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

Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, Hannaneh Hajishirzi

2023 | arXiv:2310.11511

Source: https://arxiv.org/abs/2310.11511

Reading scope: abstract-and-metadata

Editorial update: 2026-10-20

Learned reflection tokens help control retrieval and critique generation.

## Why this matters

Retrieval necessity and passage relevance become explicit decisions.

## The proposal

A fixed retrieval budget can bring irrelevant context into a response. Self-RAG trains a model to retrieve on demand and emit special tokens for reflection.

## What was inspected

The authors report improvements across several tasks, including factuality and citation evaluation. This abstract-based note does not verify benchmark setup or compare present-day model versions.

## Editorial interpretation

Treat reflection as a learned control signal. A model judging its own answer remains different from an editor checking the cited source against the claim.

## Mechanism

Learn when to retrieve.

Generate with retrieved passages.

Use reflection tokens to assess passages and output.

## Limits

Self-critique is not independent verification.

Training supervision and evaluation details need full-text review.