Author: Vincent Diepeveen
Date: 16:41:08 04/25/01
Go up one level in this thread
On April 25, 2001 at 19:22:56, Vincent Diepeveen wrote: >On April 25, 2001 at 13:30:03, Robert Hyatt wrote: > >>On April 25, 2001 at 12:19:01, Bruce Moreland wrote: >> >>>On April 24, 2001 at 23:58:11, Robert Hyatt wrote: >>> >>>>On April 24, 2001 at 22:49:32, Bruce Moreland wrote: >>>> >>>>>On April 24, 2001 at 14:05:07, Robert Hyatt wrote: >>>>> >>>>>>I think the other question is the issue... can GT withstand DB? Remember that >>>>>>in the original games played against commercial programs in 1997, DB was >>>>>>crushing them all. And it was crushing them all by playing kingside attacks. >>>>>>That was Hsu's main criticism of Genius and Rebel.. they were totally oblivious >>>>>>to the coming storm, until it hit. The games were not endgames. They were >>>>>>middlegame crushes. GT seems to be no better than anybody else at seeing >>>>>>attacks against itself. In fact, it often self-attacks by being too aggressive. >>>>>>I think against DB that would be suicide. >>>>> >>>>>I think I missed seeing the PGN's of these games, Bob. I don't doubt that there >>>>>were some games played by something against something, but I don't think that a >>>>>third-hand evaluation of these matches is worth anything, and I think that the >>>>>best thing to do would be to forget that they were ever played. Anything said >>>>>about those games is not just hearsay, it's hearsay about hearsay. >>>>> >>>>>bruce >>>> >>>> >>>>I agree. But sometimes "hearsay" is all there is. And _many_ people heard the >>>>hearsay details from Hsu or Campbell at talks they gave. They both reported >>>>that the major problem was micro king safety. Which is not a surprise at all, >>>>IMHO. I had watched the micros get rolled by attackers over and over and >>>>over on ICC... >>>> >>>>Suggesting that GT would do better than others is a speculative discussion. I >>>>have watched a lot of GT games as well, and while it does play very >>>>aggressively, it does so in many cases at the expense of its own king safety. >>>>Against an opponent that has good king safety itself, this can be dangerous, >>>>which was my point. >>> >>>My point is that it's impossible to make any claims about any of this without >>>making people extremely angry. The reason they get angry is that there is no >>>way to refute conclusions about something that was only seen in private, and >>>there isn't even any way to publicly question the conclusions, because they >>>aren't even your conclusions. >> >>Note that I generally don't "make claims" about DB. I do quite often >>respond to outrageous claims made by _others_. If you think about it, Hsu >>and SMK are in the same "boat". Both are very good at what they do. Both >>get little credit from being good, except for the people that will remove >>their blinders and take an honest look at what each accomplished. >> >> >> >> >>> >>>This isn't the way things are supposed to be. I refuse to sit in the shadow of >>>that thing for the rest of my life. I think the thing to do is grant it the >>>accomplishments it actually achieved, but refuse to extrapolate by giving it the >>>world championship by acclaim from 1999-2099. We never got to test the thing's >>>capabilities. That's not our fault, it's IBM's fault. >>> >>>bruce >> >> >>I don't sit in its shadow either. We are slowly "catching up" although that is >>possible _only_ because they are standing still and we are not. We will catch >>it one day. Absolutely no doubt about that. _unless_ IBM decides to make a >>DB 3 for another match... then we would be 10+ years behind them once again. >> >>I don't think the 2099 number you gave is reasonable. But until the latter part >>of the current decade, I doubt there will be a machine comparable to them... >>After 2010, I think DB can be relegated to "old news". But at present, it is >>still interesting, because we _do_ see pieces of it from time to time. IE DB >>Jr was on ICC during the Kasparov/Kramnik match, giving analysis (Murray had >>it on). > >Yes that was very cool from Murray to analyse with it on CCC. Note he >reported it got to 12 ply after some minutes and the scores it gave >were always moves behind DIEP seeing the same score + PV. I need to add to that that it was quite logical that DIEP saw more as DBJR there because i simply searched deeper at a dual 800PIII and i had more extensions probably as i'm not limited to what DB processors are limited to. I also do dangerous extensions in the root and i got like 13 to 15 ply in the positions where the DBJR score was reported. I see DB as a machine which design when it was proposed (around 1994?) was going to beat seemingly everything, in 1997 it searched fullwidth at a depth where only progs got to in those days WITH big forward pruning, not even counting nullmove. But then nullmove overtook them very rapidly when also machines got rapidly faster. As you can't fight against a better branching factor! If he was to design a machine now i don't doubt that with nullmove it would reach unimaginable depths with 1000 billion nodes if he somehow would manage to implement it in hardware (would be a single chip machine then though). It's questionable whether a program which doesn't know anything from bad bishops and doubled pawns, nor actively plays in opening, whether such a program has a chance against todays progs which are very optimal optimized. IMHO evaluation is more important now. When DB started to get designed, yes even half a year later in the WMCCC1997 paris, i still played games with DIEP getting 8 plies of search at a PII300. So i lost games because of that stupid depth as 8 plies loses tactical. Later World champs, tournaments i never again saw tactical problems, except if the programs were so stupid to play into an anti-positional line where both played one anti positional move after each other, putting pieces on dangerous posts, thereby creating tactics. Yet despite all that i never felt i was losing because i got outsearched or because of tactics. In wmccc1997 i clearly had that feeling though. Now an extry ply is of course not bad, because it gives with a good eval of course security. Especially if my eval is better as that of my opponent and i get to the same depth i win with induction of course, at auto232 player i'm used to get outsearched between 2 and 4 ply on average. Diep getting 10-12 ply there. Tiger starts with 14 usually pretty soon on a K7-900, then it goes further 15 , 16 etc. Idem Deep Fritz (single cpu). Depths of 17 ply not uncommon for it in middlegame. Yet i prefer my 15 times slower searching prog above that :) I do not doubt that if one designs with todays algorithms and ideas a program in hardware, that one might get huge search depths. Yet to win games you need nowadays more as just a tactical strong program. If we see chiptest (8 ply with 500k nps) then deep thought 1, dt2 and then DB1 and DB2, i get the impression that the machine definitely improved bigtime each version, but that software progs got a lead to it when the new generation of books entered computerchess. At paderborn 99 i'm pretty sure that DB would not have had a chance to win the title! Designing a machine now would definitely of course boost a commercial program, but i feel that search depth is no longer the weakest link of any program. Getting a few ply more for free is of course cool if you can get it. If SSDF gets itself a modern CPU instead of the 1997 cpu they use now i feel that if they would ship a DB version to that list, that it would face even more problems. Learning, mean auto232 players, result collecting problem etcetera. Computerchess world has become very hard for a machine which just focuses at nodes a second rate. Note that in paderborn99 also the 512 cpu zugzwang was getting completely outsearched by fritz when talking about search depth. And that though zugzwang is known to forward prune quite a lot with its dubious FHR pruning. At depths above 10 ply definitely harder work is needed as in the past. My hope is that next year despite all that Zugzwang will participate despite that Feldmann announced its retirement. Otherwise we might not see a single university program participate except hopefully crafty. Best regards, Vincent
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.