🧠 Food for thought

Implications, context, and why it matters.

The startup rests on FDA-cleared and Nature-published AI

• professor James Zou’s lab has built medical AI that reaches real use. Its EchoNet algorithm has FDA clearance, according to Zou’s lab website 1.
• Another project, SleepFM, is a multimodal foundation model trained on nearly 600,000 hours of sleep data from 65,000 people. It is described as able to predict future risk for more than 100 diseases 2.
• The lab also works on agentic AI, which uses multiple software agents to handle research tasks. In one virtual lab, AI agents devised nanobody designs for SARS-CoV-2, the virus that causes Covid-19. Those variants then went into a physical lab for testing. Stanford researchers said the designs were feasible in experiments and bound tightly to a SARS-CoV-2 variant 3.

Health AI is moving toward specialized models built on proprietary data

• Human Intelligence fits a broader move toward foundation models trained for healthcare. Generic models can make serious errors on specialized scientific tasks 4.
• Many startups now try to stand out with proprietary multimodal data, gathered in several forms such as sleep recordings or neural signals.
• Apple is following a similar path. It uses wearable data from the Apple Heart and Movement Study to build health AI models 5. The result is a race for unique datasets to build specialized AI that can automate diagnostics and research and development.