Yale researchers from the Laboratory for Intelligent Global Health and Humanitarian Response Technologies (LiGHT) at the Section of Biomedical Informatics and Data Science are collaborating with a team at EPFL and the International Committee of the Red Cross to build Meditron, currently the world's best-performing open-source large language model (LLM) for medicine. Meditron was recently included in the 2024 AI Index Report released by the Stanford Institute for Human-Centered Artificial Intelligence.
The index is recognized as a global leader in AI insight and has been cited by policymakers, the New York Times, and hundreds of academic publications.
LLMs, like ChatGPT, are algorithms with the power to compress complex information into a conversational interface. This is useful in medical contexts, especially in resource limited settings, according to Meditron’s developers. The student-powered collaborative teams at Yale School of Medicine and EPFL hope that Meditron will make high quality medical information more accessible.
Meditron is currently the leading open-source LLM for medicine, and performs within 1% of multi-billion dollar models like GPT-4 on tasks of medical reasoning. Meditron is adapted from Meta’s Llama 2 model and is open source, meaning that its underlying code is available for the public to evaluate and adapt. Within the first 3 months of its launch, Meditron was downloaded over 30,000 times.
“With so many start-ups using open models like Meditron as a base, venture capital and donors should recognize the value of investing in academia to create ambitiously rigorous and cost-effective development of large-scale artificial intelligence,” said Mary-Anne Hartley, MD, PhD, MPH, Assistant Professor of Biomedical Informatics and Data Science and director of LiGHT. “Academia plays an essential role in ensuring that this transformative technology is not only accessible, but openly validated in a neutral space for public good.”
The AI Index was launched in 2018.