simple endgame bridges have nothing in common with kingside fianchetto, done on a regular basis.Evert wrote:You can't conclude that from looking at games.Lyudmil Tsvetkov wrote: judging by R games, something I can certainly do, as the games are available, in distinction to the code, I can pretty much safely assume R does introduce quite some knowledge previous engines at the time did not have, so maybe I am more right after all, and real nps numbers are closer to outputted ones.
I do not know why everyone assumes R has more search innovations than chess knowledge ones, where is your proof?
or, most probably, as with you and a range of other members, to turn you the attack favour, you are mixing supposition/personal predilection with proof?
look at R games:
- fianchettoeing its king bishop sometimes, no other engine did that at the time, even not now
- excellent handling of connected passed pawns
- some smattering of imbalances, lacking in other engines
- more reasonable psqt for pawns and pieces alike, etc., etc.
this definitely seems to me like some knowledge base, does not it?
A well-tuned evaluation and sufficient depth will give the illusion of knowledge being present that cannot be traced to a single line of code, it becomes emergent.
Simple example: you do not need to teach a computer how to "build a bridge" in KRPKR. It will figure it out.
and a good smattering of imbalances.
you can not prefer N+R vs Q in complex positions, unless you have sufficiently strict knowledge conditions in code. search will never see that.
similarly, there is no way for search to find g3 + Bg2 in a wide range of openings, simply because there is nothing to see by search.
it is clear there was some knowledge base in R that other leading engines at the time did not possess, and also Fruit does not have.
when you look at a big number of engine games, and know the styles of the engines, it is easy to know when a move is based on evaluation, or purely on search.
simple rule: search helps only in tactically relevant positions, where there is a clear-cut advantage; search is able to see good moves there.
search can not help a lot in tactically neutral positions, with no clear edge, because there is nothing for search to see, and you have to rely on eval.
