Author: Mark Young
Date: 01:04:12 12/18/97
Go up one level in this thread
On December 15, 1997 at 15:00:19, Howard Exner wrote:
>Here are some examples from the most recent SSDF test games.
>
>8/R7/P6k/1K6/5p2/r7/8/8 b - - id"Genius 5 P90 - Rebel 9 P90";
>2R5/3K4/2P4k/8/7p/8/2r5/8 b - - id"Hiarcs 6 P90 - Rebel 9 P90";
>
>In the above two examples Rebel went on to lose both. These are easy
>draws
>for humans who are aware of the common drawing theme of advancing your
>own pawns then sacrificing the rook for the advanced passer. Can this
>kind of knowledge be programmed? I get the impression that these
>positions
>are difficult for all computer programs.
>
>8/8/Pb1p3p/3P2kP/4K3/4pB2/8/8 b - - id"Nimzo 3.5 P90 - Rebel 9 P90";
>
>Here Rebel 9 played Kh4 and went on to lose. Probably on faster hardware
>it would avoid such a move. But again I'm not sure if endgame knowledge
>of opposite colored Bishop endings is easily implemented.
>
>Games like these really hi-light the "idiot-savant" nature of
>computer chess programs. One minute they're brilliant and the
>next ...
What time control was those games played at. I put the Nimzo-Rebel9
position in my rebel9 and it plays Kg5-f6 at 17 sec on a p II 300 so on
a p90 it should find the same move at a 1 min. or so.
mark
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.