I did get the point and I'm commenting that the point is missing the point. There is nothing new in learning that a large neural net can approximate the output

I did get the point and I'm commenting that the point is missing the point. There is nothing new in learning that a large neural net can approximate the output of a classical system. This has been done many times before. The real point is that DeepMind build a system that is half-search and pretend it's no-search. You cannot get the "high-quality data" without a classical system- not in chess.