Mid-Training with Self-Generated Data Improves Reinforcement Learning in Language Models Paper • 2605.08472 • Published 14 days ago • 3
