Synthetic Data RL

Models at Google DeepMind generate their own synthetic data via reinforcement learning to extend token limits and advance capabilities without external datasets.