Is Anthony ... back ?!

[CarstenL]

Just saw that Anthony Cozzie has registered on CCC Forum.

[Damir Desevac]

No he is not back. It is just someone impersonating him... He had his CCC account long ago. The user who signed up with his name did it recently for the first time...

[CarstenL]

No he is not back. It is just someone impersonating him... He had his CCC account long ago. The user who signed up with his name did it recently for the first time...

Really ??

Look here: http://www.talkchess.com/forum/viewtopi ... 51&t=38440

[BB+]

At this point I am out of computer chess with no intention of returning. I think as a field it is almost solved at this point,

From the "sporting" point of view, this has bit of truth to it -- you get huge hardware, a massive opening book, and it often comes down to whether or not you get wins as opposed to draws against the bulk of the field. Cozzie can probably cite the WCCCs of 2006/7 as examples of this. Which means MichaelIsGreat is probably correct about the TCEC Elite Match being superior.

[BB+]

Well, he got lured into a response at TalkChess.

Yes, that is correct Albert, except that computer chess is in fact mostly finetuning and relatively few optimizations. This is also why I believe that computer chess is mostly a solved problem: most of what we do is exactly this fine tuning. Add a pattern here; tweak a value there. 'Oh, maybe I should only reduce 3 ply instead of 4 under case X'. You see 450 elo; I see 'engineering problem'. If you don't believe me, compare the Toga code to Fruit. It's quite a bit stronger in ELO, but with relatively few lines changed.

I admit that this perspective is rather extreme and comes from my background as PhD. For comparison, my current research involves writing programs from English. We have no idea how to do this well currently.

Think about what you say when you talk about his 'programming skill'. Remember a doubling of speed only achieves 60 elo or so. What chess engines really are is a collection of heuristics to estimate solutions to an exp-time problem. So either Vasik has some sort of special talent in selecting and evaluating such heuristics, or he has a better algorithm for selecting them. I find #2 much easier to believe, but that's just me.

Miguel: I agree with the search thing, but the space of search parameters is much smaller. You can afford to do ordinary blitz testing. But when you have 10,000 evaluation parameters you just don't have the CPU time. But hey, I've been wrong before.

P.S. I don't want to imply that Rybka's 5 year domination of computer chess was any less impressive because of this. In computer science you always want to find general solutions. The point of that section is that for most people 'strong engine -> great programmer -> no need/possibility of cloning'.

He uses "exp-time" in a sloppy manner here (one could alternatively say that chess is a O(1) problem, and thus "solved in theory"), but I can certainly can find much to agree with here. The question of "generic" versus "specific" methods in games programming has long been a divergence point for the engineering/theoretical crowds. For instance, the theoretical guys take a question like "Is alpha/beta better than proof-numbers?" and then try codify for what types of games/positions (or search trees) this is likely to be true. The engineering crowd simply asks what heuristics work for a given game [or how they might be realised from a high-level description in a specific case], with theoretical questions at best guiding the hunt for such heuristics.

See for instance this comment from Lefler about LMR in general. [Rajlich noted in that thread: "Re. LMR - these tend to work in games where sibling moves tend to be similar to each other. In chess, quiet moves tend to be similar to each other, while captures and checks tend not to be similar. It might be interesting (but not easy) to classify games by whether LMR works or not - this will be a lot less obvious than whether null move works"].