Browse Teach Curate

Self-Refinement in Large Language Models

Post-hoc self-correction methods (e.g. Self-Refine) in which a language model iteratively generates output, produces feedback on its own output, and revises it — and the Degeneration-of-Thought limitation when a single agent grades itself.

jeremy Video

Self Refinement and Self Correction Methods in Large Language Models

Self-refinement and self-correction in large language models constitute iterative mechanisms where a model generates an initial sequence, evaluates it against explicit or implicit criteria (critique)…