Retrieval-Augmented Large Language Models for Evidence-Informed Guidance on Cannabidiol Use in Older Adults

A team of researchers from the University of Toronto and KITE Research Institute has published a novel framework that uses retrieval-augmented generation (RAG) to provide safer, evidence-based guidance on cannabidiol (CBD) use for older adults. The system addresses a critical gap where stigma and limited health literacy often prevent proper understanding of CBD dosing, titration, and drug interactions. By combining structured prompt engineering with a curated database of CBD evidence, the framework generates context-aware guidance tailored to individual factors like symptoms, cognitive status, medications, and caregiver support.

The researchers tested multiple state-of-the-art models, including standalone LLMs and their novel ensemble retrieval architecture that integrates multiple retrieval systems. They created 64 diverse user scenarios and employed three automated, annotation-free evaluation strategies to benchmark performance. Across all metrics, the RAG models—particularly the ensemble approach—consistently produced more cautious and guideline-aligned recommendations than standalone models like GPT-4 or Claude 3. The study also introduces a reproducible evaluation framework for AI tools in sensitive health contexts, moving beyond simple accuracy metrics to assess safety and reliability in real-world applications.