Training language models to follow instructions with human feedback
Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida et al. · NeurIPS · 2022
Abstract
Documents the three-stage recipe that turned a raw language model into something that follows instructions: collect demonstrations and fine-tune on them, have humans rank competing outputs to train a reward model, then optimise the policy against that reward with reinforcement learning. Human raters preferred a 1.3-billion-parameter aligned model over the 175-billion-parameter base one.
Why it matters
The single most consequential result on this shelf for how AI products actually feel. It established that instruction-following is a separate axis from scale — and made the case that a smaller aligned model can beat a much larger unaligned one.
https://arxiv.org/abs/2203.02155