Nora Belrose comments on Interpreting Neural Networks through the Polytope Lens