Author: Uri Blass
Date: 23:06:37 05/03/01
Go up one level in this thread
On May 03, 2001 at 16:48:20, Christopher R. Dorr wrote: >On May 03, 2001 at 00:59:14, Robert Hyatt wrote: > >>On May 02, 2001 at 22:43:55, John Dahlem wrote: >> >>>On May 02, 2001 at 22:05:43, Robert Hyatt wrote: >>> >>>>On May 02, 2001 at 18:39:24, Fernando Villegas wrote: >>>> >>>>>Hi Bob: >>>>>Probably I missed this, but: it is clearly in the contract that the Fritz K. >>>>>will play must be the very same as the one that played DJ? There will be an >>>>>engine frozen in a safety box in a bank or in the Pentagon to make sure that >>>>>condition? It is not possible that Morsch can add some things and substract >>>>>other, forbidden or not? By example, to add a match algorythm to accept or not >>>>>draw offers according the history of the match? There will be, if such rule of >>>>>"no changes in the engine" exist, a way to to examine the source code? And if >>>>>not rule of that sort, how we can know for certain the degree of improvements >>>>>Morsch can invent in three long months? >>>>>Regards >>>>>Fernando >>>> >>>> >>>>I don't know. IE prior to the finish of the 'qualifier' I don't see how rules >>>>could have been agreed to since no one knew which program would win. But if the >>>>rules say the same version must play Kramnik that played DJ, then I would >>>>personally say "shove it" and do something else if I were the programmer(s). >>> >>> >>>As I understand it the rules currently state that the program has to be done by >>>July and then shipped to Kramnik, and no changes be made after that. >> >> >>Wouldn't matter to me. I would refuse _any_ such requirement. If he wanted >>to download available versions of my program, fine. But I would not let him >>see what he would be playing... any more than he would show me _his_ >>preparations for the match either... >> >>This makes little sense... > >To me, this is much ado about nothing. Even if Kramnik had to settle for Deep >Fritz 6.1 instead of DF7, how much will he miss? As Bob has said, it's the >positional stuff that the GM cares about; the evaluation functions at the leaves >of the tree. No GM in their right mind is going to go head-on into tactics >against the box anyway. > >How much will the positional evaluations change between vesion 6.1 and version >7? Not much generally. I do not know. I know that the change between Junior5 and Junior6 or the change between tiger12 and GambitTiger was a significant change. If the positional error is big enough to be easily >caught, it would be quickly changed. But this isn't the stuff he is looking for. >He is looking for the subtle stuff that only a world-class GM can find, and only >by deep and serious study. This isn't stuff the programmer will likely think of >to work on. The programmer may work on other stuff that kramnik did not think about in order to have better evaluation function than Kramnik in part of the cases. One of the problems in chess is to give the right value for things. I know that Junior6's value of pawns is smaller than Junior5's value of pawns because Amir found it productive for Junior. I believe that programmers can find improvements by a lot of testing and they can get evaluation that in some cases is even better than Kramnik's evaluation. There are many things that a program evaluates when it evaluates a position and it is the reason that Fritz's evaluation may be better than Kramnik's evaluation in part of the cases. I know that when I calculated a line in a game I did the mistake of not paying attention for the fact that my opponent has a passed pawn in the end of the line. Something like this cannot happen to chess program. Kramnik is better than me but there are so many things to consider when you evaluate a position that it is possible that he may miss something that computer is not going to miss(maybe not a passed pawn but another positional factor). Uri
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.