1 paper across 1 session
This paper proves that self-verification prevents model collapse in recursive training without relying on real data.
