Author: Robert Hyatt
Date: 07:31:20 06/19/99
Go up one level in this thread
On June 19, 1999 at 08:29:09, Vincent Diepeveen wrote: >On June 18, 1999 at 20:16:44, Robert Hyatt wrote: > >>On June 18, 1999 at 18:31:05, Roger D Davis wrote: >> >>>I found this message on the Rebel Site where the events of round six were >>>reported: >>> >>>Ed and I took the opportunity to play some 10 minute blitz games against Deep >>>Blue Junior. Amazingly Rebel crushed the IBM supermonster with 3-0! Deep Blue >>>Junior had no chance in these games, so one can have his doubts about the >>>playing strength of this machine. >>> >>>I haven't seen anyone else mention Deep Blue Junior. >>> >>>Roger >> >> >>Ed didn't tell "the rest of the story" so I will... >> >>I ask Hsu about this 'machine' a few weeks back, and here is what he told me >>about it: >> >>Some internal IBM folks asked him to develop a 'demo' facility to show off DB. >>He elected to do a web-based interface, which is "stateless" if you know what >>this is all about. In essence, this machine won't play a "game" at all, it >>simply takes a position, searches it for 1 second (which includes mostly the >>time needed to download the chess processors with the state information) and > >it sometimes takes minutes for it to come back, but it doesn't search for 1 >second, but like 15 seconds. it shows the time using the time at the ibm >server. it shows this. > >a major problem for it is the opening. it plays really silly moves. >after 1 move it seems out of book. Vincent, do you understand the concept of "stateless"? It doesn't know it is playing a _game_. It only sees a single position and computes for less than one second to produce a move... So it is more than possible that it doesn't use a book. Or it might have a book so impossibly wide that it plays all sorts of stuff... > > >secondly, it handles pawnstructure like a child, so even if it searches >like only 7 ply fullwidth, then still it's awful handling things positional, >without talking about tactics even. If it is searching way less deeply >than any other program, then this evaluation is already a major problem. > >How can someone handle positional things *that* bad? > >>es a move. No repetition testing at all, no game history, no >>nothing except for a near-instant search. However, it can take quite a while >>to make a move because _many_ web browsers get pointed at this thing by the >>IBM guys doing demos... >> >>Hsu estimated that it might play at 2200 or so. Which was all that was needed >>for the demonstrations it is used for. It is _not_ "deep blue junior" by any >>measure you would care to name. >> >>And putting such nonsense on Ed's web page is _highly_ misleading. >> >>To say the least. >> >>Bob > >it is showing at the screen: > >DEEP BLUE JUNIOR in a major font > >No question about it. > >Either IBM is lying on its screen, or perhaps it's really as bad as >i'm telling for years? read what I wrote. I specifically ask Hsu about it a while back. That is his explanation _exactly_ as to what it is. > >it's just playing with its pieces, some things it seems to understand, >like that a rook in the corner can be bad when its closed in, >see ferret - fritz of today where fritz got after some bad moves >(personally i'm still considering fritz was won after the opening) >got a rook closed in the corner. it seems this deep blue junior knows this. > >So even if it thinks for say 15 seconds, and practically a few seconds, >then still there is a lot to explain. > >Greetings, >Vincent > >Greetings, >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.