Machine learning - the unreasonable effectiveness of trillions of matrix multiplications, and backpropagation as a universal function approximator. Every time

Machine learning - the unreasonable effectiveness of trillions of matrix multiplications, and backpropagation as a universal function approximator. Every time they pull off another thing that's just "too dang silly to work", and yet it does... it really makes me smile. I took the free Machine Learning course at Stanford ways back, it was fun to get another toolkit, should I ever actually need it.