Author: Eugene Nalimov
Date: 13:59:16 01/05/99
Go up one level in this thread
On January 05, 1999 at 16:14:42, KarinsDad wrote: >On January 05, 1999 at 15:50:51, Christopher R. Dorr wrote: > >> >>I agree that they would be able to do some things better, but how much >>improvement in I/O or tablebase handling is needed? Crafty and Winboard have >>been done (basically) by one person each, and it they handle I/O and EGTB's >>fine. I haven't seen too many commercial programs lately that need much >>improvement in parsing user input. >> >>I agree that a good opening book would be a benefit, and this they could >>certainly do, but could they do it better that Rebel, for example? I don't >>really see how. Jeroen Noomen has worked very hard, and produced a top-notch >>opening book; perhaps it could be improved, but by how much? How long would it >>take? >> >>Documentation? Absolutely! There a big company could shine. But will they >>improve the software itself? I really do doubt it; Crafty is already one of the >>top 10 speed players in the world on fast hardware (at least on ICC, it outrates >>almost all of the GM's at blitz - I'd take Crafty on a Quad Xeon 450 vs. anyone >>on earth in a match!). Perhaps another 10 or 20 or 50 points can be squeezed out >>of the same hardware, but I don't think some kind of 'Manhattan Project' >>approach is what it will take to accomplish that. >> >>I honestly feel that software increases in strength are coming very slowly now; >>I have a 8 or 10 year old program (Zarkoz 2.61) that can hang with the best >>commercial programs of today (it loses, but not terribly. I played it against >>Rebel 8 in a short match, and it lost something like 8-5 or 9-5). The increases >>that have happened over the last few years have been spurred more by hardware >>increases than software. This is not to disparage the programmers at all; there >>have been definite software improvements; but they are not dramatic, and they >>are more the result of having very good and advanced chess programmers (who have >>spent *years* studying this particular type of programming) like Bob Hyatt and >>Ed Schroeder really direct their talents at an extremely high level of >>sophistication. >> >>You can't toss 50 programmers in a room, and expect them to advance on this kind >>of work by sheer force of numbers, no matter how good they are as programmers. >>I've done my share of programming, and realize that the general programming >>skill of a person is often not as important as a deep and advanced understanding >>of the topic at hand. > >Your observations are correct when talking about general programmers doing >general work. However, Microsoft could hire chess programmers like Bob or Ed >(not necessarily these 2 individuals) who would make excellent team leads. They >could also purchase the source of dozens of chess programs out there. They could >also purchase chess product companies. They could also create an R&D >organization with 5 grandmasters, 5 well known chess programmers, 5 industry >expert computer programmers (like Bob), and 5 Phd AI experts whose charter is to >implement true AI into a chess program (true chess knowledge and understanding >is still considered a real weakness in the current chess program industry). > >When Bill Gates decided to get into the Internet business, he had 4.5 Billion >(that is billion with a b) dollars in cash assets in his company from which he >took 1 Billion dollars and started Internet technology throughout his product >lines. Has he been successful there in an industry that was basically just >starting? Just ask Netscape. > >> >>Yes, the interface might be made better, and yes the databases might become more >>user-friendly. perhaps the focus could be on tutorials and developing ways to >>use these programs to train better. But will the program play significantly >>better than the top programs today? I really do doubt it. >> >>Chris Dorr In Russia I worked in small company. In the United States I'm working at Microsoft. I suspect that companies that make modern chess programs are very similar to the Russian company I left (~5 very experienced and talented programmers, reasonable amount of modern computers and documentation, etc). Company specialized on the optimized compilers for specialized languages (main customer was hugee telecommunication company). From my point of view, it's not easier to write the *good* optimizing compiler than to write a *good* chess program. So I think I can compare and make reasonable conclusion. And it is very simple - Microsoft can easily beat almost any small company. It can hire people that are at least not worse than people in the small company, and give them much better support, hardware, etc. Those people will be able to use existing organization structure to help in testing, performance improvement, etc. You want examples? I worked on EGTB generator/probing code (that's now part of the Crafty) for 8 months. It's not ready yet - compressed TBs are not part of Crafty distribution. Had it be my primary job, it'll be done in less than a month, and with better quality (I ignored 50 moves rule; had I machines with more RAM available for generator, I'd properly implemented it). Peter Kappler's inline assembly stuff (that speed up Crafty by at least 15%) could been implemented not now, but at least 2 years ago. Please note that all that stuff was made not by chess programmer genius, but by ordinary good programmers. I'm absolutely sure that if you'll give Crafty source to the performance evaluating group in the MS division where I'm working, they'll significally speed it up. Murray Campbell and Hsu during theirs' talks repeatedly spoke about more than 8,000 factors in the DB evaluation function. They said that several of those factors are very important, and are parts of the general chess knowledge, but are not part of the chess programs they know about. I think that if you'll hire grandmaster (or several), and let him - with Ed, Bob, or other author of the state-of-the-art program - work on the evaluation for several months (as it was done with DB), resulting evaluation function will be much better than the original in Rebel, Crafty, or other program. Eugene
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.