A General Theoretical Paradigm to Understand Learning from Human Preferences | Dark Hacker News