Don't believe the AI doom hype - researchers understand and steer models more than ever before using techniques like reinforcement learning from human feedback.