Charlie Steiner comments on No convincing evidence for gradient descent in activation space