ConceptAI & ML
Rejection fine-tuning
A method to suppress reward hacking by sampling responses, detecting and rejecting hacking responses, and fine-tuning the model only on non-hacking responses.
Know first
Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.
Lecturers
1Connections
3The lectures that define it or assume it, and the concepts next to it.