I looked at this explanation of the universal approximation theorem again today, and it's enormously helpful. Thanks @michael_nielsen! I need to remember to include this perspective in my intro talks.
http://neuralnetworksanddeeplearning.com/chap4.html
Thanks! I've run through that explanation in ~10 mins (actually showing the widgets on-page) in talks. At least some audience members seem to like it! (Of course, if others hate it, I wouldn't usually know.)
