Author: Uri Blass
Date: 22:45:42 09/03/01
Go up one level in this thread
On September 03, 2001 at 22:02:35, Mike S. wrote: >On September 03, 2001 at 19:25:38, Uri Blass wrote: > >>(...) >>I believe that using computers only to check blunders is a common mistake >>and humans simply underestimating the computers. >> >>They are not good only in tactics and they can give you after a long search a >>good positional move. > >What is your impression, after which time a noticeable increase of the >positional qualitiy occurs in general (please give your hardware data also)? Is >this different, depending on each program? It depends on the program and on the position. There are no clear rules. There are positions when one program needs 20 hours to find the right positional move and there are positions when it needs 20 minutes to find the right positional move. I posted a position when gandalf needed only few seconds to find the right move when Deep Fritz could not find the right move even after many hours. note that Tiger has also problems with the positions but I did not tell it to search for many hours like Fritz(it does not choose the same move of Fritz but also does not choose Bc5). The main problem with Deep Fritz's analysis is that the score went down every iteration so I decided to give it to analyze for a long time later positions and when the score continued to go down I decided that Deep Fritz's move is not a good idea. I am not sure if it is losing but the analysis convinced me that it does not give me the advantage when I believe that Bc5 gives me a better position. The hardware that I use is p800(128 mbytes RAM) and p850(256 Mbytes RAM) when p800 is used most of the time. I now have only 3 games so even without using the hardware most of the time for my games I still has more time than my opponents(one of them told me that he plays 48 games in the same time and he also cannot give his PII450 24 hours to analyze one interesting position because of the fact that the computer is used for his children to play other games). > >Furthermore I'd be interested, if you can give examples (or explanations) of >which kind of positional insight the programs get at correspondence analysis >time, which they usually don't get under normal time conditions. Are these still >"static" positional things, or does a kind of planning become visible etc.? Not only "static" positional things. > >Can you express a general opinion of what current programs can do positionally, >and what they can't (yet)? I know about one case when none of my programs could find a move because they underestimated the endgame. I believe that the move that programs suggested is also enough to win but I could prove a bigger score by my move and it is not because of not analyzing the positions after the program's move. I believe that there are some endgames that all programs fail to evaluate correctly and they may fail to find the best move in the middle game because of not evaluating correctly the endgame. I have no clear rule about positions when an endgame is not relevant in the near future. 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.