Decentralized open source AI for Rare Illness medicine at Stanford
Research to the People at Stanford is a unique patient partnered research organization that focuses on mobilizing our diverse community of scientists, clinicians, and technologists to address unaddressed issues of patient care quality for rare illnesses. To that end we are currently running a Decentralized Hackathon in partnership with Roche, Genetech, and Huggingface... where contributors from a diverse range of backgrounds are coming together to train and validate rare illness specific expert models. Once trained the models will be evaluated in a model arena style ELO system by our partners at the Ehlers Danlos Foundation, which will presents a novel approach to validating medical-grade AI. Format: 10-15 minute talk with interactive capstone. Will start by summarizing the technology behind fine tuning open source LLMs, and the collaboration problem it presents to tightly integrate machine learning process with scientific expertise. At end of talk will load two models up in model arena, and quiz them in collaboration with the audience... asking them questions and then grading their responses. Should be a fun look at the deep need for Decentralized organization in training an validating these next generation intelligence tools. š¤