Author: Robert Hyatt
Date: 14:04:33 09/09/01
Go up one level in this thread
On September 09, 2001 at 03:30:59, Uri Blass wrote: >On September 09, 2001 at 02:14:30, Dave Gomboc wrote: > >>On September 09, 2001 at 02:13:07, Dave Gomboc wrote: >> >>>On September 09, 2001 at 01:51:24, Uri Blass wrote: >>> >>>>On September 08, 2001 at 23:37:18, Robert Hyatt wrote: >>>> >>>>>On September 08, 2001 at 15:25:47, Uri Blass wrote: >>>>> >>>>>>On September 08, 2001 at 12:18:45, Gian-Carlo Pascutto wrote: >>>>>> >>>>>>>On September 08, 2001 at 11:46:09, K. Burcham wrote: >>>>>>> >>>>>>>> >>>>>>>> so i have concluded after lots of time analyzing deep blue positions >>>>>>>> that todays programs seem to be very close or equivelant to deep >>>>>>>> blue in playing strength. >>>>>>> >>>>>>>The problem with DB and the main reason why this debate has been >>>>>>>going on since the start of CCC is that theres just not enough >>>>>>>data. 6 games doesn't seem to be enough to get a decent idea to >>>>>>>compare DB to others. So people start making all kinds of assumptions, >>>>>>>and arrive at even more shaky conclusions. >>>>>>> >>>>>>>I personally do not believe that the top comps of today are >>>>>>>equivalent to DB as far as search is concerned. This is based >>>>>>>on the data I have seen and what I know of DB's design and search. >>>>>>> >>>>>>>As for eval, well, I think that is another matter. While DB no >>>>>>>doubt had a very sophisticated eval, and contained more than >>>>>>>nowadays micros can do, I'm not sure if it was tuned as well as >>>>>>>todays comps are. >>>>>>> >>>>>>>They may have had a team of grandmasters and good programmers, >>>>>>>I think tuning a top program is something that must be done >>>>>>>over time and based on loads and loads of games. It is wellknown >>>>>>>that DB wasn't actually 'final' when it played Kasparov. So >>>>>>>their tuning wasn't probably all that great either. The 'smart' >>>>>>>parts of the eval may have interacted in a less than ideal way. >>>>>>> >>>>>>>Whether or not that added up to something that was weaker or >>>>>>>stronger than current top is something I don't know. Nobody >>>>>>>else here knows either. And you won't be able to tell from >>>>>>>6 games, no matter how long you argue (its 5 years and counting...). >>>>>>> >>>>>>>Fact is, DB did what it was supposed to do. It beat Kasparov >>>>>>>and generated a huge amount of publicity. >>>>>>> >>>>>>>Robert may not like the fact that many people (I won't call >>>>>>>names, you know who you are) like to compare their programs >>>>>>>to DB or even say they're better to build onto the huge >>>>>>>amount of publicity DB generated. But somehow this is >>>>>>>justified. Not because their programs are stronger, but >>>>>>>because DB disappeared after it gave the impression comps >>>>>>>topped humans. But a champion is not champion if he does not >>>>>>>play. >>>>>>> >>>>>>>Deep Blue is the Fischer of computer chess. >>>>>>> >>>>>>>He did something cool, disappeared and left the rest of >>>>>>>the world arguing instead of moving on. >>>>>>> >>>>>>>The Fritz match will be interesting. If Fritz beats Kramnik, >>>>>>>that'll be a very good argument against DB. But I expect >>>>>>>Kramnik to toast the comp actually. >>>>>>> >>>>>>>What bothers me about that match is that they made it look >>>>>>>like Kramniks demands were redicolously unfair, so the meaning >>>>>>>of the match in the comp/human/Kasparov/DB debate is reduced, >>>>>>>but it seems that they aren't going to abide by the terms >>>>>>>anyway. This is probably good...It'll do Kramnik more justice >>>>>>>when he toasts it even then. >>>>>>> >>>>>>>Oh, and if Hsu publishes his book, that will also be >>>>>>>very intersting of course...but when, if ever? >>>>>>> >>>>>>>> in other words i am looking for any positions >>>>>>>> that my system will not choose deep blues next move. or does >>>>>>>> not see deep blues next move as an equivelant eval. >>>>>>> >>>>>>>[D]r4bk1/5rpp/1Bppbp2/4n3/N7/1PP5/P1B2RPP/4R1K1 b - - 7 27 >>>>>>> >>>>>>>From DB's ancestor. You need to >>>>>>> >>>>>>>a) find the best move (easy)b) find that it wins a knight (eval >2.xx) within 3 >>>>>>>minutes >>>>>>> >>>>>>>The 3 minutes should actually be divided with the speed difference >>>>>>>between DB and Deep Thought. >>>>>>> >>>>>>>-- >>>>>>>GCP >>>>>> >>>>>>This position was discussed a long time ago in CCC >>>>>>The conclusion of me and Amir Ban and a lot of other people was that black does >>>>>>not win a piece because no human could prove that it wins a piece. >>>>>> >>>>>>If you want to find an impressive move of Deep thought then you need >>>>>>to find something that humans can understand. >>>>>> >>>>>>If humans cannot understand that it is winning a piece after going forward and >>>>>>backward with their program then the argument is not convincing. >>>>>> >>>>>>Uri >>>>> >>>>> >>>>>Until 20 years ago humans thought that in KQ vs KR the king and rook _must_ >>>>>stay together for best defense. After a computer demonstrated that this is >>>>>not correct, humans _finally_ figured out why. >>>>> >>>>>The number of things humans are not going to understand is going to go _up_ >>>>>and not _down_ over the next 20 years. If you think that just because a human >>>>>can't understand something, it can't be correct, then humans are going to get >>>>>wrecked by a _lot_ of "incorrect" play over the next 20+ years and beyond. >>>> >>>>The main problem is the fact that humans together with programs ,time and the >>>>game could not understand the evaluation of Deep thought. >>>> >>>>If Hsu has Deep blue Junior that is supposed to be better than Deep thought then >>>>I invite him to prove the +2 evaluation against one of the top programs when he >>>>gives the top program some hours per move. >>>> >>>>I am interested to know what he thinks today about that position. >>>>Does he think that Deep thought really outsearched Cray blitz by 20 plies or >>>>does he think that Deep thought had a bug in the evaluation that caused it to >>>>believe that it wins material(maybe it had a big positional score)? >>>> >>>>I believe that Cray blitz was better than Deep thought and I think that Deep >>>>thought was simply lucky to get a position when Cray blitz blundered because it >>>>was more easy to find the right moves for Deep thought and not to find the right >>>>moves for Cray blitz. >>>> >>>>Uri >>> >>>Hsu hasn't worked for IBM for quite some time. AFAIK he's at Compaq... oops, >>>HP. :-) >>> >>>Dave >> >>Actually, now that I think about it, didn't Compaq sell off their processor >>stuff to Intel? Maybe he's there now. >> >>Dave > >I know that hsu used Deep blue Junior based on posts that I read few monthes ago >and tested it in test positions. > >I did not see the results as convincing because of the following facts: >1)I am interested in practical games and not in test positions and the position >after the c5 together with analysis of Deep blue Junior is more interesting. > >2)I doubt if the Deep blue Junior that is used today by Hsu is identical to the >Deep blue Junior that was used at that time and it is possible that it is simply >better so even if Hsu has a version of Deep blue Junior that is better than the >commercial programs it does not prove that the old Deeper blue from 1997 is >better. > >Uri In other words, the is _no_ way this can ever be proved, correct? Which does tend to make arguments futile...
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.