Author: Uri Blass
Date: 15:02:49 09/08/01
Go up one level in this thread
On September 08, 2001 at 16:58:56, Gian-Carlo Pascutto wrote: >On September 08, 2001 at 16:34:45, Uri Blass wrote: > >>Several weeks? >> >>Remember tht it was only deep thought and everyone agree that programs of today >>are not so slow that they need weeks to see things that Deep thought could see >>in few minutes. > >Note that Bruce isn't doing the same kind of SE that they were doing. > >'Everyone agree' > >I obviously don't. A major plus of DB was their use of singular extensions. It >pays of _massively_ in positions like this. It's a large overhead in others. > >As I said before. It's a 20 ply brickwall. If your SE is good you will punch >through it and see it fast enough. If your SE isn't as good you will need >loads of plys more. And time doubles with each one. > >>Common sense tells me that not in the position that you posted and my experience >>tells me that even in the nolot test when there were singular extensions were >>relevant the difference was not 20 plies. > >Robert reported that Cray Blitz needed 10 more moves (thats 20 ply) >to arrive at the eval they did. > >Their speed advantage was not massive at that point (7 to 1 I believe) >so most of this is due to smart use of extensions. > >>I need to see positions when not singular extensions do difference of 20 plies. > >Here! Here! I meant convincing positions from games. Most of the nolot positions are convincing positions from games in the meaning that people could prove a winning evaluation by a tree even if programs cannot see the win from the root. programs can see that every alternative for the opponent out of the tree is bad and they can see that in the leaves of the tree the opponent's position is also bad. The only problem with them is that the difference in plies for singular extensions may be 10 plies or 12 plies but certainly not 20 plies. > >>I have a lot of examples when they were clearly worse than the current comp in >>tactics. > >I don't doubt you have those for DT, but I dont think >you have them for DB2. > >This argument is about moves that other comps can't find. >This is one. > >I don't say it didnt make mistakes. > >>I cannot believe that the only example that you can find when they are better >>in games is something that no program can understand even after you show it the >>lines and go forward few moves in the game. > >Oh, I could try to find others perhaps if I bothered >to look hard enough. But this one is perfect for me. > >They can't find it. DB could. > >You can complain all you want about not being to able to verify >it and you'll only be illustrating my point. > >If you ask for a position that current comps can't find as >a conditione sine qua non for respecting DB's strength, you >aren't going to be able to verify the ones you get, because >that is _exactly_ what you asked for. > >>>I understand the latter explanation is unacceptable to some people here. >> >>It is unacceptable because it against the common sense of people. > >Whatever you name what _you_ think. > >PS. From Crafty's current output, I'd estimate it being able to >find it around ply 21-22. That'll take another 100-300 hours. > >So, perhaps 5-13 days total will do. > >That'd be just lovely. > >Imagine... > >7 days vs 7 seconds > >;) Note that in order to save time you do not need to analyze the root position and seeing the same score in the position one ply after the root position is enough. If you remember the discussion there was an alternative to Cray blitz move in the game few moves after the root position and the first step to convince me( showing me that the alternative is losing) was not done. 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.