Author: Robert Hyatt
Date: 07:56:49 02/19/02
Go up one level in this thread
On February 19, 2002 at 01:30:42, Uri Blass wrote: >On February 18, 2002 at 23:02:49, Robert Hyatt wrote: > >>On February 18, 2002 at 15:33:14, Uri Blass wrote: >> >>>On February 18, 2002 at 14:24:18, Robert Hyatt wrote: >>> >>>>On February 18, 2002 at 12:47:19, Uri Blass wrote: >>>> >>>>>On February 18, 2002 at 08:57:47, David Dory wrote: >>>>> >>>>>>On February 18, 2002 at 06:35:53, Uri Blass wrote: >>>>>>> >>>>>>>I do not think that all of the programmers of today are stupid. >>>>>>>I guess that they found better ideas. >>>>>>> >>>>>>>Uri >>>>>>> >>>>>> >>>>>>Better ideas? That's a relative thing, and we have no way to really compare >>>>>>their ideas (on DB's hardware and software), with ideas used in Rebel, Fritz, >>>>>>CM, ChessTiger, etc., on a PC, except in a very artificial and abstract way. >>>>>> >>>>>>I believe in each case, the programmers found ideas that were APPROPRIATE for >>>>>>their system. After all, the GREAT idea's of today, would have been disastrous >>>>>>to implement on a Fidelity Chess Challenger running with a Zilog Z80 CPU at a >>>>>>BLINDING 4Mhz. >>>>>> >>>>>>The reasons DB would still be kicking ass today, were it still around and being >>>>>>updated, are: >>>>>> >>>>>> 1) Hsu and his team had a history of creating a fantastic chess computer, DT. >>>>>> For all intents and purposes, they really had a doctorate in chess computer >>>>>> science! >>>>>> >>>>>> 2) They used the considerable resources IBM gave them, not just for software >>>>>> improvements, but to build a bunch of custom high speed micro-chips and >>>>>> integrate them into the fastest chess computer of all time. (so far :-)) >>>>>> >>>>>> 3) After a long time working out the bugs, they brought in GM Joel Benjamin >>>>>> to fine tune the openings, etc. >>>>>> >>>>>>How many other developer's do this, to this extent, Uri? >>>>>> >>>>>>It isn't just that Hsu & team were brilliant, or had a TON of resources, or had >>>>>>such sensational experience building a custom chess computer. It was all these >>>>>>things together, and I believe the whole was equal to more than the sum of the >>>>>>parts, which were considerable, in this case. >>>>>> >>>>>>If you had several million to invest in a new fantastic chess playing computer, >>>>>>wouldn't you do what the DB team did? >>>>>> >>>>>>Dave >>>>> >>>>>No >>>>> >>>>>I believe in pruning rules and I would use the money to find better pruning >>>>>rules. >>>>> >>>>> >>>>> >>>>>Uri >>>> >>>> >>>>Pruning rules introduce error. As you go deeper, those errors are >>>>summed. The DB guys instead chose to design hardware to let them go >>>>deeper with _zero_ error in the software search, and some unknown level >>>>of error in the hardware search due to whatever kind of pruning they chose >>>>to implement there... >>> >>>I believe that the right pruning rules practically almost do not introduce >>>errors and searching not deep enough produce more errors. >>> >>>See the mistake of deeper blue in game 2 when it could not see the draw. >>> >>>Uri >> >> >>No program will see that draw during the next 10 years. We don't have >>_any_ program that can search to 60+ plies down every critical path. > > >We do not need to search 60 plies in every path but only in the important lines >and it is doable. > >I believe that if you give the top programs of today 24 hours they will see >close to 0.00 evaluation. Test your hypothesis. Others already have and after a _week_ of searching they found exactly _nothing_. 60 plies is too deep for something that is not narrow and forced. And even for something forced, 60 plies is very tough to touch except for what are often called "straightjacket" positions where both sides have one legal move for an extended period of time. You aren't going to find _this_ draw in 24 hours from the original position where they moved their king to the wrong square. I don't care what kind of hardware you have. > >No way to see a forced draw because white can escape the forced draw by a line >when programs evaluate it as a small positional advantage for white. > >> >>Forward pruning, by its very definition, is error-prone. No way around >>it. > >I believe that if the forward pruning is done correctly then it is not >error-prone Sorry, but I believe that is nonsense. Otherwise you are saying that it is possible to do a perfect chess player with _no_ search at all, because the pruning rules are not based on search results. In short, a 1 ply search should be able to play perfect chess with a pruning rule that never makes a mistake. It won't _ever_ happen. >You will need a lot of testing in order to correct cases when the forward >pruning generates errors but I expect after analyzing millions of positions and >correcting the mistakes in the pruning that caused errors to have forward >pruning with no errors or maybe only one error of practical demage in 1000 >games. I expect you will end up with billions of lines of code, and if you think anyone can write billions of lines of code with no errors, then why can't we write thousands of lines today with no errors? > >one simple example: >If you prune only rook promotions and bishops promotions when the number of >pieces of the opponent is big enough to avoid stalemate danger you will have >practically no errors. > >Uri And I'll bet someone can create a position where this fails miserably. And that one error is all it takes...
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.