“Reinforcement learning techniques and post-training for specialized models have become mainstream.”