A tabular foundation model built for messy, real-world business data
Emerged finds new products worth a look, daily — what’s new today
It is a rare open, side-by-side test of classical, deep and foundation models for type 1 diabetes forecasting, and it reports where they all fall short.
T1DMAI pits 26 blood glucose forecasters against each other on 115 days of the maker's own CGM, insulin and carb data, then on 279 people from the public MetaboNet set. The field runs from persistence and ARIMA through XGBoost, LSTMs and PatchTST to Chronos and TimesFM, plus a 31k-parameter transformer of the maker's own. XGBoost mostly wins; the small model catches the most lows. MIT licensed, mostly AI-written, and plainly not a medical device.
More like this
How long until a local LLM machine pays for itself against API bills
Rides, meals and chain wear in one SQLite file you host yourself
A terminal monitor that names the local models sitting in your video memory
Run trillion-parameter models on a laptop by streaming weights from NVMe
Follow the models, agents and labs you depend on and hear when something breaks