Evan R. Murphy comments on Pretraining Language Models with Human Preferences