What is RLHF?
RLHF stands for Reinforcement Learning from Human Feedback.
A training method where people compare or rate model answers, and those judgments teach the model which answers to prefer. Most AI trainer work on rating and ranking tasks feeds RLHF.
RLHF is how most chat models learn to give answers people actually find useful. The model writes several answers to the same prompt, people compare them, and those comparisons train a reward model that scores new answers. The main model is then tuned to produce answers that score well.
What the work looks like
As an AI trainer on an RLHF project you usually see one prompt and two to four answers. You pick the best one, rate each on a scale for accuracy, helpfulness and safety, and write a short reason. Expert projects ask you to judge answers in your own field, such as a contract clause, a dosage question or a proof, where a generalist would miss the error.
Why it matters for pay
RLHF needs judgment a model cannot supply for itself, which is why domain experts are paid well above general labeling rates. Platforms such as Mercor, micro1 and Turing hire for it under titles like AI trainer, evaluator or reviewer.
Example
Two answers explain capital gains tax. One is fluent but uses last year's threshold; the other is plainer and correct. In RLHF the correct one wins, and your reason names the outdated figure.
Looking for this kind of work? AI Safety & Evaluation roles
More terms: how models are trained
Full glossaryReady to look at roles?
Every listing shows its pay and who it is open to.