NN learning is not search, and most NN architectures don't encode Turing machines. And gradient descent is not brute force, it's actually very efficient. So this is a lot of nonsense.

→ View original post on X — @fchollet