Large language models (LLMs) like ChatGPT have become increasingly popular for answering questions online. But how reliable are they, especially when it comes to complex fields like biomedicine? A new research paper explores how to make these AI-powered answers more trustworthy by teaching LLMs to provide references. The challenge with current LLMs is that while they can generate convincing-sounding answers, they sometimes fabricate information or provide inaccurate references. This is particularly concerning in biomedicine, where factual accuracy is paramount. The researchers developed a system called retrieval-augmented generation (RAG) to tackle this issue. RAG combines the language skills of an LLM with a specialized search engine focused on biomedical literature. When asked a question, the system first retrieves relevant abstracts from the PubMed database. Then, it uses a fine-tuned LLM to generate an answer based on these abstracts, including references for each statement. This allows users to verify the information directly. The results are promising. The researchers' retrieval system is significantly more accurate than a standard PubMed search, and their fine-tuned LLM performs comparably to GPT-4 Turbo in referencing relevant abstracts. While some inaccuracies in generating reference IDs still exist, the researchers are working on improvements. This research is a significant step toward building more reliable, transparent AI systems for answering complex biomedical questions. By providing verifiable references, it allows users to trust the information and make informed decisions about their health. The next step will be scaling this system to handle a broader range of medical questions, while continually refining its accuracy and reliability.