07Notes
AI Research Notes
(02)Daily breakdowns of new AI papers: what each one found, how it works, and where its claims stop.
Search by topic, paper, or author
Teach the small model the teacher's ranking, not its probabilities.
Order-Only On-Policy Distillation (O³PD)
Your fine-tuning data can make a model less safe without a single harmful example.
Follow the Gradient: Tracing Safety Degradation Back to Training Data
No notes match your search.