AI models often give the right answers but point to the wrong sources

In three linesLeading AI models like GPT and Gemini routinely cite text passages that don't support their answers, even when answers are correct. Researchers at Peking University term this "attribution hallucination" and introduce CiteVQA benchmark to systematically test for it.

Summary generated by Claude — human-verified