This article distinguishes between citations and provenance in retrieval-augmented generation (RAG) systems. While citations confirm that a source supports an answer, provenance tracks the entire lifecycle of how that answer was generated, including the specific source version, retrieval event, and subsequent modifications. The author proposes using the W3C PROV data model's distinctions between entities, activities, and agents to better record these relationships, emphasizing that a citation alone does not guarantee a causal explanation or a complete history of the generation process. AI
IMPACT Highlights the need for robust provenance tracking in RAG systems to ensure reliable and traceable AI-generated information.
RANK_REASON Article discusses concepts and proposals for RAG systems, not a specific release or event.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →