Ex-Postdoc @MIT_CSAIL, helping computers to see. Now AI person @Xyla and @EvidenceOpen, helping computers help humans and become true bicycles for the brain.
In Offcall's new 2025 report surveying 1,000 doctors: 44% now use OpenEvidence daily. It's the clear #1 by a mile.
The report sums it up perfectly: "Physicians are adopting AI on their own, often using personal subscriptions to the hottest AI tools, because their organizations can't move fast enough."
Clinicians are voting with their keyboards.
Powerful data from @grahamwalker and @OffCallDotCom
Introducing ClinicalKey AI by Elsevier Health!
A breakthrough in clinical decision support, combining trusted medical content with advanced AI capabilities.
Discover how it's set to help change the way medical info is accessed. ๐ https://t.co/pspLGjZAG1
#ClinicalKeyAI#AI
This paper asks โDo We Still Need Clinical Language Models?โ in the era of ChatGPT? Their analysis finds the answer, at least for now, is yes: https://t.co/68614xLVSY
@Michael_J_Black Also, teaches you how to make an argument (quantitatively through evaluation or otherwise) for your approach / algorithm / intervention; something that most fancy LLM demos here on twitter are very much in need of.
As language models get ever larger, is there still a place for smaller specialized models? In this work we find that there is, and that their performance is comparable to much larger general language models while requiring less data + computation.
Recent results have suggested that LLMs encode a surprising amount of clinical knowledge. These results raise an important question about the role of smaller specialized clinical models: do we still need clinical LMs? We explore this question in our paper: https://t.co/CK7ZAoBy7W
@drjwrae Love the idea of "compression for intelligence". Heard Daniel Burfoot talk about this (eg https://t.co/W5mU1xQnHj) at CVPR a long time ago, fascinated by this ever since. Will the talk be broadcast?
@igilitschenski I think the starting point makes the difference - "Hm, I wonder why this happens" vs "This is annoying and I want to fix it". And ideally over the (long term) course of a project they converge.
Incredibly honored to receive the Koenderink Prize at #ECCV2022 for our work on transforming the #Sintel movie into a benchmark and dataset for Optical Flow (and more!), together with Daniel Butler, @GStanleyNeuro, and @Michael_J_Black .
Some background: https://t.co/CCV9D2hH57
Excited to share our ICLR 2021 paper on image composition in GAN latent space! joint with @jswulff@phillip_isola
paper+code+colab: https://t.co/zsxYHw2xgb
it's interactive and super fun to play with :)