Author: Robert Hyatt
Date: 06:49:42 04/23/01
Go up one level in this thread
On April 23, 2001 at 02:55:51, Uri Blass wrote: >On April 22, 2001 at 22:36:08, Robert Hyatt wrote: > >>On April 22, 2001 at 18:54:20, Vincent Diepeveen wrote: >> >>>Write it down as you want, but the play shown by >>>Deep Blue was around 2400 level. >>>The play shown by Kasparov was worse as that even, around 2300 level. >>>Yes he didn't give away a piece in any game, but that's about the only >>>good thing we can report about Kasparov. >>> >>>He even fell into a known openingstrap where i will never fall for. >>> >>>But he also lost a game in 18 moves which never happens to me >>>in slow games and i'm 2285 rated. >>> >>>Last time i lost so quick i was like 2100 rated or less. >>> >>>The games are easy to analyze. >>> >>>Seirawan has commented on the match with loads of question marks, >>>and missed wins for Kasparov. All of them giving stupid excuses. >>>Any world champion missing wins which a 2200+ isn't missing means >>>it is the worst game ever of this world champ. >>> >>>Just read the ICCA journal for analysis of Seirawan. issue june 1997. >>>to ask for this article email Jaap v/d Herik who is leading the ICCA >>>journal publications. >>> >>>The games are really bad. Very bad. >>> >>>Deep Blue didn't even know some doubled pawns are good... >> >>I See no point in arguing about DB's play or eval. _everything_ that >>you say will be nothing but speculation, and we can speculate all day with >>no resolution of any kind. >> >> >> >>> >>>Of course its search depth of between 11 to 13 ply was very good >>>considering in 1997 most programs only got that depth WITH forward >>>pruning. Programs WITH a bit of mobility never got to that >>>depth in 1997. >> >> >>That is a completely wrong statement. I have reported before, that each >>member of the DB team has confirmed that the depths reported as 11(5) means >>11 plies of software search (+ extensions of course) followed by 5 full-width >>plies of hardware search (+ extensions like in check, but not singular) followed >>by a normal quiescence search. >> >>Hsu reported this. Campbell and others confirmed it. >> >>Your 11-13 plies is simply _wrong_. >> >>As I also told you in the past, the _hardware_ does not produce a PV at all, >>it is incapable of doing this as it was designed. But behind _every_ PV you >>see in the IBM output, lies a 5-6-7 ply hardware search on top of it that you >>can't see... >> >> >> >> >>> >>>In that sense deep blue sure would have had a good shot at any program >>>in those days, when we disregard the openingsbook. >>> >>>Yet i'm pretty sure that many programs in 1997 would have beaten it. >>>In Hong Kong that was proven by Fritz. >> >>How? Fritz played a 2M NPS version of the program on the _original_ deep >>thought hardware. Hsu produced _two_ more re-designs before the final >>Kasparov match. The machine Fritz beat had _nothing_ to do with either >>the machine used in match one or the machine used in match two. >> >>I wish you would stop making the wild claim that you could beat them. Crafty >>played them 10 games in their lab and it did no better than Fritz or Rebel in >>that test. > >I agree that probably vincent could not beat them but Crafty at that time was >weaker than Crafty of today. > >Another point is that the games were not public so people have the freedom not >to believe IBM about it. > >Uri I would expect that deep blue of _today_ would be stronger as well. Even with the same chess chips, the SP has gotten significantly faster. And then again, Hsu was ready to build chess chips more than 10 times faster. Rather than a machine that averages 200M nodes per second it would be 2B nodes per second. With peaks at 10B nodes per second. Claiming I could beat the _old_ DB is a wild stretch. Because I know what 200M nodes per second will do... Claiming that their search depth was 4-6 plies less than it actually was is misleading, when the data is public. I simply want to see realism posted...
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.