RLHF Guide: What Each Post-Training Stage Asks Human Experts to Do

RLHF is now one stage of post-training. How SFT, preference optimisation and verifiable rewards use human judgment, and how each fails when it’s weak.