Author: Uri Blass
Date: 08:44:48 04/06/02
Go up one level in this thread
On April 06, 2002 at 11:11:08, Sune Fischer wrote: >On April 06, 2002 at 10:31:06, Otello Gnaramori wrote: > >>On April 06, 2002 at 05:03:29, Sune Fischer wrote: >> >> >>>I am one of those ignorant people that have some doubts about the true strength >>>of the programs ;) >>>Let me explain why... >>>We all agree that in tactics nothing beats the computers, but how good are they >>>at the positional level? >> >>Not to be boring, but I would like to stress that tactics is way more important >>in chess than positional knowledge for winning purposes I mean... > >Maybe it is an irrelevant example, but my little engine is now seeing two plies >deeper than the previous versions. I have been playing matches against TSCP and >Ozwald043, it has a good positive score, so in engine-engine matches is now >clearly stronger. >However I have done nothing about the evaluation, and all you have to do, is to >wait for it to castle, sacrifice a few pawns so you have a clear shot at its >king, and you can just roll it over! >Point is, it is not playing any stronger against humans as far as I can tell, >humans just go for the king, and it doesn't begin to defend itself until it's >within its horizon, which is often too late. I disagree if it can see 2 ply deeper then it is stronger against humans. Not all the humans just go for the king and even if they do it they can expect better defence. > >The top programs probably have this snag fixed, but there might be other similar >weaknesses that one can find. >It takes some time and elo to find these, but I bet there are >openings/middelgames/endings that computers in general play weaker than others, >it is a matter of "research". There are also opening and middlegame that known humans players play weaker and it is a matter of research and I believe that if 2400 players are going to buy gulko and play against him in order to find his weaknesses then Gulko is going to get bad results against humans. I think that using the option to buy the program in order to learn it's weaknesses is unfair way to find it's rating and the only fair way is if the program is not commercial and humans can learn about it's weaknesses only by watching games of it against other players. I believe that if some sponsor decides to give every human who win against Gulko in a tournament 1500$ bonus for a win and 500$ for a draw then the rating of gulko is going to go down so even if 2500 humans can beat programs after serious preperation it does not mean that the real rating of programs is less than 2500. The only way to find the real rating of computers is to give humans the same money for beating them as they get for beating other humans and not doing the programs public. Changes in the program between the games should be also allowed. 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.