25 July 2024 · 1 min read
It is fascinating to see, and pretty much a full time job in itself, to explore the continuous em...
I'm tracking the continuous emergence of new AI foundation models — from Mistral Large 2 to Meta's Llama 3.1 — as a billion-dollar arms race. My prediction: three types of models will emerge, and for healthcare, specialised imaging models like Microsoft's collaboration with Mass General Brigham are the ones to watch.
Author
Last updated
6 May 2026
It is fascinating to see, and pretty much a full time job in itself, to explore the continuous emergence of new AI foundation models over time. This is a billion dollar arms race, and it will be fascinating to see who will be the winners (big tech, or challenger companies, open-source vs. closed system).
In recent news, French Mistral AI has just released its latest iteration, Large 2, which seems to be competing quite well with its US peers on the usual benchmarks, and Meta just published the latest release of Llama, v3.1.
For us in the healthcare space, it will be even more interesting to watch the emergence of specialised models, like this collaboration that Microsoft just announced with Mass General Brigham and University of Wisconsin-Madison, on foundation models for imaging.
My prediction is that we'll see three types of models emerge. The ultra large ones that only few companies will be able to afford to train, smaller models to be used on end-customer devices like what Apple clearly seems to be planning for on the iPhone, and specialised models for ultra specific use cases like this one.