On: The truth about deep learning

[Xcelerate]

> Neural nets are universal function approximators.

True, but that's kind of like saying Rule 110 is capable of universal computation — possible in theory, but not really useful in practice.