Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection

What changed

Trains a generator with retrieval, relevance, support and utility reflection tokens to retrieve on demand and rank passage-conditioned response segments.

Scientific scope

Trains a generator with retrieval, relevance, support and utility reflection tokens to retrieve on demand and rank passage-conditioned response segments. Builds on existing retrieval-augmented generation. Critic training uses GPT-4 labels before generator training; no claim of training-free self-verification or guaranteed faithful answers. Multi-passage generation and ranking incur resources.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Builds on existing retrieval-augmented generation. Critic training uses GPT-4 labels before generator training; no claim of training-free self-verification or guaranteed faithful answers. Multi-passage generation and ranking incur resources.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction