Author: Alvaro Polo
Date: 08:06:13 07/19/00
Go up one level in this thread
On July 19, 2000 at 08:14:56, Amir Ban wrote: >On July 19, 2000 at 03:55:44, blass uri wrote: > >>On July 18, 2000 at 19:10:46, Amir Ban wrote: >> >>>On July 18, 2000 at 14:05:46, Jeroen Noomen wrote: >>> >>>>On July 18, 2000 at 09:29:12, Amir Ban wrote: >>>> >>>>Amir, >>>> >>>>I agree that Junior earned its points honestly. I also agree with most you write >>>>about these games. Still, you don't point out anything about the losses against >>>>Kramnik and Piket. And that was exactly what I had in mind writing this thread. >>>>Those two games showed exactly where chess computer programs still can be >>>>improved. And HAVE to be improved, otherwise human GM's will have good chances >>>>to get more points next year. And they will, because they have learnt. >>>> >>>>IMO if you solve most of the problems about king's attacks and closed positions, >>>>then it will be almost impossible for the strongest GM's ta beat a computer. >>>>Because in that case they have no advantage in any type of position anymore. But >>>>in 2000 there is still not much to be done when a clever player manages to block >>>>the position or start a slow attack: The programs do not know about this and >>>>only human mistakes will save them. >>>> >>>>So the crucial question is: When will one of the leading programmer stop >>>>searching for higher NPS, better searching techniques etc? When somebody will >>>>REALLY tackle the 2 problems I mentioned? Because otherwise a computer can still >>>>be beaten in 2010, running on 500 GHz. But as I already mentioned: This is the >>>>computerchess paradox: NOBODY wants to sac NPS for more knowledge. And as long >>>>as nobody wants to quit this 'rule', human GM's are still superior in knowledge >>>>and understanding of the game. >>>> >>>>Jeroen >>>> >>> >>>The speed vs. knowledge dilemma is a false one. It may apply to Rebel and other >>>programs, but it doesn't apply to Junior, where I have a framework to code >>>evaluation stuff virtually for free. >> >>2 questions: >>1)I guess that the fact that you can add evaluation stuff virtually for free >>in run time make adding knowledge to the evaluation less simple and you need >>more time to do the design decisions to change the evaluation function relative >>to other programs. >> >>Am I correct? >> > >No I'll believe that adding new knowledge to Junior is almost free. I have then two questions. 1.- Why isn't then Junior's evaluation much better? Please don't misunderstand me. I am sure it has a great evaluation but, one may think that when things are almost free you could just add any bit of knowledge that you might consider useful under any circumstance and have a really astounding, hypergreat, out of this world evaluation. 2.- Assuming that DJ already outsearches GMs and assuming that its evaluation will soon be better than GMs also, when do you believe DJ will beat the reigning WC in a match? Alvaro Polo > > >> >> >>2)I know that Junior6a does not have a trapped knight code at least in part of >>the positions. >>Here is one position that demonstrates it from the game Junior6a-tal(15 minutes >>per move with no pondering) >> >>[D]r4rk1/pp1n1pbp/2p1p1p1/3nP3/2B2P1q/2N4P/PPPBQP2/2KR3R w - - 0 1 >> >>Junior blundered by 14.Qg4 and only after 14...Qxf2 it understood that it lost a >>pawn(15.Bxd5 cxd5 16.Nxd5 h5 17.Qg1 Qxg1 18.Ne7+ is bad because the knight is >>trapped in e7). >> >> >>Can you evaluate this stuff virtually for free? >> > >In principle yes, but I wouldn't do it in this case because it's not clear how >to formulate it. The knight is not really trapped. It can be defended by >friendly pieces. There may be a bishop on c8 that could be exchanged for it. In >the late endgame with few pieces on board the knight would not be trapped at all >because the back rank is free, etc. etc. > >It's a common error to code an evaluation term with a specific position in mind, >and then find that your program applies it in dozens of positions where you >don't want it applied. > >Amir
This page took 0.02 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.