Author: Robert Hyatt
Date: 10:27:18 05/07/04
Go up one level in this thread
On May 07, 2004 at 11:16:30, Dan Ellwein wrote: >On May 07, 2004 at 04:19:21, Vincent Diepeveen wrote: > >>On May 07, 2004 at 01:03:20, Jouni Uski wrote: >> >>>Has You looked for that afterword? There is very short comment to Kasparov >>>vs. Junior and Kramnik vs. Fritz. Hsu still claims DB was superior to >>>Fritz/Junior in tactics! Really? >>> >>>Jouni >> >>Let me quote Bruce Moreland: "i would love to have a shot at deep blue in blitz, >>i tactically will destroy it". >> >>This was with Ferret at a 4x 400Mhz PII machine. >> >>I agree with Bruce. >> >>We must be realistic. Deep Blue needed 3 minutes to get to 10 ply in openings >>positions. In endgames it finished 12 ply a lot. Most middlegame positions >>however it searched 10 - 11 ply. > >Vincent > >if that was the case... > >how was Deep Blue able to win a match from Garry K. > >my best > >Dan >> Pick from one or more of the following: 1. It didn't happen. It was done via smoke and mirrors. 2. humans helped DB win. 3. Kasparov threw the match to play again and get another big chunk of money. 4. Kasparov played like a 2000 player. 5. add any excuse of your choice here... >>On average they claimed a search depth of 12.2 ply but this is not iterative >>depth but 'observed' depth. So the singular extension depth added to it (not >>qsearch i guess). >> >>10 ply with a singular extensions and threat extensions and mate extensions is >>in theory tactical very strong. Certainly for 1997 standards. >> >>However in hardware they cannot do any dangerous extension. Not only Hsu >>explicitly mentions it, also Chrilly has done very clear statements that >>hardware search is *that* inefficient that he had to forward prune in hardware >>in a very primitive way. Same for Deep Blue. In its 4 ply of hardware search it >>forward pruned, and *had* to of course. Both cannot use any dangerous extensions >>in hardware search. Deep Blue triggers the last one at 4 ply depth left. >> >>This where software products pick up incredible tactics last few plies. They see >>just near to shit last few plies. >> >>So you can extend a lot in mainsearch, but 10 ply - 4 = 6 ply. So within 6 ply, >>it should see everything then. >> >>Let's be clear, this in 2004 is not a realistic scenario. Hsu still lives in the >>80s. He did live there in 1997 still. His machine didn't even use nullmove which >>by 1995 had been clearly proven for every idiot on the planet as the way to go. >> >>Frans Morsch *publicly* did statements about recursive nullmove. During dinner >>every programmer has heard it, i'm sure of it. >> >>What Hsu writes is utter nonsense.
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.