RLHF is Reinforcement Learning from Human Feedback.

It usually refers to fine tuning language models using data labelled by humans.

Hugging face have a good overview in this article: https://huggingface.co/blog/rlhf