And for T1D specifically it would be very dangerous to put an LLM in control of something like an insulin pump considering that we don't have any way to do quality assurance to the level required. Simple, deterministic algorithms are preferred for safety critical systems because they're less likely to fail in bizarre or unpredictable ways.
E.g.
resistant hypertension many medications
Which led me to a couple of research articles about pheochromocytomas and a Washington Post article about someone with high blood pressure on a "bucketload of medications".
e.g. https://www.washingtonpost.com/health/2022/04/23/high-blood-...
If the LLM can provide a sample Google-like analysis, data points that my 7 specialty doctors missed, that's an improvement.