It also comes with a unique net (NNUE) by Albert Silver and some helpers that is not available anywhere else. The net size is about double the standard SF 20MB size net.
it's not so much unique because it's double the Sf size, but
more because it's apparently made from Lco games,
and as we know (eg. earlier from Alfazero games) the
Lco evals and thus resulting positions, were different from SF.
In other word (although i've not bought it so i cant test it), maybe
AS made a smart (sub)project, but then 'sold' it in a horrible way...
As for SF, if you train a network based on SF games, then you
probably get an SF type of evalution, my 2 cnts.
Maybe indeed this Nnue is some paradigm change (quoting HGM)
although obviously SF (besides the 'old' eval) still has
some awfully good search features.
Time for a new Nnue, Sf guys ?
PS wehereb i still maintain that engine eval should have an extra
additional parameter, namely, 'sharpness'. Against lower rated opponents
you (or the engine) can/should increase 'sharpness' of play (any strong chess
player knows this btw), and only then we might see some unbalanced positions rising just from engine play rather from the (oldfashioned) opening books..
www.superchess.blogspot.com