AI RESEARCH
When Retrieval Doesn't Help: A Large-Scale Study of Biomedical RAG
arXiv CS.CL
•
ArXi:2606.04127v1 Announce Type: new Medical question answering is a high-stakes setting where factual errors can have serious consequences. Retrieval-augmented generation (RAG) is widely viewed as a promising solution, and prior work has reported substantial gains for large medical QA models. We revisit this assumption across a broad range of open-weight instruction-tuned models spanning 7B to 72B parameters.