🔥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

a hype-free, sober conversation on AI for healthcare by ProfDhar with David Sontag at open.spotify.com/episode/75JIST…

In my The Boston Globe Op-Ed today, I urge the US government and philanthropists to ensure that Americans can benefit from repurposed generic drugs: bostonglobe.com/2023/12/27/opi…. Reboot Rx #RebootRx #TheSolutionMayAlreadyExist #cancer #drugrepurposing #nonprofit

Help us evaluate how well code LLM models assist humans in completing coding tasks! Register interest in our online 35 min Python study with $15 compensation forms.gle/BHwJf2b2bp4f4d… it also serves as paid LeetCode practice!

How do LLMs like #ChatGPT impact clinical note readability? LLM augmentations enhance understanding (from user study with 207 participants- 17% increase in understanding for augmentations vs control, p=0.002), but validation was needed to avoid errors: arxiv.org/abs/2401.09637

Learning to Decode Collaboratively with Multiple Language Models We propose a method to teach multiple large language models (LLM) to collaborate by interleaving their generations at the token level. We model the decision of which LLM generates the next token as a latent

Thanks AK for tweeting our work! 🤝Co-LLM trains a language model to collaboratively decode with other LMs. It decides whether to invoke the other assistant LM for each token. By optimally combining the expertise from each LM, co-llm generation can be better than using

Co-LLM can help achieve better performance than using a single model alone. We experimented with co-llm generations at different frequencies for invoking the assistant model, and visualized the generations.

Congratulations to Dr. Hussein Mozannar. Hussein Mozannar Strong dissertation research exploring multiple dimensions of human-AI collaboration. David Sontag Elena Glassman Arvind Satyanarayan Massachusetts Institute of Technology (MIT)

So happy to defend my PhD thesis and couldn't have done it without a 1 of 1 advisor David Sontag and an incredible committee Eric Horvitz Arvind Satyanarayan Elena Glassman

What personalized cancer treatment options are there when clinical trials aren't able to test every patient scenario? Zeshan Hussain, Barbara Lam, Fernando Acosta-Perez, Irbaz Riaz MD,MS,MBI,PhD, Maia Jacobs, Andrew Yee, MD, and #JClinic PI David Sontag propose a ML-based approach arxiv.org/pdf/2404.15187

Excited to present our work at #ICML2024, “Prediction-powered Generalization of Causal Inferences.” Limited external validity of randomized controlled trials (RCT) is a big challenge to causal inference. We investigate how integrating data from observational studies (OS) can

No two cases of multiple myeloma are alike, making treatment decisions a challenge. In npj Digital Medicine, ASPIRE awardee David Sontag introduces a machine-learning model to help physicians estimate disease progression & treatment outcomes. #PrecisionMedicine

New preprint dropped! Times have changed since the IDSA released tx guidelines for uncomplicated UTI in 2010. We looked at a large claims database formatted into the OMOP common data model to see if the recommendations still hold. tl;dr They do 1/7 medrxiv.org/content/10.110…

Thanks MIT for covering this just published ACL '24 paper by my students Shannon Shen and Hunter Lang, in collaboration with Bailin Wang and Yoon Kim! Paper: bit.ly/3B8GGMp

Are LLMs actually helping us code faster or is it just more fun? To answer this, we created RealHumanEval, a code editor with AI (think LeetCode+AI) designed to run human studies. We evaluated 7 LLMs and found that YES they do but it's not 10x and only for LLMs strong enough!

Large language models often "hallucinate," giving misleading/incorrect information. #JameelClinic PI David Sontag recently co-authored a new paper introducing a model called SymGen that quickly verifies LLM-generated text. More in MIT News: news.mit.edu/2024/making-it…

What if AI didn't wait for developer questions, but could anticipate their needs? Our new #CHI2025 paper rethinks designing AI assistants for programming—from reactive tools to proactive partners.

We are thrilled to announce our partnership with Intermountain Health to deploy Layer Health's AI platform for clinical registry abstraction across 33 hospitals spanning multiple states. fiercehealthcare.com/ai-and-machine…