Large Language Models (LLMs) like ChatGPT, Gemini, and Claude are designed to answer questions with words. Powerful, but not built to understand your health data.
That is why a new study in Nature Medicine stood out to me. It tested something different: a Personal Health Large Language Model (PH-LLM). Unlike a general LLM, this system was trained to read the numbers streaming from your wearable such as heart rate, steps, and sleep patterns and turn them into guidance you can use.
Here is what made it different
It performed almost as well as human experts on standardized sleep and fitness exams
It translated raw wearable data into advice that made sense for daily life
It even predicted how someone felt about their sleep based only on sensor inputs
In other words, it was not just a dashboard. It was a coach teaching daily, adapting to the person, and reinforcing healthy habits.
As a cardiologist, every day in clinic, I give patients advice on sleep, exercise, and nutrition. They nod, they mean well, and they try. But once they walk out the door, real life gets in the way.
Now imagine something different. You wake up after a rough night. Instead of your wearable simply telling you “6.2 hours of sleep,” a PH- LLM could suggest shifting your workout to the evening, limiting caffeine after lunch, or adding a 20 minute nap. Small, personal nudges every day.
The bigger picture
PH-LLMs offer a vision of collaborative intelligence with doctors and AI working together to give patients the daily teaching that busy clinics cannot provide.
Because if your wearable already knows you slept badly last night, maybe tomorrow it can also help you know exactly what to do about it.
Reference:
Nature Medicine. Personal Health Large Language Models for wearable data interpretation. Published 2025. https://www.nature.com/articles/s41591-025-03888-0 Cory McLean Justin Khasentino