Author: Robert Hyatt
Date: 11:13:09 07/18/00
Go up one level in this thread
On July 18, 2000 at 11:18:50, blass uri wrote: >On July 18, 2000 at 11:05:03, Robert Hyatt wrote: > >>On July 18, 2000 at 00:31:57, Peter Kappler wrote: >> >>>On July 17, 2000 at 22:26:53, Robert Hyatt wrote: >>> >>>>On July 17, 2000 at 22:18:50, Robert Hyatt wrote: >>>> >>>>>On July 17, 2000 at 20:40:57, blass uri wrote: >>>>> >>>>>>On July 17, 2000 at 20:12:36, Robert Hyatt wrote: >>>>>> >>>>>>>On July 17, 2000 at 19:38:01, blass uri wrote: >>>>>>> >>>>>>>>On July 17, 2000 at 19:10:29, KarinsDad wrote: >>>>>>>> >>>>>>>>>On July 17, 2000 at 11:59:35, blass uri wrote: >>>>>>>>> >>>>>>>>>>On July 17, 2000 at 09:32:18, Robert Hyatt wrote: >>>>>>>>>> >>>>>>>>>>>On July 17, 2000 at 08:05:57, blass uri wrote: >>>>>>>>>>> >>>>>>>>>>>>On July 17, 2000 at 07:22:41, Graham Laight wrote: >>>>>>>>>>>> >>>>>>>>>>>>>I'm afraid I still feel that Junior could have come out ahead (instead of >>>>>>>>>>>>>level)in this tournament by beating Bareev and Khalifman - and possibly by not >>>>>>>>>>>>>losing with such apparent ease to Kramnik. Continuing the game against Anand >>>>>>>>>>>>>might possibly have gained an extra half point as well. >>>>>>>>>>>>> >>>>>>>>>>>>>I think that Amir has an aspiration to make his program demonstably better than >>>>>>>>>>>>>Deep Blue (this certainly comes across in his interviews published on the >>>>>>>>>>>>>Chessbase Website coverage of Dortmund (www.chessbase.com) before the Kramnik >>>>>>>>>>>>>game). If so, as a (hopefully!) impartial member of the viewing public, I'm >>>>>>>>>>>>>afraid to say that I've yet to be convinced. >>>>>>>>>>>>> >>>>>>>>>>>>>As evidence, I point firstly to the games against Bareev and Khalifman. On both >>>>>>>>>>>>>occasions when Deep Blue '97 gained an advantage over Gary Kasparov (who's a >>>>>>>>>>>>>better player than anyone at Dortmund was), it parlayed that advantage into >>>>>>>>>>>>>victory - whilst Deep Junior twice failed conspicuously to "slam in the lamb". >>>>>>>>>>>>> >>>>>>>>>>>>>I would also point to the game against Khalifman. Here we see Deep Junior lose >>>>>>>>>>>>>to a combination of blocked centre and king attack - classic anti computer >>>>>>>>>>>>>methods which have both been well known for a long time. They work because, in >>>>>>>>>>>>>this case, nothing short of truly massive search depth is going to help you to >>>>>>>>>>>>>make the correct moves. >>>>>>>>>>>>> >>>>>>>>>>>>>However, for both king attack and blocked centre, Deep Blue '97 demonstrated >>>>>>>>>>>>>that it's evaluation knowledge was able to adequately handle the challenge. >>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>>I guess that the evaluation of Deep Junior could do better if Deep Junior could >>>>>>>>>>>>search the same number of nodes. >>>>>>>>>>>> >>>>>>>>>>>>I believe that Deep Junior is better than Deeper blue if you assume 200,000,000 >>>>>>>>>>>>nodes per second for deep Junior. >>>>>>>>>>>> >>>>>>>>>>>>Uri >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>>I believe pigs can fly. But only if you increase the density of the atmosphere >>>>>>>>>>>by a factor of 10,000 or so. >>>>>>>>>>> >>>>>>>>>>>DB has two almost insurmountable advantages: (1) it is faster than anything is >>>>>>>>>>>going to be for a _long_ time; (2) using special-purpose hardware they did >>>>>>>>>>>everything in the eval that was suggested by GM players, because they could do >>>>>>>>>>>so with no speed penalty. >>>>>>>>>> >>>>>>>>>> >>>>>>>>>>Deeper blue had one significant disadvantage. >>>>>>>>>>They had no time to test their evaluation. >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> DJ and every other PC program has _many_ >>>>>>>>>>>"concessions" in the evaluation due to speed considerations. DJ's king safety >>>>>>>>>>>would fail if it was 1,000 times faster... because there are some things that >>>>>>>>>>>speed won't help until we reach the point where the computer can see 30-50 plies >>>>>>>>>> >>>>>>>>>> >>>>>>>>>>I think that these things are not relevant in the games that it lost. >>>>>>>>>>I think that in the game against kramnik the mistake of deep Junior was Kh8 and >>>>>>>>>>Deep Junior could see 4 plies after it that it is in trouble. >>>>>>>>> >>>>>>>>>You seem to be contradicting yourself. >>>>>>>>> >>>>>>>>>You are stating that the lack of the king safety failure due to the event >>>>>>>>>horizon is not relevant and then turn around and state the DJ found out it was >>>>>>>>>in trouble 4 ply later (once king safety failure was in scope of the event >>>>>>>>>horizon). >>>>>>>>> >>>>>>>>>4 Ply is 6^4 or about 1300x faster hardware required. >>>>>>>> >>>>>>>>I disagree >>>>>>>>computers can see 4 plies in the important lines often by being 50-100 times >>>>>>>>faster or even less than it. >>>>>>>> >>>>>>>>The 4 plies are not quiet moves so I will not be surprised if being only 20 >>>>>>>>times faster is enough. >>>>>>>> >>>>>>>> Or 2000x more processors. >>>>>>>>>Event horizon for king safety is totally relevant here, otherwise, DJ would have >>>>>>>>>probably played a different move. >>>>>>>>> >>>>>>>>>If I was a GM, I would attempt to add positional elements that the program would >>>>>>>>>not detect until way later in the game such as permanent weaknesses for the >>>>>>>>>program and permanent strengths for the GM. >>>>>>>> >>>>>>>>I agree that adding knowledge is important but my point was that I believe that >>>>>>>>Deep Junior is better than Deeper blue based on equal numbers of nps >>>>>>>> >>>>>>>>Uri >>>>>>> >>>>>>> >>>>>>>Based on what??? >>>>>> >>>>>>I think 5.5 out of 9 in this tournament is better than 3.5-2.5 against kasparov >>>>>>inspite of the fact that 5.5 out of 9 is slightly worse performance because >>>>>>kasparov could not train at home against something similiar to deeper blue when >>>>>>the players in this tournament could train at home against something similiar to >>>>>>Deep Junior that can produce almost the same moves. >>>>>> >>>>>>I think that Deep Junior could get 5.5 out of 9 with the same number of NPs as >>>>>>Deeper blue because my Junior5.9 see the problem in the game against kramnik 4 >>>>>>plies after the losing mistake and I guess that seeing more 4 plies that 3 of >>>>>>them captures can be done with 200,000,000 nps. >>>>>> >>>>>>I also guess that Deep Junior has a chance to find 8.h4 that is probably >>>>>>winning(based on alterman's words) against piket because the difference in >>>>>>evaluation between this move and the game move was small. >>>>>> >>>>>>Junior had also good chances to translate the advantage in other games with >>>>>>better hardware so I guess that it could get at least 5.5 out of 9 with the same >>>>>>number of nps. >>>>>> >>>>>>Uri >>>>> >>>>> >>>>>That is all well and good... But _what_ does it have to do with deep blue vs >>>>>junior? By the time Junior can hit 200M, DB would be doing 400B nodes per >>>>>second, assuming they wanted to continue development. So making comparisons >>>>>between a deep junior that won't exist for about 10 years after the development >>>>>of DB2 doesn't make a lot of sense... as there is no reason that DB2 would be >>>>>the 'last' of the line, if they wanted to continue. >>>>> >>>>>Amir wasn't talking about being better than DB at 200M nodes per second, he >>>>>seems to believe he is better at under 2M nps. That I have a very difficult >>>>>time considering, even under the wildest of circumstances... >>>> >>>> >>>>Here is a reasonable way of thinking about this: >>>> >>>>Let's take Dortmund and the DB97 match and consider the games. >>>> >>>>two questions: >>>> >>>>1. In the Deep Blue games, if you go over them, what serious weaknesses do >>>>you see? Not the overlooked draw as _no_ program has ever come close to finding >>>>that. In the 96 match, Kasparov found a problem and wore the machine out. In >>>>1997 what did he find? >>>> >>>>2. In the Deep Junior games, if you go over them, what serious weaknesses do >>>>you see? One obvious one is king safety. What do you think would have happened >>>>had Dortmund been 12 rounds rather than 9? Did you notice a trend over the last >>>>few games? Do you think that would have continued? >>>> >>>>Based on the above, I challenge you to justify any claim of equality or >>>>superiority between DJ and DB97. I don't think there is any comparison at >>>>all. DB outplayed Kasparov in blocked positions, in open positions, in >>>>endgames Kasparov thought he should win, etc. >>> >>> >>>I don't think DB outplayed Kasparov to the extent that you think. Certainly not >>>in game 1. Yes, definitely in game 2. But in games 3-5, it was GK who had the >>>better chances throughout. In those 3 games I think you have to give a slight >>>edge to GK. Game 6 was clearly the worst game of his career, but it's not like >>>it showcased any brilliant play by DB - it just was just a case of Garry >>>suffering a total psychological breakdown and handing the game over. >>> >>>All that having been said, I do agree that DBs play in '97 was more impressive >>>than DJs in Dortmund. >>> >>>--Peter >> >> >>I wouldn't begin to claim that DB "outplayed" kasparov in 97. I do claim that >>it "beat" him, of course. :) >> >>But in the above, the point is can you find any specific weakness in DB that >>would lead to GMs discovering that and beating it like a drum? Can you find >>any weakness in Deep Junior that would lead to GMs discovering that and beating >>it like a drum? >> >>That is the main difference I see. We _all_ saw the king safety/blocked >>position problem in Dortmund. We didn't see any such problem in DB'97. > > >It does not prove that the evaluation of DB'97 was better about king safety. >I guess that with 200M nps Junior could find better moves and avoid the king >attacks in the games. There you are simply wrong. No matter how deep you search, there is _always_ a position that doesn't expose the tactics until one ply deeper. If your evaluation doesn't recognize trouble without seeing it in the search, then it is going to make mistakes. I have lots of details about what kinds of things DB's eval had for king safety. I can tell from the speed, that DJ doesn't do the same things at all. Nor do I. Nor does Hiarcs. Or any other program, fast or slow. There are limits to what can be done within software without crushing the speed to the point where the program plays great positional chess but falls for elementary tactical traps. > >I found that my Junior5.9 could find 9.Bh6 to trade the bishops instead of 9.Bg5 >and with this move I suspect that piket can have problems with the king and not >Junior That isn't a tactical issue (depth) at all. It is a positional judgement issue. Different versions of Junior might play Bh6 in zero time. Other programs might or might not do the same. There is _no_ tactics in the Bh6 move that a program will see. It is all about the sense of the weakness of the squares around the black king with the g7 bishop gone. > >I also found that in the game against kramnik Junior5.9 can start to see >problems 4 plies after the mistake Kh8. >3 out of these 4 moves are capture so I guess that with 200M nps Junior has good >chances to avoid Kh8. > >Uri
This page took 0.01 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.