Author: Vincent Diepeveen
Date: 08:49:14 01/12/00
Go up one level in this thread
On January 12, 2000 at 09:43:24, Robert Hyatt wrote: >On January 12, 2000 at 09:04:42, Vincent Diepeveen wrote: > >>On January 11, 2000 at 18:14:46, Dann Corbit wrote: >> >>>On January 11, 2000 at 16:11:54, Amir Ban wrote: >>>[snip] >>>>Don't be silly. That's the end position of DB-GK game 2. Deep Blue didn't see >>>>the draw, and actually made the mistake of walking into this position. People >>>>found the draw half an hour after the game and used micros to verify it. >>>Not at all unlikely that I am wrong about all of them. >>> >>>At any rate, there was a thread titled "DB vs Kasparov Game 2 35. axb5" which >>>had computers running for days on end. I see that you were a contributor. The >>>other thread authors were: >> >>>Blass Uri >>>Bruce Moreland >>>Dan Homan >>>Dave Gomboc >>>Ernst A. Heinz >>>Ernst Walet >>>Eugene Nalimov >>>Graham Laight >>>James Robertson >>>Jeremiah Pennery >>>Mark Young >>>Michael Cummings >>>Peter McKenzie >>>Robert Hyatt >>>Scott Shepherd >>>Vincent Diepeveen >>>Will Singleton >>> >>>I have put all relevant messages into a zip file on my ftp directory: >>>ftp://38.168.214.175/pub/AXB5.ZIP >>> >>>for any who might be interested. You seem to have had some special insight on >>>this position, and even had access to the printouts. >>> >>>Is this position reproducable by PC's? >> >>Yes it is, for the same reason that deep blue played the Be4? (Qb6 wins >>as shown in numerous analysis) move. > > >I don't believe this has been proven. DB failed low on Qb6. It is therefore >likely it isn't "good enough". > > >> >>Deep Blue finds opposite bishops a big draw, even when being a passed >>pawn up. Combined with the fact that deep blue doesn't like a king >>without pawns around it (and a possible queencheck) this makes programs >>find axb5 without problems. > > >This is _absolutely_ false. Hsu and several of us had a long discussion at >one ACM event about this, as Cray Blitz was in a position where it thought it >was drawish for the same reason. Hsu mentioned that deep thought had come up >on a similar position and had a similar eval, and that a GM had talked with >him at length. And that as a result, he had modified the eval, as opposite >bishops don't always draw. > >Why make statements that you absolutely can't prove, that you have absolutely >no foundation for? Deep Blue had a good sense about which of these type >positions are winnable and which are not... There are simply bugs in deep blue singular extensions that made it play your so beloved c6-c5 move with a big score. Just give the pseudo code of your singular implementation in cray blitz. that makes things a lot easier for the audience. I can prove my statements just as easy as you can disproof them. OBVIOUSLY deep blue shows for the Qb6 line before axb5 a line that gets the opponent queen to the white king with king in center. When i changed that in my evaluation diep still didn't want to play the axb5 line, because there is another line possible after which white ends up with opposite bishops but a passed pawn up at b-file. We're not talking about a 'small penalty' here. we talk about a HUGE penalty here. Add the 2 together and a program plays the move. Note that the ab5 move was just played because deep blue score went down from 74 to 48, so it didn't see any drawing scores there. it was positional determined. then axb5 was 63. So that's half a pawn up for white on average. The Qb6 move next move was first 55 points worth. that went down to 32 then ONLY 5 points more was the positional determined move Be4, which is a bad move. Qb6 there wins clearly, as Seirawan and some other grandmasters clearly indicate, and later Kasparov also referred to as: "missing a win with Qb6, Be4 is a human move". So my point is still strong. Note that old DIEP version also play Be4, as in past i gave more penalty for opposite bishops. However when search depths get huge (much bigger than the 11 ply that deep blue searched here) then score of Qb6 gets slowly up when it starts to realize that it also losed the b5 pawn with black. >> >>The reason deep blue played axb5! is the same reason why it doesn't play Qb6! >>the next move. >> >>It plays Be4 because after Qb6 we get again an opposite bishops ending, >>but this time it's won, that's the big difference with the axb5 move.
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.