Author: Uri Blass
Date: 08:23:28 12/09/02
Go up one level in this thread
On December 09, 2002 at 11:14:42, Bob Durrett wrote: >On December 09, 2002 at 09:30:39, Uri Blass wrote: > >>On December 09, 2002 at 09:22:27, Bob Durrett wrote: >> >>>On December 09, 2002 at 09:10:16, Uri Blass wrote: >>> >>>>On December 09, 2002 at 08:33:39, Bob Durrett wrote: >>>> >>>>>On December 09, 2002 at 06:47:24, Kurt Utzinger wrote: >>>>> >>>>>>Why last hope? We should be aware of the fact that improvements can/will only be >>>>>>(very) small and average chess people will even not be able to notice. >>>>>>Kurt >>>>>> >>>>>>http://www.beepworld.de/members39/utzinger >>>>> >>>>>So, is it reasonable to infer that chess engines have come to a dead end, where >>>>>the only significant improvements will come from hardware improvements? >>>>> >>>>>Bob D. >>>> >>>>I think that you need to wait some years for it to happen. >>>>Chess engines are not close to their real potential. >>>> >>>>Uri >>> >>>Well, that makes me feel better. : ) >>> >>>Could you say WHY you believe that chess engines are not close to their real >>>potential? What sort of improvements, not resulting from hardware improvements, >>>do you anticipate? >> >>I believe that programs generates trees that are too big and it is possible to >>get the same target by clearly smaller trees. > >That seems to be the main thrust of the "tricks," like alpha/beta. But these >sorts of tricks have been around quite awhile now. They are almost becoming "Ho >Hum." Wouldn't it seem unlikely that anybody will discover any surprising new >trick for making the trees smaller? [Small evolutionary improvements but no big >revolutionary one.] Bob Hyatt, for example, seems to be saying that alpha/beta >cannot be generalized. It's a dead end. "No room for revolutionary >improvements there," according to Bob H. : ) I believe that there is a room for revolutionary improvements. We only need good functions to find if a move is illogical in order to prune it. > >> >>I believe that the evaluation also can be improved significantly. >>Programs do not know about theoretical draws of KRPP vs. KR and it is only one >>example. > >There is still a lot of discussion here about "knowledge," whatever that is. [I >know the definition of "knowledge" in human terms, but not sure of the >definition in the language of chess software.] Maybe that's where the advances >will come from in the evaluation? I think that big part of the advances are going to come from the search by better pruning and better extensions. 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.