Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Not so fast

Author: Vincent Diepeveen

Date: 13:51:35 09/18/01

Go up one level in this thread


On September 17, 2001 at 19:45:50, Bruce Moreland wrote:

>On September 17, 2001 at 17:29:33, Vincent Diepeveen wrote:
>
>>On September 17, 2001 at 14:31:30, Robert Hyatt wrote:
>>
>>Bob, take it from an FM, after gxf4 the game is very complex.
>>yes it's lost, but it's *not* easy. And suppose you make a passer
>>mistake, well in that case you still draw it, has happened before...
>>
>>However in the case of f3 e3 then it's a simplistic win. With Bh5
>>you get the same endgame but even better at the board because you threaten
>>to create a mating attack after which it's completely game over in the
>>endgame with a piece up!
>
>It doesn't matter which move is best.  What matters is what the program thought
>and why.
>
>I thought that initially you wanted to see if DB was simply overlooking Bh5 in
>response to f3.  The way I saw things, you found this somewhat difficult move in
>response, and you wondered if DB saw it.
>
>When I saw that post, I wondered if it could be true that other moves lost as
>badly or worse, *according to the programs*.
>
>I believe that this is the case.  We've seen evidence that gxf4 also produces a
>very bad score.
>
>People have talked a lot about preprocessors and singular extension, but to my
>way of thinking, that has nothing to do with anything.
>
>If there are two moves that produce a bad score at deep search depths, it
>doesn't matter what scores they produce at shallower search depths.
>
>I think that your implication that DB missed some tactics here is without basis.
> That it understood that it was losing and chose between several bad moves, is
>much more likely than that it is a tactical midget that can't see anything, or
>that it chose this move based upon an incomplete or flawed evaluation.

5k2/7R/4P2p/5K2/p1r2P1p/8/8/8 b LCTFIN04 (...h3!)

Are you missing h3 to be a draw here with Ferret?

God knows how many nodes a second they burned for this position.
half a million, 2 million a second, 10 million a second?

Who knows? If so many nodes a second doesn't give h3 as best move,
even if it is for the wrong reason perhaps, then a program has
a problem!

This is Karpov-DeepThought btw. A few years after they published their
singular extension article.

It's an endgame and even not extending a thing finds this tactical
thing!

There are loads of examples. Most are visible in the logfiles from
the DBII match. Uri showed some things which they miss in their
mainlines which are idiotic to miss, either from positional *or*
tactical viewpoint!

But the discussion here is on the DBII side led by people who never
checkout any logfile on a chessboard, so i bet you'll not understand
my point either!

The DBII machine searched 12 ply usually. I don't doubt they saw
some capturing lines and matethreads way deeper than we would
expect from a 12 ply search, but what it missed is so huge, and
considering the progress that has been made the last years in
computerchess, that claiming that a thing from
6 years ago still would see things which some commercials hardly
see today that's simply insane!

>bruce



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.