Author: Uri Blass
Date: 11:06:23 07/14/02
Go up one level in this thread
On July 14, 2002 at 10:22:02, Rolf Tueschen wrote: >On July 14, 2002 at 09:47:55, Uri Blass wrote: > >>On July 13, 2002 at 15:23:08, Rolf Tueschen wrote: >> >>>On July 13, 2002 at 08:51:33, Uri Blass wrote: >>> >>>>On July 13, 2002 at 05:47:29, Rolf Tueschen wrote: >>>> >>>>>On July 13, 2002 at 04:02:29, Jan Kiwitter wrote: >>>>> >>>>>>On July 13, 2002 at 00:20:02, Dann Corbit wrote: >>>>>> >>>>>>Hi Dan >>>>>> >>>>>>>I think he is around IM level, if I recall correctly. (2100 German rating) >>>>>> >>>>>>In this case your information is wrong. I myself have about 2150 German rating >>>>>>and I am far from playing at IM level. >>>>>> >>>>>>Regards >>>>>>Jan >>>>> >>>>>Some simple truths: >>>>> >>>>>° Human players with FIDE (not US or Britain or some national rankings) Elo >>>>>level of about 2000 or 2100 are incredibly strong amateurs. They all know more >>>>>about chess than any commercial or amateur program. >>>> >>>>I do not believe in it. >>>> >>>>I had fide rating of slightly more than 2000 and I lost it. >>>>I believe that I am at the level of 2000-2100 fide rating(unfortunately a >>>>tournament when I did good results against players with fide rating was not >>>>included in the fide rating). >>>> >>>>My israeli rating that is eqvivalent to fide rating is again more than >>>>2000(2021). >>>> >>>>I expect chess programs to beat me with no opening book >>>>after 1.a3 a6 or 1.h4 h5. >>>> >>>>There are positions when humans with 2000-2100 can play better than the machines >>>>but there are also a lot of positions when machines play better than 2000-2100 >>>>humans. >>> >>>I hope you didn't read me as if I had said that amateurs could beat commercial >>>progs at will. >>> >>> >>>> >>>>> >>>>>° The most important difference between such human amateurs and the progs is not >>>>>their chess but the huge opening books stuff. No amateur is able to know the >>>>>details of so many openings. >>>> >>>>The biggest difference is the speed of the machines. >>>>machines calculates too fast for humans. >>>> >>>>No amatuer is able to search even 1K nodes per second. >>> >>>Why should he need to do that? This is only important in sharp tactical >>>positions. >> >>This is important in almost every position and not only in sharp tactical >>positions. >> >>Humans can find better positional moves by searching deeper even in quiet >>positions. >>Chess programs can also find better moves by searching deeper. > >All true but I must insist on that humans can see certain plans a tempo and the >rest is just the confirmation by some calculations. Machines are unable to do >this. You criticised Crafty's general forbiding any takes on g5/g4. Of course >this is premature. But this is my point. Either you tell them or they must >calculate. But certain position cannot be calculated to a good end. > >> >>I believe that programs have no chance against GMs if you let them to >>change their mind only when they see a move that is at least 0.3 pawns better. > >This surprises me. Thanks for the hint. I never read something about such >numbers. This is not something that I read but a guess. There are a lot of cases when programs change their mind for a move that is less than 0.3 pawns better. It is easy by changing the beta value to tell the program to change it's mind only if it finds something that is at least 0.3 pawns better. I admit that I did not try it in my chess program movei(I guess it can get a Fide rating of 2100-2200 against humans if I am allowed to change my small opening book between games but it is only a guess) but I am almost sure that I am going to find that if I try it the new version is going to play significantly weaker(I suspect that it may be faster in some tactical test positions when it is impossible to find the right move for positional reasons but games are more important for me). I think that I should try it because it is possible that it is better not to let programs to have small changes in the evaluation in the last iterations. Searching a move that is at least 0.3 pawns better than the move that the program suggests is clearly easier task than searching a move that is at least 0.01 pawns better than the move that the program suggest. I strongly believe that not allowing 0.29 pawns change is too much risk but it is still possible that it is better to get some speed improvement by not allowing small changes in the last iterations(I did not test it and I need to test it in games in order to know). 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.