Author: Robert Hyatt
Date: 21:14:26 01/21/00
Go up one level in this thread
On January 21, 2000 at 23:39:00, Albert Silver wrote: >On January 21, 2000 at 22:59:32, Robert Hyatt wrote: > >>On January 21, 2000 at 18:07:45, Albert Silver wrote: >> >>>On January 21, 2000 at 17:28:08, Amir Ban wrote: >>> >>>>On January 21, 2000 at 10:50:16, Albert Silver wrote: >>>> >>>>>On January 21, 2000 at 09:51:26, Tom Kerrigan wrote: >>>>> >>>>>>On January 21, 2000 at 09:33:22, Robert Hyatt wrote: >>>>>>>I don't think there is any doubt. But it will likely be at _least_ another >>>>>>>10 years and probably longer. >>>>>> >>>>>>You said earlier that the DB team discovered glaring holes in the evaluation >>>>>>functions of PC programs. Glaring enough that a seriously retarded version of DB >>>>>>could still whomp on them. >>>>>> >>>>>>So my question is, why doesn't FHH make a PC program with this ueber-function? >>>>>>It wouldn't be much work for him, and the cost is zero. Okay, it would run >>>>>>significantly slower in software than it does in hardware, but if the function >>>>>>is THAT much better, it would still be a win. He could throw in null move and >>>>>>probably achieve partiy. >>>>>> >>>>>>I think this is a real no-brainer, and the only reason he hasn't done it already >>>>>>is possibly because the evaluation function isn't all that it's cracked up to >>>>>>be. >>>>>> >>>>>>-Tom >>>>> >>>>>It could also be that the 'patches' for the eval function would be to taxing on >>>>>a PC system. How expensive would certain things like the x-ray effect of pieces >>>>>be? You know, lining up a rook-rook-queen battery behind pieces and pawns for >>>>>devastating effect, or pawn-bishop-queen. I once proposed this to a programmer, >>>>>suggesting values for who controlled a square through this battery effect (even >>>>>though the piece at the end would be quite a distance from the controlled >>>>>square). The idea was to speed up certain tactics this way, and the positional >>>>>understanding of the program on who had better square/space control. When I was >>>>>told this was too costly, I realized that systems that had super hardware >>>>>offered possibilities one could only dream of with PCs. I have no doubt that DB >>>>>probably had MANY such dreams implemented. >>>>> >>>> >>>>If they did they would show up in DB and DBjr games, and made a difference. If >>>>they didn't show up in the games, then they must not have been very important. >>>> >>>>Amir >>> >>>Possibly, but I have a theory, and I'm still waiting for someone to explain to >>>me I'm dead wrong. I have often read here from Hyatt that DB2 had an enormous >>>amount of knowledge in it. I have no reason to doubt this, but have been >>>confounded by some of DB's decisions, and some of what I saw seemed to >>>contradict this. I don't mean to start a discusion on this, I'm just giving my >>>feel on the matter. In other words, it seemed its knowledge or its use thereof >>>was inconsistent at times. Perhaps this was just the World Champ's play that >>>caused this, but what I believe is that perhaps the balance of all this >>>knowledge was less than ideal. It was very much rushed and I imagine the ideal >>>balance was just a little hard to reach in the same year all this was >>>implemented. I have been led to understand the knowledge was at _least_ ten >>>times what any other program has to offer. If this were true then balancing it >>>all would be that many times harder, and many problems could stay well hidden >>>for a long time. This belief has been reinforced by the fact that the final DB2 >>>chips were rushed so much at the end. >>> > >> >> >>Remember (and note I didn't know this until reading Hsu's book) that the DB2 >>chips were delivered _very close_ to the 1997 match. They didn't have time to >>do alot of testing. Joel Benjamin mentioned that in game one mobility was >>turned up way too high and they hadn't noticed. He said that Campbell >>attributed the early queen move to this. >> >>I can _easily_ imagine why a very complex eval could produce some very >>bizarre things, with the small amount of testing they did with the real >>machine... > >It would seem safe to say that even if Hsu couldnt make any hardware >improvements, he could probably seriously improve DB2 given a year with it. > > Albert Silver based on how little it was actually tested, and how conservative they were in using the new hardware, I would think that with static hardware, they could make it better every year for 5 years with no trouble at all..
This page took 0.01 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.