🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Quitting programming as a career right now because of LLMs would be like quitting carpentry as a career thanks to the invention of the table saw.

'One Patient, Many Contexts: Scaling Medical AI Through Contextual Intelligence' Michelle M. Li (李敏蕊), Ben Y. Reis, Adam Rodman, Tianxi Cai, Noa Dagan, Ran Balicer, Joseph Loscalzo, Isaac Kohane, Marinka Zitnik doi.org/10.48550/arXiv… (9/21)

Embracing your inner robopsychologist. zaklab.org/blog/embrace-y… I'm going to have to refresh this take of having theory-of-mind of our AI colleagues. But Susan Calvin from Asimov's "I Robot" is becoming more relevant than his 3 Laws of Robotics.

Excited to share our new work on knowledge tracing approach to personalized education, with a real-world, prospective evaluation in Harvard Medical School AI in Medicine 2 course zitniklab.hms.harvard.edu/AIM2, led by Grey Kuling 🧠📚 arxiv.org/abs/2507.00032 What's new? Knowledge tracing blends

What's in a smile? A diagnosis? 10 years from now what percentage of chronic neurological diseases will be first diagnosed at home by patient-controlled AI? ai.nejm.org/doi/full/10.10…

The authors of a new Perspective write about technology that alleviates the current mental health workforce shortage by focusing on the initial stages of mental health treatment for both primary care providers and mental health professionals. Learn more: nejm.ai/4k3NUSy

Antonio Regalado Matthew Herper Post reads like a Zagat review

Game theory meets LLMs. I think this makes the revenue cycle challenge with dueling AI's quite vivid CC Sam Finlayson HT Rachel Ramoni

Perhaps the terroir of the substituted blood products was lacking?

On the latest episode of the NEJM AI Grand Rounds podcast, Dr. Alan Karthikesalingam (Alan Karthikesalingam) shares why he believes synthetic data — and long-tail thinking — are essential for building more robust clinical AI. Listen now: nejm.ai/ep31

🌍 Excited to open up our global evaluation of AI for drug decision-making and therapeutic reasoning Shanghua Gao Want to shape the future of therapeutic AI, from understanding existing medicines to developing new treatments for diseases with limited options? Start here:

Isaac Kohane cc

Welcome to order of AI mystics: phase changes like Grokking related to density of prime numbers in model weights

Interesting study explaining why generative AI is brittle with unknown unknowns: Out-of-Distribution Detection Methods Answer the Wrong Questions arxiv.org/abs/2507.01831

Had a great time at AACR AI conference sharing vignettes about our recent research 1) TxAgent - AI agent for therapeutic reasoning across a universe of tools. Start here to test TxAgent and submit your most challenging therapeutic questions: txagentevals.curebench.ai Shanghua Gao

Programmers who are not intimately familiar with their full stack will benefit more. Alas I am not the coder I was when I did my doctoral work on a Lisp Machine (Dandelion) and knew the stack from top to bottom. So I am seeing very large gains in programming productivity. HT

Base model supremacy hypothesis: Half-way through the podcast we learn that many of the enhancements to the conversational medical diagnostic AI—AMIE—were no longer required because of improved performance of the base model. It's a version of the scaling hypothesis:

“Cookie banners make the internet ugly and waste countless hours daily. They fail to protect privacy and harm the web as a platform, despite widespread knowledge of their ineffectiveness. It's a tragedy that cookie banners made the entire internet ugly in one fell swoop”

Top talent cycling between AI companies.

Looks like AI-augmentation of medical student eduction when thoughtfully applied works better than the alternative NEJM AI ai.nejm.org/doi/full/10.10… Will we start scaling education? CC Ethan Mollick