As I shared in the NYT, models often see the data but fail to weigh it like a physician, drifting toward generic "average patient" responses. Context window ≠ Clinical reasoning. www.nytimes.com/2025/12/03/w...

“‘Just because you’re providing all of this information to language models,’ @daniellebitterman.bsky.social says, ‘doesn't mean they're effectively using that info in the same way that a physician would’. And once people upload this kind of data, they have limited control over how it is used.”🧪🛟

Check out our editorial on Zazzetti et al (2025)'s paper on synthetic data generation for breast cancer, in JCO CCI! Synthetic data could help with many gaps in clinical AI research, but challenges remain especially (IMO) issues with out-of-domain generalization @shan23chen.bsky.social

Super proud of @shan23chen.bsky.social for his podium presentation on his research into LLM sycophancy in the face of illogical medical queries at #AMIA25! Full paper: www.nature.com/articles/s41... Also cited yesterday in the NYT! www.nytimes.com/2025/11/16/w...

LLMs tend to prioritize helpfulness > reason. We show that safety-aware, compute-efficient fine-tuning helps models reason more critically in healthcare domain, and generalizes to improved safety alignment across other domains. www.nature.com/articles/s41... @shan23chen.bsky.social

An overemphasis on helpfulness makes LLMs vulnerable. Research shows models will comply with illogical medical requests, generating false information. This sycophantic tendency can be corrected with specific prompting and fine-tuning. #MedSky #MedAI #MLSky

Mass General physician-scientist @daniellebitterman.bsky.social discusses how AI assists the clinical data pipeline leading to better treatments for patients. Listen to unNatural Selection & register for #WMIF2025 at the link in bio to hear more : www.unnaturalselection.net/podcast/s1e19 #MedTech

Our paper on multilingual reasoning is accepted to Findings of #EMNLP2025! 🎉 (OA: 3/3/3.5/4) We show SOTA LMs struggle with reasoning in non-English languages; prompt-hack & post-training improve alignment but trade off accuracy. 📄 arxiv.org/abs/2505.22888 See you in Suzhou! #EMNLP

Are you driven to use AI to transform patient outcomes in oncology? My lab in the AI in Medicine Program (Mass General Brigham, Harvard Medical School) is seeking Postdoctoral Fellows to pioneer applications of AI—especially LLMs—in cancer care. More here: www.linkedin.com/posts/daniel...

Reliability of Large Language Model Knowledge Across Brand and Generic Cancer Drug Names | JCO Clinical Cancer Informatics ascopubs.org/doi/abs/10.1... #JCOCCI @daniellebitterman.bsky.social

Does your LRM reason in your language? Check out new preprint led by ✨ @jiruiqi.bsky.social & @shan23chen.bsky.social. Implications for safety/human oversight & accuracy!

Agents are all the rage and we need to track their abilities in the medical domain. Enter MedBrowseComp, the 1st benchmark to assess agents' abilities to reason, navigate the web, and search for verifiable med info! Preprint: arxiv.org/abs/2505.14963 Site: moreirap12.github.io/mbc-browse-a...

I’m thrilled to be in San Francisco for @statnews.com's Breakthrough West Summit! I’ll be bringing my firsthand perspective as a physician-scientist to speak about how AI is transforming cancer care, alongside leaders in the field. Let's connect if you're here! #STATBreakthroughSummitWest

AI in Cancer Care Artificial intelligence has the potential to upend oncology, changing everything from diagnosis to treatment options. Get a wide-ranging view of how the use of technology could play out over the next few years. Moderated by @angusrohan.bsky.social #STATBreakthrough

Exciting news: we are organizing a shared task – 2nd edition of the Chemotherapy Treatment Timelines Extraction from the Clinical Narrative (text mining task) -- collocated with the Clinical NLP Workshop. Do LLMs solve the task? Check out bit.ly/ChemoTimelin...

A pie graph worth keeping in mind as the NIH budget plummets jamanetwork.com/journals/jam... for 356 new FDA drugs approved

Conference and professional societies: PLEASE make hybrid options available for attendees and presenters at your conferences so that scientists from HHS-funded agencies can attend. These are unmissable opportunities to promote all the great intramural science and scientists from our government.

My Perspective in @NEJM_AI. AI could distort clinical decision-making in ways that prioritize profit over patient care. Oversight & regulation must go beyond performance metrics alone to address hidden commercial forces that could shape decision support. ai.nejm.org/doi/full/10....

My opinion as an actual NIH-funded researcher (unlike Vinay) at ucsf: his lies about how NIH dollars are used reflect a complete lack of understanding of how research is performed, a lack of respect for research, and are harmful to the entire biomedical research enterprise #grifter

Budgeting for the next year of my grants and they will all need to be rescoped, even before the 15% IDC rate. NCI funding at 83% for new awards and another 10% reduction for renewals (current state). Essentially, we are getting 50% of what we asked for...how is this sustainable? @carlbergstrom.com

As a cancer doctor I see every day how NIH-funded clinical trials save lives and has made the U.S. a leader in medical innovation. Here's one example: In the 1970s, childhood cancer survival was only 58%. Today it is 85%, largely thanks to NIH/NCI funding of Children's Oncology Group trials.

Congressional delegation outside USAID now: “We are here to shed a light on a crime unfolding before our eyes.”

Senator Andy Kim just went to the USAID building, talked to the security guard there to confirm employees are being barred entry, and then did a press gaggle right there in front to call it out. This is doing something. This is making an effort on messaging. Other Democratic lawmakers: take notes.

Gay? Lesbian? Trans? Intersex? NYC Health has health information for everybody. 🏳️‍🌈🏳️‍⚧️

LLM attachment styles: Secure - Claude, Anxious - DeepSeekR1/o1, Avoidant - Gemini Am I wrong?

In addition to the research funded by the NIH, I am grateful and indebted to the dedicated NIH scientists & staff. Their work advances breakthroughs, scientific careers, and improves & saves lives across the U.S

Also applies broadly to treating trainees/junior faculty kindly and collaboratively.

Love this paper, thanks for this contribution @roxanadaneshjou.bsky.social and @rajpurkar.bsky.social. This is the way to start to understand how LLMs can and should perform for diagnostics - clinical vignettes don't reflect real world medical practice.

We have a NEW PAPER in @naturemedicine.bsky.social on reporting recommendations for addressing the unique challenges of #largelanguagemodels (LLMs) in biomedical applications www.nature.com/articles/s41... #MLSky #StatsSky #medSky #AISky #artificialintelligence #generativeAI #transparency

TRIPOD-LLM is out! Check out our consensus guidelines for reporting #LLM research in biomedicine. TRIPOD-LLM is intended to be a living guideline to keep up with the rapid advances in LLMs. Kudos to lead author Dr. Jack Gallifant

Physician scientists have my heart full of gratitude. I cannot overstate how important this career is for humanity’s well-being!

Our latest update of the HemOnc knowledgebase is ready and available at Harvard Dataverse: dataverse.harvard.edu/dataset.xhtm.... Have a look! @peteryang.hemonc.org @ecquis.bsky.social

I have not listened to the full lecture, but the content on the NeurIPS keynote slide is xenophobic and unacceptable. There is growing prejudice against Chinese students and academics - let's let this be a catalyst to begin addressing it more deeply. Listen to the eloquent audience member below.

The extraordinary recent takeover of ML/AI by #NLP is well-known but insufficiently reflected on. Look at the @neuripsconf.bsky.social tutorials in 2024! neurips.cc/virtual/2024... 14 tutorials; 6 have "LLM" in the title; 4 more cover foundation models, with large NLP coverage. That's > 70% 😲

I won't be @ NeurIPS, but Jack Gallifant is representing the lab and some of his other great papers. Stop by to chat w him if you're interested in our work! #NeurIPS2024 CrossCare studies how biased pretraining data -> biased LLMs: neurips.cc/virtual/2024... 12/10 4:30-7:30p W Ballroom A-D #5203

I am always worrying about Benzene (my cat)! www.nytimes.com/2024/12/05/w... But please don't stop wearing sunscreen! Sun exposure is a known cancer risk, benzene risks unknown. This article has good tips if you want to minimize benzene exposure. Obligatory Benzene (cat) pic ⬇️

It's that time of year to bring out our @bmj.com xmas paper On the 12th Day of Christmas, a Statistician Sent to Me... from 2022 (w/ @richarddriley.bsky.social) with our list of commonly seen statistical faux pas🎄 www.bmj.com/content/379/... #StatsSky #EpiSky

Check out some of our early experiments on SAE transferability across modalities - moving toward more interpretable vision-language models 🧠

Right there with you... An important point that we all need to reflect on: How are we supporting the next generation of scientists? Young scientists most immediately affected, but there will be ripple effects throughout society

If you are going to NeurIPS, stop by our poster to hear about our Coss-Care project, in the Benchmarks and Dataset Track! 12/11/24 from 4:30pm-7:30pm neurips.cc/virtual/2024... @neuripsconf.bsky.social #NeurIPS

Looking forward to discussing risk management of large language models at the FDA Digital Health Advisory Committee Meeting on November 20th. The meeting will be webcast and is open to public comment through January 1, 2025: www.fda.gov/advisory-com...

🩺💡The Bitterman lab has spent much of the past year researching #LLMs for healthcare. This post summarizes our inroads into making LLMs safer and reliable for clinicians and patients: huggingface.co/blog/shanche.... We'll be at #EMNLP2024 - come chat if you have similar interests!

🎉 Incredibly proud of @shan23chen.bsky.social for being selected for the 2024 Google PhD Fellowship in Natural Language Processing: blog.google/technology/r... !!! So excited to see how Shan's contributions will continue shaping the future of clinical NLP #HealthAI #NLP 🌟

Here are some reflections on many studies we did this year. Tons of progress has been made, but there are still safety concerns..🧐 Poster 10:30 riverfront at EMNLP2024 🏖️ Happy to chat and connect! 📃 huggingface.co/blog/shanche... 🔊 tinyurl.com/aimpodcast24 @daniellebitterman.bsky.social