Along with the rest of the industry, I am pretty excited to see the impact LLMs are gonna eventually have in healthcare, the potential is very significant if done right.
Keep an eye out for a full/extended manuscript forthcoming! (3/3)
@sdas617@Colin_Magdamo@Lily_You_Cheng
Been sitting on this a while but the work from my thesis is finally up as a preprint on arXiv!
Given the speed things have been moving in the LLM space, BERT with its context window of 512 tokens already feels woefully out of date. (1/3)
https://t.co/cZBLRtAX0H
Even with this limitation, we were able to recover structure over much longer documents, and use these embeddings from EHRs to cluster MGB patients with Alzheimer's disease and other dementias. (2/3)
A few technical reports/papers to read for your long weekend:
Google's Gemini 1.5: https://t.co/yFnihVh3eK
OpenAI Sora: https://t.co/5TZAv5WaM0
UCBerkeley's RingAttention multimodal model: https://t.co/jC6vrNsg3o
Google's Mixture of expert model: https://t.co/qGEtspbvhS
Happy to say this work is finally out in JAMIA Open!
TL;DR: Together with collaborators at BMC, we identified a new EHR metric predictive of burnout among practicing physicians.
https://t.co/ox2FaTdcoM
In the NYT today, Cade Metz implies that I left Google so that I could criticize Google. Actually, I left so that I could talk about the dangers of AI without considering how this impacts Google. Google has acted very responsibly.
@notkavi I suppose it's better to be exposed to a perpetually increasing market than renting in one, which is subject to the same perpetual proportional increase.
New survey paper! We discuss “emergent abilities” of large language models.
Emergent abilities are only present in sufficiently large models, and thus they would not have been predicted simply by extrapolating the scaling curve from smaller models.
https://t.co/qX3OtaPQH9
🧵⬇️
Thank you to all of our presenters and attendees from @MIT, @MassGenBrigham, and beyond for making this year's conference both exciting and meaningful. We hope that this lays the much-needed groundwork to take the next concrete steps in clinical AI. (2/2)