## AI and rare diseases: a complex challenge The differential diagnosis of rare diseases represents a significant challenge in the medical field. Despite affecting a considerable portion of the population, their identification remains complex. Large language models (LLMs), thanks to their recall abilities, have been explored for this purpose, but with not always optimal results. ## MIMIC-RD: a new benchmark A new study has introduced MIMIC-RD, a benchmark for the differential diagnosis of rare diseases, constructed by directly mapping clinical text entities to Orphanet, a reference database. The methodology involved an initial LLM-based mining process, followed by validation from medical experts to confirm the actual presence of rare diseases. ## Limits of current LLMs The evaluation of various models on the dataset of 145 patients showed that the most advanced LLMs perform poorly in the differential diagnosis of rare diseases. This highlights a gap between the current capabilities of AI and the real clinical needs. The research suggests several future directions to improve the diagnosis of these pathologies.