Author: Robert Hyatt
Date: 10:40:32 08/21/02
Go up one level in this thread
On August 21, 2002 at 07:53:40, Vincent Diepeveen wrote: >On August 20, 2002 at 23:14:51, martin fierz wrote: > >But this is ?? BLUNDER moves. The most horrible to make. Moves that >lose a game directly. I don't know a single engine that >makes such moves anymore. What do you know about the circumstances where those moves were played? Anything at all? IE we know that in the blunder against fritz in 1995, a communication problem was directly responsible for it. I saw them make at least one other blunder due to a similar problem beyond their direct control. I also saw _my_ program have its share of such problems, since we all used simple modems and long-distance phone lines across the country. > Crafty finds 3 instantly bad and >one it to my amazement doesn't find. Says something about how bad >crafty is (but if you look to its primitif evaluation you see >why). If it says "how bad crafty is" then what does it say about _your_ program? IE for the last 20 games (standard time control, 60 10 or 30 30) we have played on ICC, on hardware that probably favors you somewhat. IE I would rather use a dual 1.4 than a quad .7. Crafty has won 10, drawn 5 and lost 5 vs diep. Perhaps you should stop saying how bad Crafty is until you improve your program enough to beat it more than it beats you? It would look a bit better then, when you make such ridiculous statements... > Even then it solves 3 out of 4 positions. So it prevents >3 clear losses there. > >I concluded the same as Uri. The level was so bad in these days >that you won't find any good move like the programs play now. > >Take the rxc4 sacrafice of junior. We can explain why it plays it, >but doesn't take away that it is a real good move. > >If you look to the 1997 match dbii - kasparov you won't find many real >good moves. The majority is found within a second and kept by the >current software generation. Some others which Seirawan comments with >'!' do not deserve a !, because they are too simple to find. Like 2 ply >moves which a very old genius version (called differently) from around 1986 >at my pc also finds under DOS instantly > >Best regards, >Vincent > >>On August 20, 2002 at 20:36:02, Uri Blass wrote: >> >>>On August 20, 2002 at 20:27:18, martin fierz wrote: >>> >>>>this kind of test is fundamentally flawed by being 100% biased: you are >>>>presenting a selection of positions where deep blue failed, in every single one. >>>>of course, if a program of today solves a single one of these, be it by luck or >>>>by better knowledge, it already looks good. >>>>there are surely lots of positions where deep blue would look good in comparison >>>>to a micro, but they are not included. >>>> >>>>for any meaningful comparison, you should get a set of test positions and run DB >>>>and your micros over it. of course you can't do that now. but if you can't make >>>>a meaningful comparison, the next best thing is to make none at all. not to make >>>>a meaningless comparison :-) >>>> >>>>aloha >>>> martin >>> >>>The problem is that I know of no good moves that deep thought played and the >>>micro cannot find. >>> >>>If I find a lot of blunders that deep thought played when most of the top >>>programs of today avoid most of them when nobody can show me good moves of deep >>>thought that most programs need hours to find them,then it suggest that deep >>>thought was inferior relative to the programs of today. >>> >>>Uri >> >>1) you did not find "a lot" of blunders. you posted about 4. >>2) your conclusion is completely invalid. i could do the same with ANY of the >>top programs, and come to the conclusion that each single program is much worse >>than all others. one example that comes to mind is the loss of tiger against >>smirin, which probably all other top engines would have avoided, if i remember >>right. i bet you can find 3 more examples like that if you try, and then >>conclude that tiger is inferior relative to the top programs of today. which is >>obviously just wrong. >> >>aloha >> martin
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.