Author: Robert Hyatt
Date: 06:19:16 06/29/00
Go up one level in this thread
On June 29, 2000 at 08:54:07, Jarkko Pesonen wrote: >On June 28, 2000 at 15:51:33, Robert Hyatt wrote: > >>On June 28, 2000 at 12:35:30, Bruce Moreland wrote: >> >>>On June 28, 2000 at 11:48:46, Robert Hyatt wrote: >>> >>>>I understand... but _here_ is the real question: In how many similar positions >>>>will it have the same eval, and it be wrong? I've seen it play brilliantly in >>>>one game, and then play like a patzer in the next four games. It is nice to >>>>find positions where a program _really_ seems to get the right idea. But then >>>>reality sets in, as you find similar (but not similar enough) positions where >>>>the program comes to the same conclusion as in the brilliant game, but it is >>>>dead wrong. >>>> >>>>This happens way too frequently with computer chess programs, unfortunately, >>>>mine included. It will play a brilliant endgame against a GM, then come back >>>>and play a completely insane endgame that no 2000 player would even consider. >>>> >>>>I don't like to brand programs as "brilliant" until they handle things most >>>>of the time, not just some of the time... >>> >>>I think it probably understands it. I know that CST does things that are very >>>speculative in the middlegame, but this may be a null-move killer. Does Crafty >>>know to play 1. h5 in the following position? >>> >>>[D]2R5/4k3/p2Np2p/4P1p1/p5pP/q1P1P1P1/2P5/1K6 w - - >>> >>>Mine doesn't. Force the move though, and boom. >>> >>>bruce >> >> >>Mine neither. Too many zug positions there.. > >But Crafty did know how solve this up to version 17.4 > >17.5 didn't solve this anymore. >From main.c > >rewrite of outside passed pawn/pawn majority code. it is now >much faster by using pre-computed bitmaps to recognize the right >patterns. That isn't the problem here. This is broken by zugzwang. If an older one found the right move, it was just lucky. In fact, it probably found it due to improper evaluation of the blocked pawn position.
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.