Proud to share our latest publication in npj Digital Medicine! 📢 As you may have heard, our team is working to automate suturing skills assessment through AI models. 📢 This NEW work builds on prior work showing that sub-skills within suturing are dependent on each other. Better

💊 We took your language model to the drug store… and it knew about acetaminophen (generic name) better than Tylenol (brand name)! Hugh Zhang Scale AI developed GSM1K last month, where they found many of the open-source models are overfitting to the benchmark. We wondered how

I'm excited about this new clinical trials dataset from Jimeng Sun: 479K clinical trial outcomes from diverse sources, enabled by LLMs and validate din a verity of ways. Wild and cool idea. arxiv.org/pdf/2406.10292 #AI #Healthcare #ClinicalTrials #DrugDiscovery

🌟 A neural network learns words through a baby's 👶 eyes and ears 🎥! Trained on 61 hours of one child's video to maps words to visuals and enables zero-shot generalization. So cool. science.org/doi/10.1126/sc… #AI #LanguageLearning #MachineLearning

Models of the human brain’s cortex have been made by combining cells from up to five donors. This approach could enable genetic background to be accounted for in studies of brain development and disease go.nature.com/3RGQnaf

We are thrilled to welcome Matthew McDermott, PhD (Matthew McDermott), and Xuhai “Orson” Xu, PhD (Xuhai “Orson” Xu), to the DBMI faculty in 2025! Columbia University Columbia Medicine Columbia Vagelos College of Physicians & Surgeons dbmi.columbia.edu/matthew-mcderm…

How constrained is each gene in the human genome? Our work on estimating gene constraint using loss-of-function mutations along with gene features (expression patterns across tissues, protein language model embeddings, etc...) is now officially out! 🧵/8 nature.com/articles/s4158…

New preprint! We revisit the decades old idea of predicting the proteome from the transcriptome, but as far as I know, we are one of the few to use deep learning doi.org/10.1101/2024.0… A thread 🧵

Our new paper in JAMIA, led by Jason Patterson, predicts individual risk of adverse drug events for MS therapies, leveraging EHR data and knowledge graphs for explainable AI. Check it out: doi.org/10.1093/jamia/… #MS #AdverseEvents #AI #EHR

"the tobacco, ultra-processed food, fossil fuel, and alcohol industries together account for some 19 million deaths worldwide every year" In The Lancet Diabetes & Endocrinology, authors discuss a new WHO report on the commercial determinants of non-communicable diseases: hubs.li/Q02GcFWR0

Mongolia's Olympic uniforms are lovely. But what's up with the wording on this tweet? Is the insinuation that we don't do the same? Let's talk about how the USA Olympics uniform connects to our heritage, culture, and history. 🧵

📣 Life update: I’ve joined OpenAI and am hiring researchers! 💥 I’m immensely grateful to all of my teammates at Google/DeepMind over the last ~5 years–you all have taught me so much. I’m excited to continue marching towards our shared mission to enable universal access to

And we are there! check this out science.org/doi/10.1126/sc… Eric Topol Harvard T.H. Chan School of Public Health Harvard Data Science Initiative

My dissertation won the AMIA 2024 Edward H. Shortliffe Doctoral Dissertation Award Honorable Mention! Look forward to giving a talk on my dissertation "Causal machine earning for reliable real-world evidence generation in healthcare" at #AMIA2024 Annual Symposium. Dissertation is

An unedited photo of Brazilian surfer Gabriel Medina after receiving the highest surfing score in Olympic history (a 9.90) 📸 Jerome Brouillet Photography

Subtitle - how to build fake data that is useful for society

UPDATED: we have identified 313 (!) funding opportunities and fellowships for postdoctoral fellows. Download this database freely here: research.jhu.edu/rdt/funding-op… For each entry, we provide description, amount, deadline, eligibility, and link to funder. Good luck!

LLMs are going to need as much verified human generated content as possible and companies that build and use them should be using their revenue to support independent journalism and publishing. Seems obvious. nytimes.com/interactive/20… via @NYTimes

Excited to present CarDS Lab screening with wearables projects this afternoon at 3:45 (Digital Health Stage) and compete in the Young Investigator Award competition tomorrow at 12:15 (Science Box 3) discussing developing digital twins of RCTs! Stop by and say hi!👋🏽

Back together!