Large language models (LLMs) can store and recall vast quantities of medical information, but their ability to process this information in rational ways remains variable.