james.lucassen comments on Why does gradient descent always work on neural networks?