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

1 Introduction

1. What problems of standard retrieval-augmented generation does Self-RAG address?
   Standard RAG retrieves a fixed number of passages for every input, whether or not retrieval helps or the passages are relevant, which can hurt versatility and quality; and the output isn’t guaranteed to agree with the retrieved passages, since models aren’t trained to follow them.
   • Asai et al., 2024, Self-RAG (The Twelfth International Conference on Learning Representations (ICLR 2024)), §1
2. What is Self-RAG in one sentence?
   A single language model trained to decide when to retrieve, to generate from retrieved passages, and to critique the passages and its own output by emitting special reflection tokens, which also make its behaviour controllable at inference time.
   • Asai et al., 2024, Self-RAG (The Twelfth International Conference on Learning Representations (ICLR 2024)), §1