Author: Kai Lübke
Date: 02:39:03 03/05/98
Go up one level in this thread
On March 04, 1998 at 22:17:53, Howard Exner wrote: >8/6p1/6k1/p7/2p5/r6P/P1RK2P1/8 w - - id"Fritz 5.0 - Rebel 9.0"; > >This is a game from Kai Luebke's 40/2 tournament run on two P200's. >Here Fritz played Rxc4 and the game continued with Rxa2+ Rc2, Rxc2. >Of course this entire sequence for white is a terrible blunder. Could >some of the programmers explain what is going on here. Are these >positions still too difficult for computers when they are so simple for >humans? Or is knowledge code required to solve these positions? Fritz >being a deep searcher misses this so I'm guessing some sort of >endgame knowledge is needed. Any explanations for Fritz's choice? To give some more data (I was shocked by this situation myself): Up to that point, both programs thought the game was drawn. I haven't checked if already Rxc4 is that bad, but Rc2?? is the big lemon for sure. Even after Rxc4 Rxa2+, Rebel thought of the game as quite even. Now Fritz took a 2 minute search (approx.) going to 14 or 15 ply IIRC, yet returned a draw score for Rc2. Rebel instantly said +4,30 for Rxc2 then. *Then*, surprisingly, Fritz needed only a few seconds to report a -4 score itself! Really strange why it didn't see it before, that one extra ply can't be the only reason. But I've seen similar (though not quite as drastic) things time and again, that one program made a deep search, returned an even score, predicted its opponent's move correctly and then suddenly dropped to -3 in seconds... This game was a major disappointment (also because it cost Fritz the qualification to the semifinals for which a draw would have been sufficient). Maybe Fritz still lacks endgame knowledge - it also performs below average on the endgame part of the Louguet suite. Worse than Crafty, that is. --- Shep
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.