Tweet by bballinger
December 10, 2025
Proof small teams can win in AI: We trained a health foundation model on 3M person-days of data, comparable in scale to the larger labs. It detects high blood pressure with 87% accuracy, alongside atrial flutter (70%), ME/CFS (81%), and sick sinus syndrome (87%). The model was inspired by Yann LeCun's JEPA architecture, but adapted to multivariate irregular timeseries. We called it JETS: Joint Embedding for Time Series. The input is 63 channels of sensor data: heart rate, oxygen saturation, sleep stages, VO2Max, etc. The model was tested on both diagnosis and biomarker prediction tasks. The “AI revolution” won’t be limited to chatbots, images, or code assistants. We think physiological ground truth is the next frontier for health superintelligence.
- Author
- bballinger
- Date
- December 10, 2025
- Canonical URL
- /tweets/bballinger-1998786141395333445-50b070