Author: Robert Hyatt
Date: 19:24:24 01/10/98
Go up one level in this thread
On January 10, 1998 at 18:31:39, Don Dailey wrote: >Hi Bob, > >Yes, singular extensions are certainly complicated to implement. I'm >sure there are a lot of ways to do it wrong and hurt the program and >I believe it can be a win if you do it right. > >You said we were both absolutely destroyed by Deep Blues extensions. I >don't remember this at all. We have played them twice and neither >game went like this. The first game was considered the best game >ever by 2 computers in a competition (which I don't beleive) and >the 2nd game was very normal, no one getting a big advantage quickly. > the game I recall had *Socrates white, DTII (or deep blue prototype) as black. There was some sort of attack by DT on the queenside and at the time, there was quite a discussion as their PV and score was much higher than the one coming from *socrates. The discussion thought that the DT guys had a bug. A few moves later it was obvious they didn't. The next night, Mike Valvo made the comment that "last night was a remarkable game. Deep Thought gave *socrates a chess lesson that won't ever be forgotten." We were all impressed during the game, when their eval held solid, and *socrates' eval slowly slipped, move by move. I'm trying to remember if you were there, but I think you were. I know Bradley was as well as Larry. I want to say this was in Cape May, where, if you recall, they forfeited round 1 to Marty's program when they lost power at IBM Watson, but they blew thru the next 4 rounds like Karpov through a high- school chess simul... >But I have a feeling they still would have a chance against us without >the singular extensions. I think they'd beat any of us no matter what they do. They have too much search speed, and all the eval tricks they want for free. But while I didn't think much of singular extensions when I tried them, the idea seems to be more useful than I had thought, based on their results and the number of games where it finds a critical move. IE most would say that endgame databases don't affect many games, but I am finding a significant number of KRP vs KR endings. Actually there are many more pieces on the board, but Crafty finds clever ways to trade into winning or drawing positions when it isn't obvious. I am beginning to believe that singular extensions are also capable of finding something important in every other game or so... but until I try them again, I can't promise anything. > >- Don > > >P.S. I have a question for you: Do you believe S.E. helps more >against humans than computers or visa versa? I think this is >definitely a possibility. > I'm not sure. However, were I to guess, I'd guess the other way. Why? Back to my KRP vs KR database. It comes up *much* less often against humans than against computers. Against good computer programs, I generally see Crafty come out a pawn up or a pawn down or even, most of the time. And that tends to make KRP endings more likely (again, based on the dozens of games I play against computers on ICC). But against humans, things then to turn tactical or something else, and end up more unbalanced somehow. You could easily be right of course. Since I don't use 'em, I don't have nearly as much data as I do for the things I am doing... But clearly playing against computers and playing against humans are two different things for the most part...
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.