Marius Hobbhahn comments on We need a Science of Evals