🔥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

In our new preprint, we evaluated GPT-4V on 934 challenging NEJM medical image cases and 69 clinicopathological conferences. GPT-4V outperformed human respondents overall and across difficulty levels, skin tones, and image types except radiology, where it matched humans. GPT-4V

Do large language models have a probabilistic understanding of disease states? And what does this mean for the future of diagnosis and clinical reasoning? I explore this with Thomas Buckley, Arjun (Raj) Manrai, and Dan Morgan in our new paper in JAMA Network Open. A brief 🧵⬇️

(1/4) Excited to introduce ProCyon: a multimodal foundation model to model, generate, and predict protein phenotypes, led by stellar Owen Queen Yepeng Robert Calef Valentina Giunchiglia 👉 biorxiv.org/content/10.110… ProCyon is an 11B parameter multimodal model that integrates protein

Is a well-placed word worth a thousand pictures for an AI? "the provision of images either reduced or had no effect on model performance when text is already highly informative." arxiv.org/abs/2311.05591 Kudo Thomas dbmi.hms.harvard.edu/people/thomas-…

Preprint out today that tests o1-preview's medical reasoning experiments against a baseline of 100s of clinicians. In this case the title says it all: Superhuman performance of a large language model on the reasoning tasks of a physician Link: arxiv.org/abs/2412.10849 A 🧵⬇️

o1-preview is far superior to doctors on reasoning tasks and it's not even close, according to OpenAI's latest paper. AI does ~80% vs ~30% on the 143 hard NEJM CPC diagnoses. It's dangerous now to trust your doctor and NOT consult an AI model. Here are some actual tasks: 1/5

‼️"o1-preview demonstrates superhuman performance in differential diagnosis, diagnostic clinical reasoning, and management reasoning, superior in multiple domains compared to prior model generations and human physicians." And this is using vignettes, not multiple choice.

How does o1-preview compare to hundreds of clinicians in medical reasoning? We explore this in our new preprint, led by stars Thomas Buckley of DBMI at Harvard Med and Peter Brodeur of BIDMC. Full text: arxiv.org/abs/2412.10849 Great thread by co-conspirator Adam Rodman below 👇

Thanks for the sneak peak of this last month, Adam Rodman. This work from Adam and Arjun (Raj) Manrai (co-led by the inimitable Thomas Buckley) is really worth a read. The most astounding result highlighted in the thread below: 👇🏽

Excited to share recent results on the impressive performance of 01-preview on medical diagnosis & management challenges arxiv.org/abs/2412.10849 Additional reflections on developments in my LinkedIn article: tinyurl.com/za3nj5yw

promising results of o1-preview for medical reasoning (though still lots of work to figure out how to integrate with the healthcare system)

Open-source LLMs have narrowed the gap with leading proprietary models surprisingly fast. But how well do they do on tough clinical cases? Excited to share our new study out today in JAMA Health Forum led by Thomas Buckley with amazing physician coauthors. Detailed thread to come

Full study: jamanetwork.com/journals/jama-… News coverage from Harvard Medical School: hms.harvard.edu/news/open-sour…

An open-source LLM performed comparably to GPT-4 in generating differential diagnoses for complex cases, indicating that high-performing, locally-deployable custom models can enhance clinical decision support while ensuring data privacy and flexibility. ja.ma/4bGTyHx

New research letter on the diagnostic abilities of open-source models for diagnosis led by superstar Thomas Buckley -- short letter (and something that I think most researchers already know) but big implications for medicine. A 🧵⬇️

AIM DBMI at Harvard Med graduate students already pushing the frontier in medical AI thecrimson.com/article/2025/3…

Huge update to our preprint today on the superhuman performance of reasoning models in medical diagnosis! TL;DR – they don't just surpass humans in meaningful benchmarks, but in actual medical care from unstructured clinical data: A 🧵⬇️: x.com/AdamRodmanMD/s…

We just added a major new experiment to our o1 study, comparing o1 and attending physicians at key diagnostic touchpoints on REAL cases from the BIDMC ER Stellar work led by Thomas Buckley and Peter Brodeur as part of a rapidly growing DBMI at Harvard Med and BIDMC collab 🚀

Updated paper by physicians at Harvard, Stanford, and other academic medical centers testing o1-preview for medical reasoning & diagnosis tasks: “In all experiments—both vignettes and emergency room second opinions—the LLM displayed superhuman diagnostic and reasoning abilities.”