A post sharing a video describes the Jev creator’s argument that reinforcement learning from human feedback (RLHF) is “guaranteed to disappoint” because it tells people what they want to hear. The post contrasts the ambition to automate everything with models being tuned to optimize for human feedback.