Buck comments on Polysemanticity and Capacity in Neural Networks