Author: Uri Blass
Date: 12:14:43 09/06/01
Go up one level in this thread
On September 06, 2001 at 14:36:12, Robert Hyatt wrote: >On September 06, 2001 at 13:24:48, Uri Blass wrote: > >>On September 06, 2001 at 11:37:06, Robert Hyatt wrote: >> >>>On September 06, 2001 at 10:48:42, Uri Blass wrote: >>> >>>>On September 06, 2001 at 10:00:52, Robert Hyatt wrote: >>>> >>>>>On September 06, 2001 at 08:22:23, K. Burcham wrote: >>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>>>amazing game. i always look for moves that you played for the computer, >>>>>> to get the results you want. >>>>>> >>>>>>in this game, deep fritz chooses all the program moves that you list here. >>>>>> >>>>>>in your games that you sacrifice material for position, and your "every move >>>>>> advance toward king" methods, this makes your games look very easy for >>>>>> a human. and this makes these games look like you are playing an >>>>>> easy opponent. >>>>>> >>>>>>it seems that if these (all) programs are this vunerable to kingside >>>>>> attacks, then it would also seem that huebner (spelling?) could >>>>>> have used these methods in his comp games. and it would also seem >>>>>> that kramnik could use these methods in his upcoming match. >>>>>> >>>>>>and if kasparov is so good, and the king, and he used a comp for studies, >>>>>> then why didnt he use these methods with deep? are you going to >>>>>> say that deep would not choose these moves. are you going to >>>>>> say that deep would not fall for ...h5, and would develop >>>>>> its own attack. >>>>> >>>>> >>>>>Deep Blue is a different animal from Deep Fritz. First, it was at least >>>>>a hundred times faster. Which plugs several holes. Second, its evaluation >>>>>was developed and tuned while playing against human GMs, not against other >>>>>computers. That plugs several more holes. >>>> >>>>We do not know if it was a different animal because kasparov did not try h5 and >>>>he had the opportunity to try in game 2. >>> >>>OK. I will re-phrase that. _I_ know that deep blue is a completely different >>>animal from deep fritz. 100X faster. Better tuned against GM players. No >>>null-move. Singular extensions. The differences are almost too many to >>>mention. >>> >>> >>> >>> >>>> >>>>I also doubt if GM's tried the idea of 1.e4 e5 2.Nf3 Nc6 3.Bb5 a6 4.Ba4 h5 >>>>against deeper blue. >>>> >>>>I doubt if tuning the evaluation based on games against humans could help >>>>because there is a lot of things that humans did not try. >>> >>>You only need reasonable king safety. With that, 4... h5 is simply going to >>>get busted on principle. >> >>I do not know how can you be sure that it is so simple. >> >>Do you know what is the losing mistake of White in the game? > >This has nothing to do with "the game". It is simply about having watched >lots of Fritz games, and lots of DT/DB games. > > >> >>Is it so clear to know when sacrificing a pawn for a king attack works and when >>white can stop the attack and wins thanks to the pawn advantage? > >A deep search, + a good evaluation can do this. Whether DB could do this in >every game is unknown and not very likely to boot. But that it is probably >better at doing it than the current micros is a pretty sure bet, as I watched >lots of "attempted attacks" fail in the various demos I saw played. The difference is that the top programs of today are available and people can learn their weaknesses by not public games Deeper blue was not available so players did not know it's weaknesses. I am also sure that there are a lot of attempted attacks against the top program of today that fail but if an attack fails it is no news when every success is news. <snipped> >No program of today will see that repetition. Try forcing Kf1 and >seeing if it then realizes that Black can repeat. I did not say that programs can see the repetition but that they can see Kh1 for good reasons. > > > > >> >>It does not need to understand that Qe3 gives black a draw but only needs to >>understand that Qe3 gives black better chances when after Kh1 black has no Qe3. > >That does not compute, to me. The queen at e3 is not particularly meaningful >in most positions. So it has to see the _reason_ that the queen reaching e3 is >good here. And it can't. It might not like the queen that centralized, for >general positional reasons, of course. But that doesn't mean that it is playing >Kh1 for the right reason in this position. The reason that Kh1 does not give black the chance of Qe3 is good enough. Humans are also going to choose Kh1 even without seeing the draw by Qe3 if they understand that after Kf1 Qe3 black has chances when after Kh1 black has no chances and has to go to a losing endgame. It is about king safety's evaluation The micros can see that the white king is not safe after Kf1 and black has chances by Qe3 when deeper blue could not see it. It is possible that deeper blue saw that both kings are not safe and simply added king safety scores. If it did it then it is clearly wrong to do it because if both kings are not safe you cannot be sure about the result and the evaluation should be closer to a draw. 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.