Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Sometimes computers still aren't better than 2400's

Author: Joshua Lee

Date: 22:37:20 01/16/02

Go up one level in this thread


On January 16, 2002 at 23:19:04, Uri Blass wrote:

>On January 16, 2002 at 16:52:44, Joshua Lee wrote:
>
>>The ivanov game always impressed me, white didn't do anything special to win yet
>>it is way beyond the speed of current computers.
>>Perfect example why Commercial Programs may not play better than Deep Thought,
>>atleast not yet.
>
>You cannot decide by one game that programs of today are not better than Deep
>thought.
>
I did not decide this by one game this is just one example of a game against
Deep Thought that computers can't understand mistakes in without a very very
long think. You could say that at this speed 500K-700Knps which Fritz gets
around 500Knps on a 1Ghz is not any better than Deep Thought's nodes.

Alot of this is actually understanding that it might normally be a great idea to
castle and mathmatically this looks good but reality is that when all the pieces
come off the castled king will be a few extra steps away which equals a lost
game. I don't doubt that any commercial or non commercial can't realize this
with a deep search probably 30+ ply but then again that just shows that they
aren't better in this instance. Tactically might be a different story
Positionally also seeing how programmers increase the level of their program
20-30 elo per year, (i think this is just an estimate) in 6 years since Wchess
Drew Deep Blue Prototype this would amount to 120-180elo and an aditional 270elo
from the hardware fromPentium 90 to a dual 1400 for instance  this is in all
390-450elo   tack this on to to draw 1 game out of 100 at a minimum you have to
be within 396 points  and the rare instance of 1 game out of 1000 399.6 points.
This put's any program that is as strong as that Wchess (don't know if it was an
experimental or commercial version) on that hardware P90 is atleast within 60
elo of Deep Blue Prototype if that program was on a dual 1400. Essentially Deep
Thought on a 12+ Processor system 4Million+ nodes , Deep blue is another topic
all together.

Again a Dual 1400 Would technically put a program of Wchess's Strength on a P90
within 60elo if there has been a 20elo increase in strength of that program each
year for the past 6 years.  To be better would still take around 200elo over
What DB prototype strength was.




>Deep thought did a lot of tactical and positional mistakes in other games that
>the commercial of today have no problem to avoid.
>
>If you look at the games of Deep thought-kasparov you will have no problem to
>see that programs of today are better.
>
>If you look at comp-comp games of Deep thought you are going to find also
>mistakes of Deep thought that programs of today have no problem to avoid(even in
>part of the games that deep thought won but got inferior position).
>
>You can use one of the top programs(not tiger) to analyze comp-comp games of
>Deep thought and also to analyze ssdf games of tiger14 against programs that are
>at least 200 elo weaker
>(Deep thought played against programs that were supposed to be at least 200 elo
>weaker in comp-comp games).
>The target should be to find tactical blunders.
>
>I expect people to find that Deep thought did more tactical blunders.
>
>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.