RLHF (reinforcement learning from human feedback)

The technique behind most AI-training gig work: humans rank or score model outputs, and those preferences train the model to behave better. RLHF is why platforms hire thousands of contributors to evaluate responses — the ranking data is the product. Pay is generally per completed evaluation batch and rises with domain expertise.