Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: first possible example of a Blunder by Kramnik

Author: Robert Hyatt

Date: 10:18:04 10/25/02

Go up one level in this thread


On October 25, 2002 at 01:46:15, martin fierz wrote:

>On October 25, 2002 at 00:40:04, Peter Kappler wrote:
>
>>
>>I think that the 1.5 pawn threshold you guys have chosen is going to generate a
>>lot of false positives:
>>
>> - sacrifices that Crafty (or any computer) can't understand
>> - Kramnik choosing to win a rook instead of mating in 9
>> - sacrifices that probably aren't completely sound, but
>>   are perfectly reasonable in human vs human play
>
>absolutely. the idea is of course that we are checking these games, not just
>trusting crafty's 1second analysis. of the 184 games, crafty let 117 go
>unannotated. i will look at all the remaining games with fritz, and judge for
>myself if the positives are false or not (hint: they are false, most of the
>time, for exactly the reasons you list).
>
>
>>are scrutinized by players all over the world.  If he had made a blunder of that
>>magnitude (missing a 1-ply tactic when not in time trouble) it would have
>>already been published.
>i totally agree. it's something i wrote here before, but some people here don't
>understand just how extraordinary this move was :-)
>
>aloha
>  martin
\


I am re-running the annotation at 30 seconds per move, to see what deeper
searches
show up.  I already notice some "different" blunders, if they are "really
blunders" of
course.  But the depth is significantly deeper, so that a lot of the false
blunders ought
to go away (although a smaller number of new ones should be added, since going
deeper
will see some things that going still deeper will refute, and vice versa.).

I will post this .can file when it finishes, but it has been running since last
night and is
still trucking along...



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.