Self-Supervised Alignment with Mutual Information: Learning to Follow Principles without Preference Labels

Open in new window