close are they? time will answer this... >> >>In chatting briefly with Michael after the match today, he made it clear that >>he thought the game was over almost by move 20 or so. > >For this I will have to take his word if white is lost at move 20... I hope >another GM player will look at this game. Maybe the GM's at smartchess. me too. I can tell you that after black's 22nd move, where white plays ed at move 23, crafty fails low on this at ply=14 and continues to fail low for white and high for black every move beyond this... > > Even if he didn't see >>the tactical shots that the computers saw for him... One quote by a GM today: >>"What human IM or GM would _ever_ cede the two open files as Rebel did? The >>game was lost at that point, how Rohde was going to win was the only question, >>as white had nothing left to say about the course of the game after allowing >>that to happen." > >This may be correct but I have seen no winning line for black yet after 26. Bg2. > >> >>That is a pretty big positional misunderstanding. And remember that Rohde >>found some nice tactics with almost no time left. > >This I know about, he crushed the programs I played against him at speed chess, >on a PII 300!!! > > IE he played the last 12 >>moves or so in under 5 minutes total... to reach move 40 and the time control. > >To me GM Rohdes is one of the best or best at playing computer programs. He >played much better against Rebel then did GM Anand. IMO > >This is why I predicted a win for GM Rohde over Rebel.
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.