Beyond Model Collapse: Scaling Up with Synthesized Data Requires Reinforcement

Open in new window