Retrieval-augmented generation (RAG) is a technique that allows LLM-based tools to answer questions by first retrieving relevant passages from a large corpus and then generating an answer based on those passages. This approach, akin to an open-book exam, contrasts with closed-book methods where models rely solely on their training data, which can lead to fabricated information, especially for newer or highly specific content. The RAG process typically involves chunking documents, embedding passages into a vector database, and using both keyword and vector search to find relevant information before feeding it to the LLM for response generation. AI
IMPACT Enhances the accuracy and reliability of LLM-based research tools by grounding responses in retrieved evidence rather than solely relying on training data.
RANK_REASON The item describes a technical method (RAG) for improving LLM performance on specific tasks (paper Q&A), including its implementation details and limitations. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
