JA EN
Glossary › align

GLOSSARY

align

appears in 1 paper titles

Definition

To make a model's behaviour conform to human intent, instructions, or values — the step that turns a pre-trained next-token predictor into something that follows requests, carried out by RLHF, DPO, and their relatives. The verb also has a representational sense: CLIP aligns image and text embeddings so that matching pairs land near each other. What follows the verb tells you which sense is in play.

Explainers using this term