Author: Uri Blass
Date: 19:07:17 08/31/01
Go up one level in this thread
On August 31, 2001 at 20:28:57, Slater Wold wrote: >On August 31, 2001 at 19:41:41, Bruce Moreland wrote: > >>On August 31, 2001 at 19:31:12, Slater Wold wrote: >> >>>** Weird. It takes DT-2 at least 6 hours to find this, while Deep Shredder >>>finds the _EXACT_ same varation in a little over 8 minutes. However, like Bruce >>>says, there sure isn't a big score. Deep Shredder thinks it's the best move, >>>but only based on it loses the least. While actually it's winning. ** >> >>I liked your post, but I'll respond to this one small section. > >Thanks. :) > >> >>I think that one of the reasons the Nolot test is interesting is that we can >>compare our programs with DT circa 1994. > >When I started fiddiling with computer chess 2-3 years ago, I thought they were >amazing. A year ago when I found "traces" of these chess playing computers from >the 80's and early 90's and I was astouned. A computer, in 1994, playing chess >on a level that every programmer at this board is striving to acheive. Granted, >we are trying to acheive it on a more, affordable hardware. However, it seems >strikingly clear that 90% of the computer chess advances have come from HW and >NOT better code. This is _SIMPLY_ proved by seeing DT-2 vs Shredder, Ferret, >Crafty, Tiger, or Fritz on today's top HW. > >> >>Based upon the results I have seen, produced by both my program and others, I >>think we are getting close to DT. We're certainly in the same ballpark with >>regard to heavy king tactics. > >Yes, I agree here 100%. Tactics I think we have come full circle. >Unfortunatly, it's positional awareness that I think most engines lack. I believe that chess programs are clearly better positionally than Deep thought. programmers of top programs worked a lot in imporovement of their evaluation function when Hsu had smaller experience in this task and we could not compete against good opponents like programs of today can compete in comp-comp games The experience against humans was also smaller than the experience of today because playing on the internet was not common like today so they could not test their program enough against GM's. The games of Deep thought also prove that Deep thought was weaker in tactics then the top programs of today. The fact that they are in similiar level in the nolot test proves nothing because the singular extensions helped Deep thought in the nolot test more than it helped it in games. I find that Ferret is faster than Deep thought in most of the Nolot positions that were solved(5 out of 8). It is faster in 1,2,7,10,11 3,6,9 were never solved when I do not count 8 because hsu has doubts about it. 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.