Author: blass uri
Date: 04:03:20 01/14/99
Go up one level in this thread
On January 13, 1999 at 23:47:12, Don Dailey wrote: >On January 13, 1999 at 17:44:16, Fernando Villegas wrote: > >>On January 13, 1999 at 14:43:53, Don Dailey wrote: >> >>>On January 13, 1999 at 05:17:22, Will Singleton wrote: >>> >>>>On January 12, 1999 at 17:04:10, KarinsDad wrote: >>>> >>>>>>Don, let me know when your individually created chess program canconsistently >>>>>beat Deep Thought and I'll take back my words. >>>> >>>>Hmm, last time I checked, I didn't see anything about hardware design in the >>>>CilkChess description. Has Don started burning proms? >>>> >>>>Couple of more points: a theoretical discussion is more interesting when the >>>>idea under discussion has some probability of occurring. For example, what will >>>>happen if a chess set is discovered on another planet? While this has very >>>>little chance of happening, it has a greater chance than does this Microsoft >>>>chess business. >>>> >>>>However, the question of team vs individual, leaving Microsoft out of it, is >>>>pretty relevant for chess programming. Progress will almost always occur at a >>>>faster rate when more than one person is working on a project, for several >>>>reasons. For chess *engine* programming, the benefit probably tops out at two >>>>people. Numerous examples abound from the literature. >>>> >>>>Don, haven't you always had a collaborator or two on your projects? CilkChess, >>>>Socrates and Tech? Didn't you benefit from some of these colleagues? >>>> >>>>Will >>> >>>Absolutely. We have a guy working on the evaluation who is a master >>>and will do a better job than I could by myself. We have also had a >>>number of people finding speedups in the code that I missed. We have >>>3 gui interfaces and have a couple guys experimenting with Temporal >>>Difference Learning. Aske Plaat has also contributed by improving >>>our implementation of mtd(f). >>> >>>There is also another team of people who built the Cilk language. >>>Cilk was actually built around the chess program, not the other >>>way around so I would have to include the whole cilk development >>>team too. >>> >>>Most of these team members are not chess experts, but will go on >>>to be among the best in the world at whatever they do. But the >>>bottleneck of the team is me. I get to spend very little time on >>>Cilkchess and most of this time ends up being to organize these guys >>>which is the best use of the time I have. >>> >>>But Cilkchess is a poor example since I am not a good example of a >>>"complete chess programmer." I taught myself programming, was >>>never better than a 1900 player in chess and just absorbed as >>>many ideas as I could from others and tried to be logical and >>>rational. A more complete chess programmer, would benefit less >>>than I would from having a team of experts at his disposal >>>because he would have less knowledge gaps (or expertise gaps) to >>>overcome. >> >> >> >>Wait, wait, wait... this is a double edge reasonning. The higher an expertise, >>the best he can grasp new ideas, nuances, etc. True: ABC will not be useful for >>him, but why we must suppose that ALWAYS these other guys will just thinks about >>the ABC? >>Fernando >> >> >>> >>>It would be interesting for me to know how big the core teams of >>>the top programs are. I'll bet that most of them have limited >>>consultations other than a gui guy and some of them do their own >>>gui's too. It seems that many teams also have someone doing the >>>book which I'll admit can be a big help. >>> >>> >>>- Don > >I don't understand what you are asking. What do you mean by ABC? > >My core idea is that the more complete the expert, the less he >will need help from others to do the thing he is best at. This >seems like an obvious principle to me. > >If I had the very best book builder in the world I would never >ask him to only write part of my chess book, I would want him >to write the whole thing! If the chess programmer himself was >the best chess book builder too, then you just wouldn't get a >whole lot of benefit from getting a less competent book builder >involved would you? I'm better, but you go ahead and do it he >would say. The final product wouldn't be better just because >it had a greater number of people working on it would it? > >And that is all I'm really saying. As complex as a chess program >is, it's really a tiny project in the whole universe of possible >programming projects. Compared to an OS like Windows, it would >have to be considered even trivial. I think it's possible for a >single person to master the art of writing a chess program and >I think some chess programmers are complete in this sense. You >could provide them with a lot of extra help, but it would be >like me trying to advise Kasparov on which move he should play, >it's not likely I could contribute very much. > >I'll say this again. If I could hire a team of the top guys >in computer chess and motivate them to cooperate on the killer >chess program, I'm not sure at all that anything really great >would come out of it. Probably a very good program would come >out, probably not much better if any than the best of the bunch. >And I really believe this. Most of the decisions about what to >do are based on the individual philosophies of each programmer. >For instance, Fritz and Rebel are not very much alike. Is this >good or bad from the cooperative point of view? You might >argue that it's good because of the diversity but if they were to >cooperate, who decides which data structure you would use? >Perhaps the data structure Franz uses lends itself to speed and >the data structure Ed uses makes it easier to program more >knowledge? If you choose either one, not only do you defeat >the philosophy of the other, but what have you gained? If you >choose Ed's data structure then in what way did this cooperation >produce something better than what Ed already had? And why did >you choose Ed's, maybe Franz was the better starting point? I believe that if everyone understand the idea of the other they can come with a third idea that is better. A good example is the Rebel-Tiger project I read at Ed's site that christophe and Ed work on the Rebel-Tiger project(They are different program and they did Rebel 2-3 times faster by their ideas. Crysthophe gave ed an idea of a selection algoritham to do Rebel faster and ed improved this idea. They expect at least 100-150 elo improvement in a few years from this project They think to use their ideas to improve tiger and Rebel and to do a Rebel-Tiger system that select in every position which program to use Uri
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.