Author: blass uri
Date: 07:43:23 12/26/99
Go up one level in this thread
On December 26, 1999 at 07:45:39, Vincent Diepeveen wrote: >On December 25, 1999 at 03:34:11, blass uri wrote: > >>On December 24, 1999 at 20:00:48, Vincent Diepeveen wrote: >> >>>On December 24, 1999 at 18:39:00, blass uri wrote: >>> >>>>On December 24, 1999 at 17:17:01, Vincent Diepeveen wrote: >>>> >>>>>00:25 391314 (0) 8 -0.48 c2-c4 Qd8-h4 g2-g3 f4xg3 f2xg3 Rg6xg3 Nd5-f6 Ke8-e7 Qd1 >>>>>-e1 Rg3-h3 >>> >>>>This line shows that it does not see it for the right reasons. >>> >>>remember, the lines are from hashtable and it's a fail low, >>>obviously fail low doesn't show the correct line. >>> >>>>After c4 Qh4 g3 the right move for black is Qh3 and not f4xg3 >>>> >>>>The line f4xg3 f2xg3 Rg6xg3 is not clearly winning for black and I guess that >>>>deep thought expected this line when it played c4. >>> >>>>White can continue Qe1 Rh3 Qxh4 Rxh4 cxb5 and the question if g3 is better is a >>>>positional question and not a tactical question. >>>> >>>>The evaluation of programs after f4xg3 are less than 1 pawn for black. >>> >>>>c4 is a tactical blunder not because of this line but because of the line with >>>>Qh3 >>> >>>Right. It just already sees something that's evil enough. Very likely >>>not this line, but something that has to do with Qh3. We're just needing >>>a bound that's it. Then we get a fail low. >> >>I tried Genius3 with selective depth 0 after c4 Qh3 g3 Qh3 Rg1 and it needs 8 >>plies to see that Qxh2 is winning. >> >>so I guess that it needs 13 plies with no extensions before c4 to see that c4 is >>a tactical mistake. >> >>I see lines of 2 plies when I tell genius3 1 ply with no extension so 13 >>plies+no extensions is practically 14 plies. >> >>My point is that you need to see 14 plies to reject c4 for the reason of Qxh3 >>and not for a positional reason. >> >>If Deep thought had some rules not to extend wasting tempo lines that are also >>giving the queen for a pawn that it may explain the tactical mistake. >> >>Uri > >Uri your test is not accurate. It is not 13 ply. In that selective depth >of genius checks are seemingly included in that depth, where all normal >programs on the world don't count them. > >You need 9 ply with a piece square table program to see it. If you extend >*all* checks but not do it in qsearch i must figure still out whether you >need 9 or 10 ply. > >For progs with a bit more knowledge you need only 8 ply . > >Vincent I agree that normal programs do not need more than 8-10 plies but it is because they do check extensions. I guess that deep thought had some rules when not to do check extensions(if the side to move waste a tempo and also sacrifice a queen for a pawn) and it is the reason that they could not see it. Hsu knows better about it and it is interesting to know what is his explanation for the tactical blunder of deep thought. 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.