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

By Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, Hannaneh Hajishirzi. International Conference on Learning Representations, 2023

2,215 citations recorded by the source corpus.

Despite their remarkable capabilities, large language models (LLMs) often produce responses containing factual inaccuracies due to their sole reliance on the parametric knowledge they encapsulate. Retrieval-Augmented Generation (RAG), an ad hoc approach that augments LMs with retrieval of relevant knowledge, decreases such issues. However, indiscriminately retrieving and incorporating a fixed number of retrieved passages, regardless of whether retrieval is necessary, or passages are relevant, diminishes LM versatility or can lead to unhelpful response generation. We introduce a new framework called Self-Reflective Retrieval-Augmented Generation (Self-RAG) that enhances an LM's quality and factuality through retrieval and self-reflection. Our framework trains a single arbitrary LM that adaptively retrieves passages on-demand, and generates and reflects on retrieved passages and its own generations using special tokens, called reflection tokens. Generating reflection tokens makes the LM controllable during the inference phase, enabling it to tailor its behavior to diverse task requirements. Experiments show that Self-RAG (7B and 13B parameters) significantly outperforms state-of-the-art LLMs and retrieval-augmented models on a diverse set of tasks. Specifically, Self-RAG outperforms ChatGPT and retrieval-augmented Llama2-chat on Open-domain QA, reasoning and fact verification tasks, and it shows significant gains in improving factuality and citation accuracy for long-form generations relative to these models.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/ddbd8fe782ac98e9c64dd98710687a962195dd9b

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity (2024, methodology): “…previous approach applies the same fixed operations to every query regardless of its complexity but also necessitates additional specific training to LMs. Concurrent to our work, Asai et al. (2024) suggested training a sophisticated model to dynamically retrieve, critique, and generate the text.”
Retrieval-Augmented Generation for AI-Generated Content: A Survey (2024, neutral): “It follows the paradigm in SKCODER, retrieves relevant pairs of query and code, then produces sketch template for real program generation.”
The Power of Noise: Redefining Retrieval for RAG Systems (2024, positive): “These sparse retrieval methods, with BM25 being its most famous current iteration [32], characterized by their use of high-dimensional and sparse feature vectors, have been essential in developing early IR systems.”
RAFT: Adapting Language Model to Domain Specific RAG (2024, background): “…NLP tasks, including language modeling (Guu et al., 2020; Borgeaud et al., 2022; Khandelwal et al., 2019; Shi et al., 2023d; Lin et al., 2023b; Shi et al., 2023c; Asai et al., 2023; Xu et al., 2023; Wang et al., 2023) and open-domain question answering (Izacard et al., 2023; Lewis et al., 2020).”
LongLLMLingua: Accelerating and Enhancing LLMs in Long Context Scenarios via Prompt Compression (2023, methodology): “…used technologies such as In-Context Learning (ICL) (Min et al., 2022; Dong et al., 2023), Retrieval Augment Generation (RAG) (Lewis et al., 2020; Asai et al., 2024), and Multi-turn Agent (Shen et al., 2024; Park et al., 2023; Wu et al., 2023a) are driving prompts to be increasingly longer, even…”
Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning (2025, positive): “…but are difficult to scale effectively due to their reliance on large-scale, high-quality annotated trajectories and the inherent non-differentiability of the search operation, which renders end-to-end gradient descent-based optimization inapplicable (Schick et al., 2023; Asai et al., 2024).”
Large Language Models for Information Retrieval: A Survey (2023, neutral): “However, due to the limited computing resources, many methods [170, 171, 179] choose to prompt LLMs for generation as they could use larger LMs in this way.”
WebThinker: Empowering Large Reasoning Models with Deep Research Capability (2025, neutral): “Unfortunately, existing open-source deep search agents typically employ retrieval-augmented generation (RAG) techniques with predefined workflows [1, 63, 43], which limits LRM’s ability to explore deeper web information and hinders close interaction between LRMs and search engines (Figure 2 (a) &…”
Retrieval-Augmented Generation for Large Language Models: A Survey (2023, methodology): “Self-RAG [Asai et al. , 2023] introduces “reflection to-kens” that allow the model to introspect its outputs.”
A Survey on Knowledge Distillation of Large Language Models (2024, methodology): “For instance, Self-Rag initiates the retrieval operation when generating the reflection token Retrieve .”
A-MEM: Agentic Memory for LLM Agents (2025, neutral): “The note construction process involves prompting the LLM with carefully designed templates P s 1 : Following the Zettelkasten principle of atomicity, each note captures a single, self-contained unit of knowledge.”
Q-TIE: A Lightweight and Generalizable Re-ranking Framework for Temporal Information Retrieval (2026, unknown): “This efficiency makes Q-TIE a cost-effective alternative for modern RAG pipelines, which often incur multiple retrieval calls for improved RAG performances (Asai et al., 2024; Yan et al., 2024).”
From Semantic Retrieval to Conversational Agent: A Web-Based RAG Architecture for Interactive System Dynamics Modeling (2026, unknown): “Modern frameworks increasingly rely on advanced conversational query clarification and iterative agentic RAG synergies [18,19], alongside the need for principled evaluation of both the generative component [20] and the interactive retrieval process itself [21].”
Invoice Haystack: Benchmarking Document Retrieval and Visual Question Answering Under Strong Visual Homogeneity (2026, unknown): “…strong few-shot performance through Fusion-in-Decoder [29], achieving 42.4% exact match on NaturalQuestions [38] with only 64 training examples (45.1% with a Wikipedia-only index), and recent advances including Self-RAG [8], and CRAG [73] introduce retrieval-on-demand and corrective mechanisms.”
Min-Cost Flow Routing for Evidence Assembly in Long Multimodal Documents (2026, unknown): “G 2 -Reader uses a large language model (LLM) planner to generate sub-questions, retrieve evidence for each, and assess whether the evidence is sufficient (Du et al., 2026), following an iterative retrieval approach (Trivedi et al., 2023; Asai et al., 2023).”
MM-BizRAG: Rethinking Multimodal Retrieval-Augmented Generation for General Purpose Enterprise Q&A (2026, unknown): “Yet these pipelines often lack standardization (Gao et al., 2025; Zhang et al., 2025) in document representation (Sarthi et al., 2024; Jin et al., 2025; Yu et al., 2024a) and ingestion design (Xiong et al., 2024; Asai et al., 2023).”
SAGE-RAI: Design Patterns for Transparent RAG Systems (2026, unknown): “While advances in self reflective [2] and active retrieval [16] improve RAG performance, Gao et al. [11] reveal that these assume technical audiences, with citation as optional.”