Author: Robert Hyatt
Date: 14:27:04 01/18/00
Go up one level in this thread
On January 18, 2000 at 16:39:06, Amir Ban wrote: >On January 18, 2000 at 14:49:58, Robert Hyatt wrote: > >>On January 17, 2000 at 14:54:18, Amir Ban wrote: >> >>>On January 16, 2000 at 21:34:44, Robert Hyatt wrote: >>> >>>>On January 16, 2000 at 14:28:09, blass uri wrote: >>>> >>>>>The only way to convince me that deep thought i better than the commercial of >>>>>today is to do public games of deep thought against the commercial of today. >>>>> >>>>>I am not convinced by the performance of deep thought against humans because >>>>>I believe that humans know today better how to play against computers. >>>>> >>>>>I am also not sure if 2550 of 10 years ago is the same as 2550 of today when >>>>>many IM's and GM's learn from computers. >>>>> >>>>>Uri >>>> >>>>It was 2650, not 2550, over 24 consecutive games. I don't think it dates back >>>>to 10 years, but don't have my ICCA journals handy to see when they were awarded >>>>Fredkin II. I think about 5 years ago roughly... >>> >>>2551 USCF >>> >>>(http://aaai.org/Magazine/Issues/Vol10/10-02/Berliner.pdf) >>> >>>Amir >> Do you ever read the things you give URL's for? The very first paragraph of this short report said _exactly_ what I had mentioned below: that DT had _several_ bugs that really hurt its performance in the early going when establishing the USCF rating... If you look at the last paragraph, first page, you find another thing I quoted: "produce a performance rating of 2500 over 25 consecutive games". If you take the best 25 consecutive games, you get 2650. If you take _any_ 25 games, you get 2551, which was just good enough. >> >>That had nothing to do with Fredkin. Fredkin stage II required "at least 2550 >>over 24 consecutive games". Not 2550 USCF or FIDE. DT had a 2650 rating over >>it's "best 24 consecutive games vs GM players". >> > >The link is the reference. It's Berliner's report on the Fredkin stage II prize. >As I've already said here once, there's no way to make you see something you >don't want to. > >It's deja vu anyway. This conversation took place two months ago, only then you >were lobbying for a 2600 DT rating. > BTW I lobbied that the _data_ showed that DT had a performance rating of over 2600, based on games very similar in scope as the current Rebel challenge series. Except that it _definitely_ went over 2500 over all the games it played, and went over 2650 if you weed out the first half-dozen games where it got beat up badly due to brand new software/hardware glitches. Berliner made this point as well. > >>2551 was for deep thought from Day 1. Cray Blitz's USCF rating is only 2258, >>because it played in only two human events (the 1981 Mississippi closed >>championship tournament which it won with a perfect score). I doubt anyone >>would think it was 2258 in 1986 on the XMP, or in 1989 on the C90. But >>original ratings included _all_ games including the ones where a program did >>badly. Which included Cray Blitz losing two games due to a horrible parallel >>search bug in late 1984 in the only other event it played in. There were lots >>of versions of deep thought, some with horrible bugs. Some without. Using a >>rating that spans all of those is not very accurate, although it is remarkable >>to me that they sustained 2551 with some of the glitches they had... > >Another pompous irrelevancy. > >Amir I haven't been arguing for _any_ DB rating. DT absolutely had a 2650 rating over 24 games. This was published in the original fredkin announcement. Berliner's "2551" is meaningless as that wasn't the requirement. The 2650 comes _directly_ from Hsu. Whom I "assume" knows how to compute a TPR over 24 games. It is also in his new book. I can manually type in the opponents/ ratings of the opponents if you want. As far as "pompous irrelevancy" I suppose that is what I should expect. I tried to explain how a "USCF rating" and a "performance rating" could be way different, and that the Fredkin prize was awarded on the basis of "performance rating". Even though DT's real USCF rating was also over 2500, which is all the more remarkable.
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.