Daniel Kokotajlo comments on New Scaling Laws for Large Language Models