Author: Robert Hyatt
Date: 13:17:17 01/11/00
Go up one level in this thread
On January 11, 2000 at 15:08:07, blass uri wrote: >On January 11, 2000 at 12:59:33, Graham Laight wrote: > >>On January 10, 2000 at 21:34:40, Robert Hyatt wrote: >> >>>>I see what you are saying... >>>>> >>>>I did find this on the rebel site though... >>>> >>>>"Deep Blue - Kasparov the re-match of 1997 won by Deep Blue. >>>> >>>>In this famous position Gary Kasparov resigned while he could have a draw with >>>>Qe3! Like Gary for most computers this move is also very hard to find. >>>>Rebel 10.0 (normal) Qe3! not found after 13 plies and 30 minutes. >>>>Rebel 10.0 (anti-GM) Qe3 found at ply 12 at 7:22" >>>> >>>>looks like the anti-GM feature did find the move if I'm reading this right... >>>>> >>> >>>Aha.. now I understand. Crafty (at the time) found Qe3 easily. But it thought >>>it was losing. It never understood that the score was anywhere near 0.00. That >>>was the issue I was addressing. It is one thing to find the right move, another >>>to find the right move for the right reason. Kasparov said that the draw was >>>obvious after he looked at it and couldn't believe the computer had overlooked >>>it. He was wrong. The draw is over 60 plies into the future, not "a few". >> >>So how is it possible that within hours of the game, people from all over the >>world had found the draw with their PC based chess programs? >> >>-g > >I found the draw by giving Genius3 to play against itself after the interesting >Qe3 and trying to improve the line for deeper blue with no success. > >I saw an evaluation of draw after many plies in the relevant lines and I found >no improvement for deeper blue by trying next best. > >I think that the draw is obvious for a team of human+microcomputer+some hours. > >I was disappointed from deeper blue because I expected deeper blue before the >match to find everything that human(with elo less than 2000)+micro+some hours >can find about tactics. > >Uri I once had a position that I don't recall exactly, where Cray Blitz won a piece against a human. And overlooked a terrible repetition. But what it saw, with its singular extensions going like gangbusters was a series of checks, where its own king was hemmed in on the kingside by black's rook at e8. And when I saw it I said, aha, it will see this as the king gets checked up the board and when it tries to escape the repetition will show up. I failed to notice that it could play kg1-h2-g3-h4-g5-h6-g7-h8-g8-h7-g6-h5-g4-h3-g2-f1-f2-etc..... the checking sequence was _so_ long, yet I saw it almost instantly because I could see that the king had no way to get off the f/g/h files, _ever_. CB couldn't see deeply enough to search this, and it didn't evaluate such things, so it stumbled into a draw where it should have won easily... Those kinds of positions happen at times... IE I look to see "Can I _ever_ get out of this?" The computer takes a completely different approach to this by saying "Can he force a repetition within my search horizon?" The two questions are different. The answers are often different too, but not always. And when they are the same, the computer looks like a genius. When they are not, it looks like an idiot. :)
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.