Why fine-tuning is not enough and how reinforcement learning with human feedback shapes smarter models.