Retrieval-augmented generation, or RAG, integrates external data sources to reduce hallucinations and improve the response accuracy of large language models.