I can't find the reference. But some time ago there was a post in HN about how different subjects require different strategies. It was written by a big shot AI or Math professor. Probably someone from MIT. I can't quite remember.