Roman Leventov comments on The alignment problem from a deep learning perspective