Author: Robert Hyatt
Date: 14:45:20 01/25/00
Go up one level in this thread
On January 25, 2000 at 16:51:09, Mike Castañuela wrote: >On January 25, 2000 at 15:26:45, Robert Hyatt wrote: > >>On January 25, 2000 at 12:17:02, Mike Castañuela wrote: >> >>> >>><snip> >>>Not my desire to stir the things, but it's a >>>funny post with plenty of fine irony. >>>I'm stay without understand as Dr. Hyatt is partial >>>(yet very intelligent persons have prejudices) towards >>>anything related to DB; is my impression that him will defend >>>with too much more decision DB that own Crafty. >> >> >>Did you ever consider that this is maybe because I know a lot about _both_ >>of them. And I know what my program can/can't do with respect to theirs. >> >>They did something _nobody_ has done. Either before them or after them. >>Something I doubt will be done for at least another 10 years unless someone >>bites the bullet and starts a hardware design project similar to theirs. >> >>The post you responded to wasn't particularly funny, IMHO. Just extreme >>hyperbole of out-of-context discussions. Which is typical... > > >First, my apologies if my previous answer appears harsh to you, >I respect your work and trajectory. > >Sure you're know a lot about DB, but the apparent (and well distinguished) >and only merit of DB is relative at their hardware approach, comparing to >micros. > >Howewer, the data provided by Chris Carson about DT performance vs. >humans is only average, not impressive in thruth. Think about the time frame. This hardware he has data for was built in 1986. In 1986 we didn't have a single program on a micro that anyone in their wildest dreams would say is a "master level program". They did this performance against humans with hardware doing about 2M nodes per second. We have programs today that can search this fast, but we aren't seeing those kinds of performances from them yet... Everyone wants to compare DT (1986 design) to todays (2000) hardware/software programs. Assuming that todays programs _could_ produce a 2650 performance rating over 25 consecutive games, what does that say? That we are _only_ 15 years behind the DB project, overall? That was the point... Sure I think I could beat deep thought today, given a 16 cpu alpha machine. I can greatly out-search them. But I couldn't in 1986. Or in 1990. Or in 1995. Nobody could. And that only gets me up to deep thought. Deep blue is 100X faster, and 10x more sophisticated to boot. Will take a while to close that gap, IMHO. > >I think that the debate between DB and competitors, must be centered, >not between results and performance (by stupid that appears this), >but between which evaluation function is better, by obvious reasons. How can we do this? We can say that todays micros can search at the same speed as the 1986 DT design. We can _hope_ that todays micros can play at the 2500+ level (maybe, maybe not, the judge is still out on that one). But does that say which has the better eval? Or the better search? Or the better computer? Or just "the better system, overall"??? I don't think DB was 'positionally stupid' by _any_ measure. Take _any_ comment by Kasparov or other GM, other than kasparov's cheating allegations. "I knew that this was something 'new and different'." His words. talking about DB being well beyond anything he had seen to date from micros. It is funny that he said this _several_ times, yet it is lost behind the cheating claim. I think Kasparov "felt" that this was some machine, and not from guessing, but from having to sit across the board from it and seeing it defend perfectly when it had to, attack vigorously when it could, and in short, play against him like no other player has come close to, to date... > >I don't think that those of best micro programs be worse. Perhaps not. But I also don't think they are _better_ either.
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.