Author: Robert Hyatt
Date: 09:46:36 01/23/00
Go up one level in this thread
On January 23, 2000 at 10:54:57, Chris Carson wrote: >On January 23, 2000 at 10:18:01, Robert Hyatt wrote: > >Big snip >> >>Note that there are _no_ deep blue vs micro games in existance. Even the >>40 games I mentioned were against a 1 processor version running in a >>crippled mode. DB has _never_ played in a computer vs computer event, only >>deep thought. > >Fair enough and I expect you to never say that DB has proven itself >better than the micros. You can express an opinion, but it is only >an opinion and not fact and you should label it as only an opinion. > >Best Regards, >Chris Carson It isn't quite an opinion. DT definitely proved itself superior to micros. Over a 10 year period. DB played several games vs Deep Thought as Hsu and Campbell were testing it. It was far superior. In this case, since DT > micros, and DB > DT, I think the transitivity holds across all three, because DB wasn't DT minus something. It was DT plus a lot. Here is another angle to consider: Deep Thought's processor was a 1987 piece of hardware. By 1995 it had proven itself to perform at a GM level, with three numbers you can choose from. An official USCF rating of 2551 over almost 50 games. A TPR over its best 25 consecutive games of 2650. And finally if you only count deep thought games without the first few which had horrible hardware problems, it had a TPR of 2590. All are impressive. _all_ are 1986 hardware (the single processor version, called "chiptest" first played in the 1986 ACM tournament.) So 1986 hardware produced a true GM-level machine, based on enough games that it can't be considered a fluke or accident. Today, Rebel is trying to produce the same result, on hardware that is 14 years newer. It is _very_ close to 2500. It isn't as close to 2650 (by taking the best 25 games vs humans). But it might make 2500 by the Summer with a little luck. With the above data, how far _ahead_ of Rebel (in terms of time) would you say DB-2 is? Rebel has yet to do what 1986 hardware has done. No other program has yet produced a GM-class rating over 25 games. My guess is, as it always has been, at _least_ 10 years behind. We can discuss any of the above that you like, of course... We have DT's results documented quite well. We have (thanks to Enrique) Rebel's results so far. A few more months will push this to 25 solid games... 1986 vs 1999 hardware. fairly close comparison in performance. Of course, we have the problem of the 1997 version of the Hardware to deal with. :)
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.