| |
| |
|
|
| |
Read about a new benchmark for clinical AI agents, a conference that accepts only AI-written papers, and how LLMs could improve population health.
|
| |
|
|
|
| |
| |
| |
|
|
| |
At the recent Open Conference of AI Agents for Science, a virtual event, attendees explored if and how artificial intelligence can independently generate novel scientific insights, hypotheses, and methodologies while maintaining quality via AI-driven peer review. The conference organizers, who mostly hailed from Stanford University, say existing norms incentivize researchers to hide or minimize AI’s contributions. This prohibition, they assert, hinders researchers’ ability to understand and shape the role of AI in future scientific investigations.
Research papers presented at the conference, abbreviated Agents4Science, were written and reviewed by AI.
Read more about the conference and watch a recording of it.
|
| |
|
|
|
| |
| |
| |
|
|
| |
The most recent iteration of OpenAI’s popular large language model poses safety risks in medicine, according to a commentary published Oct. 16 in
Nature Medicine. Stanford Medicine researchers found GPT-5 is still prone to hallucinations and is vulnerable to cleverly worded prompts or ambiguous queries that could bypass system-level rules. For example, it could bypass adherence to approved drug guidelines and assist in bioweapon development when safeguards are removed.
Stanford Medicine scholars Tina Hernandez-Boussard, PhD, Rebecca Handler, and Sonali Sharma assert that these weaknesses pose a particular problem in medicine and public health, “domains in which accuracy, adherence to safeguards, and interpretability are paramount.”
Read the article.
|
| |
|
|
|
| |
| |
| |
|
|
| |
AI agents can work autonomously, performing complex, multistep tasks with minimal human supervision. Yet no benchmark test to evaluate the quality of AI agents for clinical purposes has been available.
Now, a multidisciplinary team of Stanford University researchers has developed just such a test: MedAgentBench. In a paper published Aug. 14 in The New England Journal of Medicine AI, they report on how popular LLMs fared as clinical agents when evaluated with the tool.
Read the paper and an article about it.
|
| |
|
|
|
| |
| |
| |
Photo courtesy: Sarah Pelta
|
|
| |
|
| |
| |
|
|
| |
The inaugural AI and Cancer Research Symposium brought together roughly 300 researchers, clinicians, engineers, and data scientists, including 115 virtual attendees, on Nov. 4 to explore the transformative role of AI in cancer research and care.
The event, held at the Stanford Cancer Institute, featured panel discussions about AI applications in drug design, basic science, clinical care, and translational research. From improving diagnostics to accelerating drug discovery, the speakers and poster presentations underscored how AI is catalyzing meaningful progress across cancer research.
Read more about the symposium.
|
| |
|
|
|
| |
| |
| |
|
|
| |
Registration is open for AI+HEALTH 2025, a virtual conference designed to equip participants with actionable insights and strategies for navigating AI adoption in health care.
The event, set for Dec. 9-10, is presented by the Stanford Institute for Human-Centered Artificial Intelligence, Stanford Center for Artificial Intelligence in Medicine and Imaging, and Stanford Center for Continuing Medical Education. Content will be relevant to clinical practitioners, researchers, executives, policymakers, and other professionals, with or without technical expertise.
Register for the event or read more about it.
|
| |
|
|
|
| |
| |
| |
|
|
| |
Stanford Medicine physician-scientists explored how large language models could improve population health by influencing health-related behaviors through personalized coaching. Such coaching could be tailored to a person’s age, gender, race, ethnicity, language, baseline psychological profile, and level of education, according to
Euan Ashley, MB ChB, DPhil,
Fatima Rodriquez, MD, MPH, and
Daniel Seung Kim, MD, PhD, MPH, a former postdoctoral scholar at Stanford Medicine who joined the faculty of the University of Washington in August.
This type of approach also could increase the representation in clinical trials of “patient populations at higher risk of developing chronic disease, who were previously implicitly or explicitly excluded from digital health trials due to a lack of English fluency,” they write.
Read the commentary.
|
| |
|
|
|
| |
| |
| |
|
|
| |
The Stanford Data Ocean, an AI-powered platform designed to help make training in precision health accessible and affordable, improved learning outcomes for low- and middle-income students in the United States and 92 other countries, according to a new study led by Stanford Medicine researchers.
The study, published Aug. 16 in Communications Medicine, asserts that AI-enabled, cloud-based education platforms like SDO can make bioinformatics education and research more equitable.
Read the study.
|
| |
|
|
|
| |
| |
| |
|
|
| |
In the Oct. 3 episode of the podcast Plain English, host Derek Thompson speaks with Lloyd Minor, MD, dean of the Stanford School of Medicine and vice president for medical affairs at Stanford University, to better understand how AI is poised to benefit medicine.
“The dream is dazzling,” Thompson, a journalist and co-author of the bestselling book Abundance, writes in his introduction to the episode. He invited Minor as a guest on the show “to separate fact from fantasy.”
Listen to the episode.
|
| |
|
|
|
| |
| |
| |
|
|
| |
A new professional development course is designed to equip health care professionals with the strategic insight and practical tools needed to responsibly implement and scale AI solutions in clinical and organizational settings.
Stanford AI in Healthcare Leadership and Strategy: from Innovation to Implementation will be offered through the Stanford Center for Continuing Medical Education. The month-long program blends asynchronous online learning, recorded virtual sessions, and a two-day onsite session, May 29-30, at Stanford University. Participants will engage with faculty and industry leaders, explore real-world case studies, and develop strategic approaches to integrating AI in health care delivery.
Read more about the course.
|
| |
|
|
|
| |
| |
| |
Explaining AI jargon, one concept at a time
|
| |
|
|
| |
|
|
|
| |
| |
| |
|
|
| |
Transfer learning is a machine learning technique in which a model trained on one task is adapted to a different but related task. Think of it like using your experience riding a bike to learn how to ride a motorcycle; you're building on foundational skills rather than starting from scratch.
Models trained from scratch require huge datasets, but fine-tuning a pre-trained model is far more efficient because it has already learned to recognize useful patterns. For example, an AI tool trained to analyze chest X-rays for pneumonia has learned to identify relevant features like tissue density and abnormal opacities. Researchers can adapt this same model to detect lung cancer by retraining just the final decision-making layers, requiring far less data and computational power than building a new model from scratch.
|
| |
|
|
|
| |
| |
| |
A joint initiative between Stanford Medicine and the Stanford Institute for Human-Centered Artificial Intelligence (HAI) to guide the responsible use of AI across biomedical research, education, and patient care.
To unsubscribe from future emails, CLICK HERE.
|
|
| |
|
|
|
