Author: Robert Hyatt
Date: 13:16:52 04/28/01
Go up one level in this thread
On April 28, 2001 at 10:14:40, Vincent Diepeveen wrote: >On April 27, 2001 at 23:57:06, Robert Hyatt wrote: > >>On April 27, 2001 at 16:53:28, Vincent Diepeveen wrote: >> >>> >>>I never got a 0.00 score. I get a near to 0.00 score, so a positional >>>draw and it is *not* a horizon effect. >> >> >>Then I would say your evaluation there is _wrong_. Just like those evals >>where you have a queen vs 2 rooks and you say you are 2-3 pawns _ahead_ >>and then get destroyed in endgames when the rooks control everything. > >Note that search lines of diep were around 60 to 70 plies in this >position because of all kind of stupid extensions. > >So the evaluation was very good. In fact Seirawan already concluded >very quick it was a draw after he saw the Re8 move and a few other >search lines. You don't extend enough lines 60-70 plies to conclude _anything_. To say so is simply funny. I don't recall _anybody_ concluding "very quick" that the thing was a draw. > >So human evaluation beats still the machine here by a large margin, >as all search lines investigated by Seirawan already are within range >of any chess programs search! Not necessarily. The "human evaluation" also includes "search" that is not a conscious thing. > >>> >>>DIEP's evaluation is biggest of the world. So obviously i have things inside >>>it which others do not have and probably never will. >> >> >>Biggest isn't always best. Did you ever consider a career in the World >>Wrestling Federation? That is the kind of comments they make all the time. >>It is the kind of comment I would _never_ consider making. > >Human evaluation is the best model we have so far, so getting a bit >closer to that is not bad. > >The knowledge in diep is of course only a very small >subset from what a human knows, assuming that human >plays masterclass chess. > >Best regards, >Vincent
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.