Reinforcement learning improves reasoning accuracy without requiring a specific wait marker, which originates from training ...