Author: Robert Hyatt
Date: 12:35:37 05/02/01
Go up one level in this thread
On May 02, 2001 at 15:27:33, Uri Blass wrote: >On May 02, 2001 at 14:52:33, Robert Hyatt wrote: > >>On May 02, 2001 at 13:48:04, Chris Carson wrote: >> >>>I would like to hear from the CCC group how much K will >>>gain from having the program for 3 months. In my view, >>>an advantage yes, but maybe not as much as I thought at >>>first. >>> >>>1. K may not have the HW for the match. >> >>That is irrelevant. That only means that the program will be a bit >>stronger tactically. But it does _not_ affect the knowledge at all. >>If it doesn't understand that a pair of isolated passers are stronger >>than a pair of connected passers in a king and pawn ending, then no >>amount of hardware is going to teach the program that, and he will >>find out such shortcomings quite easily. > >No >It is clear that if the hardware is good enough then search is going to teach >the program to avoid the mistake. no it isn't, when we are talking about a hardware advantage of barely 4. IE 8x1ghz vs 1.5ghz for a single cpu. That will fix a _few_ things. But it won't do a _thing_ to the positional holes in the program's evaluation. > >The question is simply if the hardware in the match is going to help. > >In part of the cases it can help. >Programs without the knowledge that you give in pawn endgame may find the right >move in some positions by search when they need a long search. > >I believe that I can compose a test position when programs without the right >knowledge are going to need 8 processors to find the right move at tournament >time control and you cannot be sure that my test position is not going to appear >in kramnik's game. Yes, but I can compose 100 positions where the depth is _not_ the issue. Either you understand what to do or you don't. Because you make evaluational decisions at the _tips_ of the tree. And if you don't get 'em right, you aren't going to go _another_ 30-40 plies deeper to let the search show the evaluation what is going to happen. > ><snipped> >>>2. To get a real feel for Fritz 7 he will need the HW and >>> play 40/2 games. >> >> >>Not at all. Any GM I know can play blitz games and determine program >>weaknesses. > >This was exactly the mistake of adams against deep Junior in dortmund. > >He played the same opening against junior before the match and won at blitz but >unfortunately Junior played better at tournament time control and adams could >get only a draw. > >Uri He didn't do what I suggested. I don't suggest using blitz games to find openings that you can win with. I suggested using blitz games to find out what your opponent doesn't understand evaluation-wise. Just play the games and watch its scores and PVs. You will figure out just what it has no understanding of. I would never try to find an opening and hope the program would make the same mistakes in a real game that it made in the test game. I think that was a mistake Kasparov did in game 6 of match 2. But I would definitely look at the scores and PV during a lot of blitz games to see what the program doesn't understand or misunderstands. That is _not_ hard to do. If it wasn't for the fact that I'm not particularly in the business of helping commercial programmers, I could give you some specific comments made by two GMs that have been playing my program and then others. They pick up on my holes, and the commercial holes _very_ quickly.
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.