Researchers have introduced ProvenAI, a framework designed to enhance transparency in retrieval-augmented question-answering systems. This framework measures transparency across three distinct layers: answer correctness, citation fidelity, and the influence of cited sources on the generated output. In experiments using the HotpotQA benchmark, ProvenAI achieved 53.53% answer accuracy and a 71.55% citation-fidelity score, revealing a 'citation-influence gap' where cited sources did not always significantly shape the answers. AI
IMPACT This framework could lead to more trustworthy AI systems by providing measurable transparency in how answers are generated and cited.
RANK_REASON The cluster contains a research paper detailing a new framework for AI transparency.
Read on arXiv cs.IR (Information Retrieval) →
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →
