Complexity theory and limits of machine learning

Thought this was pretty fascinating. Apparently we can apply lower bound limits to very general classes of machine learning approaches.