RLHF (Reinforcement Learning from Human Feedback)
A training step where humans rate model outputs, and the model learns to prefer the answers people found helpful.
It is a big reason modern assistants feel polite and useful rather than raw and erratic.
Frequently asked questions
How do AI companies teach a model what a good answer looks like?
One key method is RLHF, where humans rate the model's outputs and it learns to prefer the answers people found helpful. That human feedback is a big reason modern assistants feel useful rather than raw and erratic.
What does the human feedback part of RLHF involve?
People compare and rate model responses, and those judgments become the reward signal the model learns from. In effect, human preferences get baked into which kinds of answers the model favors.
Is RLHF the same as regular reinforcement learning?
RLHF is a specific flavor of reinforcement learning where the reward comes from human ratings rather than an automatic score. It is a core piece of alignment, the broader effort to make a model's behavior match what people actually want.
New to all this? Start with what an AI agent really is, browse the full glossary, or explore the learning hub.
← Back to the glossary