Author: Robert Hyatt
Date: 20:20:58 01/24/03
Go up one level in this thread
On January 24, 2003 at 07:40:14, Uri Blass wrote: >On January 24, 2003 at 07:24:48, Rolf Tueschen wrote: > >>On January 23, 2003 at 21:00:37, Robert Hyatt wrote: >> >>>On January 23, 2003 at 10:44:44, Uri Blass wrote: >>> >>>>On January 23, 2003 at 10:38:45, Robert Hyatt wrote: >>>> >>>>>On January 23, 2003 at 02:18:46, Dux Kazer wrote: >>>>> >>>>>>On January 22, 2003 at 21:24:53, Robert Hyatt wrote: >>>>>> >>>>>>>On January 22, 2003 at 14:01:09, Rolf Tueschen wrote: >>>>>>> >>>>>>>>On January 22, 2003 at 13:02:10, Robert Hyatt wrote: >>>>>>>> >>>>>>>>>On January 22, 2003 at 12:27:56, Dux Kazer wrote: >>>>>>>>> >>>>>>>>>>On January 22, 2003 at 12:06:37, Matthew Hull wrote: >>>>>>>>>> >>>>>>>>>>>On January 22, 2003 at 11:58:05, Christopher A. Morgan wrote: >>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>>Bob, >>>>>>>>>>>> >>>>>>>>>>>>It shows me the abality of GK to negoiate a rule very favorable to him. >>>>>>>>>>>>It is not at all certain that GK could, over the board, be certain of a >>>>>>>>>>>>draw in a known draw position as determined with tablebases with, at least all >>>>>>>>>>>>5 piece endings, and most likely some six piece endings. Now, in those >>>>>>>>>>>>positions the game will end in a draw, which, in my view, is correct. This >>>>>>>>>>>>does not address the situation where DJ sees a tablebase draw in its search and, >>>>>>>>>>>>if it's losing trys to steer the game to that position. >>>>>>>>>>>> >>>>>>>>>>>>I like the rule. I do not see any contest between machine and man where >>>>>>>>>>>>the machine looks up its move in a table, and waits for the human to make >>>>>>>>>>>>a mistake. >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>>It is possible the machine could see a tablebase draw which a human would not >>>>>>>>>>>know how to "solve" and thus lose the drawn position. The human would deserve >>>>>>>>>>>the loss. This is the point of the man/machine contest. >>>>>>>>>>> >>>>>>>>>> >>>>>>>>>> Oh Yes... but let the machine play without the tablebases and it will lose even >>>>>>>>>>simple knight vs rook draw for sure, not to say KRP vs KR.. >>>>>>>>> >>>>>>>>>Not necessarily. Some programs can play krp vs kr pretty well without tables. >>>>>>>>>I have >>>>>>>>>special code to handle just this case, for example. I'm sure others do too. >>>>>>>>> >>>>>>>>>I'd play _anybody_ KR vs KN with crafty having the KN side... and not expect to >>>>>>>>>lose. >>>>>>>> >>>>>>>> >>>>>>>>Another challenge to human chess players. Hopefully someone bites. I'd like to >>>>>>>>see this one too! >>>>>>>> >>>>>>>>Marvelous. >>>>>>>> >>>>>>>>Rolf Tueschen >>>>>>> >>>>>>>this one is too easy. IE I will play kn vs kr without tables. I'll also >>>>>>>play KQ vs KR without tables playing either side, knowing crafty can win this >>>>>>>ending _easily_ without tables at all. >>>>>>> >>>>>>>I don't think it much of a challenge to avoid losing kr vs kn. Any decent >>>>>>>search depth will find the simple tactics where the knight is lost. >>>>>> >>>>>> I don´t think is that simple.... i know good programmers have special code to >>>>>>handle that kind of ending but at least the engine has to think for itself and >>>>>>of course that is time consuming (so human can use that time for himself right?) >>>>>>and there is always some chance in that case.. i have seen Crafty beaten Fritz >>>>>>many times in Rook vs Knight (of course without table) and not to say so many >>>>>>blitz game where human confuse the machine to go for a dead draw KRPP vs KR!. >>>>> >>>>>Fritz is a bad example. KR vs KN is only won by zugzwang, when the weaker side >>>>>makes a mistake. Fritz is very susceptible to zugzwang positions because of the >>>>>null-move >>>>>search. >>>>> >>>>>I have seen crafty win more than one blitz game KR vs KN without tables. But >>>>>only blitz >>>>>games. At longer time controls, it simply isn't winnable unless the opponent >>>>>makes an outright >>>>>blunder. There are a "few" deep wins that a table might spot. But against a >>>>>human, I don't >>>>>think kr vs kn can be won by the kr side, without the tables, and even with the >>>>>tables, you can >>>>>look at the krkn.tbs file to see that the draws outnumber the wins by a huge >>>>>margin. >>>>> >>>>>KQ vs KR is another example that a program can handle simply and almost >>>>>perfectly with >>>>>a minimal search. >>>> >>>>In both cases you need evaluation that Movei of today does not have. >>>> >>>>In KR vs KN you need to know to keep the knight clode to the king and in KQ vs >>>>KR you need to know that the stronger side needs to reduce the distance between >>>>the kings. >>>> >>>>Uri >>> >>>I can only quote what my program can do. IE it can win KQ vs KR against a >>>program with tables, with just one second per move... The knowledge required >>>is _really_ modest. >> >>Just two boring questions as usual. :) >> >>1) Could you explain with words how this is possible now? I mean real champs >>didn't know how to solve it and you do without tables? Or did you hide the >>tables in micro format? <g> > >It is possible because the search of programs even at 1 minute per game can >discover things that the real champs have problems to see. > > >> >>2) Because you said "just 1 minute", let me ask you if you believe in the myth >>that by each generation (each year) we win a 2x hardware speed and therefore >>after a couple of years we have (allegedly) the strange effect that we could >>play the earlier tournament time mode in say 0,6 seconds for the whole (sic!) >>game. There is a debate in Germany and I also wrote aboute it in >> >>http://hometown.aol.de/rolftueschen/SmallTalk.html >> >>Could you give a few factors such a maths above perhaps had overseen/forgotten? >>I want to quote you. Thanks. > >I think that they forgot the fact that the hardware is not twice faster every >year and the progress in hardware is going to stop sometime in the future. > >300 Mhz were used in the end of 97 in the microcomputer world championship in >paris (Today, more than 5 years later we do not have 300*32=9600Mhz. > >Uri Moore's law is running on a roughly 18 month cycle. 1.5 years. 5 years == three doublings. 8 * 300 is 2400, which is a bit behind, since we are at 3.0+ today.
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.