Author: Robert Hyatt
Date: 07:29:04 01/06/01
Go up one level in this thread
On January 06, 2001 at 00:53:10, Uri Blass wrote: >On January 05, 2001 at 23:51:21, Robert Hyatt wrote: > >>On January 05, 2001 at 14:28:21, Uri Blass wrote: >> >>>On January 05, 2001 at 14:03:57, Robert Hyatt wrote: >>> >>>>On January 05, 2001 at 07:50:42, Mark Schreiber wrote: >>>> >>>>>In the match with v/d Wiel, Rebel is running on P3 866 MHz. Using a faster >>>>>computer would be an improvemnt. Maybe a P4 1.5 GHz. They could also improve >>>>>Rebel to run on dual or multi processor like Junior. The Junior that ran on an 8 >>>>>processor at Dortmund would clobber v/d Wiel. At Dortmund, Junior performed at >>>>>Fide 2700. >>>> >>>> >>>>I doubt _any_ program will "clobber" him. Speed isn't the only issue when you >>>>play a computer-savvy GM. If your program has a hole (and all current programs >>>>have many of them) then speed isn't going to help a bit if the GM knows what he >>>>is doing. >>> >>>I believe that speed is going to help because the holes of chess programs can be >>>covered by deeper search in part of the cases. >>> >>>There are positions when speed will practically not help but getting this >>>positions may be prevented if the computer is faster. >>> >>>Uri >> >> >>We've been waiting for this to happen for 30 years. We aren't there yet. I >>don't think we will be there in another 30 years. The holes _must_ be filled >>or the programs are going to have problems with anti-computer humans _forever_ >>no matter how fast they go. DB1 should have proven that. it was 200X faster >>than the fastest program of today. And it fell into the same problems in the >>first Kasparov match. > >It did not prove it. > >Kasparov is a better player than Van der Wiel and it is possible that DB1 could >win against Van der Wiel. I don't believe so. Kasparov used similar ideas to defeat DB1. DB 1 didn't really understand cramped/blocked positions at all. And if it searched 200B nodes per second, it would be a tactical monster but would be defanged by closed positions. It wouldn't have any more luck if it was 1000 times faster if the evaluation had that big a hole. > >I also believe that the programs of today have better positional knowledge than >DB1 and better pruning rules that help them to search deeper so the 200x faster >may be misleading. > >Here is a position(from game 5 of the match) when I believe that DB1 made a >tactical mistake(I did not try to prove it by a tree but it is my impression). > >[D]3r2k1/p4bp1/5q1p/8/3Npp2/1PQ5/P2R1PPP/6K1 w - - 0 1 > >White played g3 when I believe that the only move is Ne2 >The tactics is quite(white has a lot of possibilities in every move) and this is >the reason that the singular extensions could not help DB1. > >I think that it may be interesting to know how much time do programs need to >find Ne2 and what is the depth that programs of today need to see significant >difference between g3 and Ne2. > This is interesting. Until depth=11 crafty likes h3. But at depth=11, it switches to g3. And it _knows_ that king safety is important. Something about g3 seems to be almost forced here for some reason. It would seem to me that Ne2 invites f3 and trouble. It bounced back to h3 at depth=12, but the scores are _very_ close (g3 vs h3). >I guess that a lot of program may find Ne2 in some minutes(they will not see >tactical difference between Ne2 and g3 but they will see a positional >difference) > >I guess that part of them may see a difference of at least 1/2 pawn if you give >them 3*200=600 minutes on fast hardware(I mean a difference of at least 1/2 pawn >between their score and their score if they investigate only g3 at the same >depth). > >A difference of 1/2 pawn means that they will not see that g3 is a losing move >but they will see big positional reasons not to play it. > >It will be interesting to do the experiment. > >Uri
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.