Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: deep blue elo

Author: Robert Hyatt

Date: 06:49:53 10/17/99

Go up one level in this thread


On October 17, 1999 at 04:13:42, blass uri wrote:

>On October 16, 1999 at 22:52:39, Robert Hyatt wrote:
>
>>On October 16, 1999 at 18:35:11, Amir Ban wrote:
>>
>>>On October 16, 1999 at 17:16:45, Robert Hyatt wrote:
>>>
>>>>On October 16, 1999 at 15:43:59, Amir Ban wrote:
>>>>
>>>>>On October 16, 1999 at 01:29:04, Robert Hyatt wrote:
>>>>>
>>>>>>On October 15, 1999 at 23:55:47, walter irvin wrote:
>>>>>>
>>>>>>>does anyone care to speculate what elo the deep blue that defeated kasparov was
>>>>>>>playing at . i know it was too few games to pin down an exact elo . does anyone
>>>>>>>believe a micro program like fritz ect could win 1 in 5 games vs db ?? if db was
>>>>>>>on icc could it be beaten at blitz by ANY of the players there ??? i'm just
>>>>>>>curious as to what others think about this .
>>>>>>
>>>>>>
>>>>>>I would speculate, and it is _real_ speculation, that it was in the 2750-2850
>>>>>>range.  Based on lots of things including deep thought performing at 2600 over
>>>>>>25 games to get the Fredkin 2 prize, plus beating kasparov.  Whether it is
>>>>>>stronger than Kasparov or not is a good question.  It is clearly close enough
>>>>>>to worry about.
>>>>>>
>>>>>
>>>>>You were told here only a week that this was for 2500 USCF (i.e. about ELO
>>>>>2400), and you acknowledged this information. How did you manage to forget it so
>>>>>fast ?
>>>>
>>>>the requirement was to exceed 2500 USCF.  they hit right at 2600 USCF.  I
>>>>didn't acknowledge anything different.  And _nothing_ I know of says that
>>>>USCF = FIDE+100.  In fact, Ken Sloan (in the CIS department) did a detailed
>>>>study and found that above 2400 or so the ratings are far closer than that
>>>>
>>>>He published that in r.g.c.c about 2 years ago.  It hasn't changed that I
>>>>know of.
>>>>
>>>>
>>>>
>>>>>
>>>>>This is not the first time that when corrected on a piece of Deep Blue
>>>>>information, you acknowledged mistake, then immediately went back to repeating
>>>>>it.
>>>>
>>>>This is not the first time you make an error in a statement, then refer to
>>>>the original error as fact.  DT had a rating of about 2600 USCF over 25
>>>>consecutive games.  You can find the exact details in the JICCA announcement
>>>>where they were awarded the prize.
>>>>
>>>
>>>Nice try. It was in fact USCF 2551.
>>
>>ok...  what does that change in my post you jumped in on?  I was over on DT's
>>rating by 49 points.  Has little effect on my reasoning for DB's rating...
>
>I think that today humans are better in playing against computers so in order
>to learn about deep thought's ability I prefer to look at games against
>computers.
>
>I found that Deep thought lost against mephisto in 1989
>and I quess that cray blitz (without bugs) has a good chance to beat
>mephisto of 1989 100:0(I understood that it had 100% score against Genius1 and
>Genius1 had chances only when Cray blitz was significantly slowed down).


Anybody can lose 1 game.  DT lost 1 game vs fritz in 1995.  That and the
loss to mephisto are the _only_ two games that DT lost vs computers over a
15 year span... which is amazing.



>I also doubt if top programs are going to lose 1 game out of 100 against
>mephisto of 1989.


I don't follow.  You are comparing today's programs vs DT of 1989?  In any
case I would bet that they would lose 1 of 100.  I have seen more than one
reported loss of today's programs to things like the old Fidelity Mach III.
No current program will lose many games, but all will lose some.  A 2200
program isn't going to get murdered by a 2500 program.  Losing 1 of 10 would
not be a surprise.



>
>I know also that deep thought lost to Fritz3(p90) and drew with wchess(p90).
>of course deep thought won most of the games against computers but
>based on the games that I saw I was not impressed by deep thought's ability.
>
>I see stupid mistakes of deep thought against mephisto in the game that it lost.
>Deep thought against mephisto let mephisto to trade to a simple won pawn
>endgame.
>
>I tried some commercial programs and they play better.
>They see a big drop in the evaluation after deep thought's last mistake.
>
>Uri


I suspect it would see far more big drops in eval after _their_ mistakes??



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.