Author: Vine Smith
Date: 19:04:21 04/18/02
Go up one level in this thread
On April 18, 2002 at 21:37:44, Slater Wold wrote: >On April 18, 2002 at 20:46:49, Vine Smith wrote: > >>I have a theory about Junior 7 -- in some positions involving material loss for >>what it believes is adequate compensation, something prevents it from ever >>changing its mind further in the search. This may either be intentional, to get >>it to sacrifice without "wimping" out because the other side can resist for a >>while, or it may be a bug of some sort. Otherwise, how to explain its behavior >>(in analysis mode, otherwise the book would intervene) after 1.b3 d5 2.Bb2, when >>it insists, up to 21 deep (when I gave up), that 2...e5? is the best move, >>giving up a pawn for nothing. There's another position like this that I could >>dig up for you, if you're interested, where Uri Blass had Junior 7 analyze a >>position from a game I played against the Winboard engine Ufim, where Junior >>will not admit it has no compensation for a lost pawn no matter how deeply it >>searches. >> >>Regards, >>Vine > >DJ7 plays very "speculative" chess. No doubt about it. And it seems to do >rather well at it. I think most people know, it's my favorite program. > >This move isn't exactly "simple" to a human, but to most programs (commercial >and amateur) it is. > >It's like DJ7 has all this amazing talent, and knowledge. It can find things >most others will never evalaute. But sometimes, I think it lacks "common sense" >in a computer chess kind of way. > >Uh hum.......like pawn endings. I've seen it blow more of those than anything >else. The more I think about it, the more I'm convinced that its erratic behavior is the result of some brilliant, but not entirely accurate, programming "trick". That is, the same thing that allows it solve Nolot #1 is the same code that causes to blow off pawns like they didn't exist, even when inappropriate. As a practical matter, the only thing that is important is that this "trick", whatever it is, causes it to win more games than it loses. But as a trustworthy analyst, there is a problem even if you second-guess its opinion with another program. Suppose Nolot #1 was an unexplored position, maybe from one of your own games. Junior suggests this surprising sac, without showing a truly affirmative evaluation, and you wonder, "Hmm...has it gone haywire again, or is this is a little bit of Morphy?" Then you run Fritz or whatever past the position, and the other program doesn't see the sac at all. Now you have to step through various continuations move by move, and really you might as well have just analyzed it yourself at that point, because it's a time consuming process. Don't get me wrong, I like Junior very much, but I find I can be as frustrated with it as amazed. I'd like to understand how it thinks, so I can have a better idea of when it's on track or off. But I can't figure out how to reverse-engineer a look into its twisted mind by considering gaffes like 1.b3 d5 2.Bb2 e5. Regards, Vine
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.