alignment

Definition

Making a model's behaviour conform to human intent, values, or a specified set of norms — refusing harmful requests, following the spirit of an instruction rather than gaming its letter. RLHF, constitutional methods, and preference optimization are all alignment techniques. Note a separate, unrelated technical sense of the word: in multimodal and translation work, alignment means putting two representations or sequences into correspondence.

Explainers using this term

• Paper Explained: Video Generation Models: A Survey of Post-Training and Alignment — Four Toolboxes for Making Video Models Do What You AskVideo Generation Models: A Survey of Post-Training and Alignment