Author: Vincent Diepeveen
Date: 14:07:15 01/25/05
Go up one level in this thread
On January 25, 2005 at 17:03:42, Vincent Diepeveen wrote: >On January 25, 2005 at 16:53:03, Olaf Jenkner wrote: > >>> >>> >>>but to store a 32 piece tablebase would be a lot 'smaller'. >>> >>>might a 2.5 by 2.5 kilometre crystal do the trick ? >>> >>> >>>duncan >> >>We have about 10^42 positions to store. >>The third root is 10^14. >>Take a 1000^3 km crystal. You must store 100000 positions >>at one millimeter. 100 at one micrometer. Maybe, the crystal was too >>big. >> >>Perhaps 99,999999% of the legal positions will never be >>necessary to compute the tree. Than we can take a smaller cube. > >There is a huge difference between a 32 stone EGTB and calculating till the >bitter end. > >What you refer to is having a say 10^30 hashtable and calculating say 10^31 in >order to play perfect. > >It's very well possible that you can get optimal play (optimal defined as: >"winning in case position is won, and drawing in case you can get a draw". Not >as the optimal DTM which is major BS where majority doesn't care a shit about; a >win is a win) long before searching 10^43. > >However please realize that storing 10^20 in hashtable means also that you will >see certain positions within that 10^20 a million times. > >So the number of overwrites will be real huge from a single position. That means >obviously that you keep researching the same tree over and over again in all >kind of ways. > >It's unlikely you'll have a perfect move ordering and it's unlikely that you can >optimal make profit from nullmove and hashtables. > >So there always will be this unsureness when that happens. > >What we do know from other games is that software even when playing near perfect >still lost to mankind, despite that certain openings already had been solved >simply to win/draw/loss. > >Now that losing or winning from mankind is not interesting in this case, but >more interesting is that it indicates that despite nearly directly being in >EGTBs they still didn't play perfect. > >If that will be the case in chess, we sure have a long way to go! >Vincent Actually my personal guess is that when we have 20 men EGTBs that programs will always win a won position and always draw a drawn position. You just go 1.d4 or 1.e4 (most likely 1.e4 is winning) and it's game over. Find me 1 GM who believes that in 100 years from now sicilian isn't a clear 1-0. Vincent >>OJe
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.