RLHF
Reinforcement learning from human feedback: training a reward model from preferences, then optimising the policy against it.
By Lattice editorial · Entry last reviewed · method
What it implies
The method that made instruction-tuned assistants work, and now largely superseded in new work by DPO and GRPO because of its instability and cost. Still the reference point when you need to understand where the alternatives came from.
Related terms
- Direct preference optimisationAligning a model to human preferences without an explicit reward model, by optimising on chosen-versus-rejected pairs.
- GRPOGroup relative policy optimisation — a reinforcement-learning method that compares a group of sampled answers rather than learning a separate reward model.