Good to see MedGemma, a unified, open, transparent foundation model by @Google published @NatureMedicine today, with diverse medical image-language capabilities (Figure), can run on a smartphone, weights @huggingface https://
nature.com/articles/s4159
1-026-04626-w
…
@erictopol
-
MedGemma open medical foundation model published in Nature Medicine
By
–
-
AI retina imaging predicts preeclampsia and pregnancy vascular risks
By
–
The richness of the retina photo or OCT predictive data, as interpretable by AI, but not expert physicians, is extraordinary. Today a new report on retina AI for preeclampsia and pregnancy vascular risks @NatureBiotech https://
nature.com/articles/s4158
7-026-03303-0
…
My summary of the key 15 publications -
Nolla Health launches AI Derm photo-to-prescription product
By
–
For your comments. Today @nollahealth introduced their AI Derm product which goes from user photo to prescription in Utah.
-
FDA-approved AI predicts 5-year breast cancer risk from mammograms
By
–
A "normal" mammogram interpreted by AI can accurately forecast high-risk of breast cancer in the next 5 years. Approved by FDA, now becoming available. But it should be free and part of every mammogram readout.
Front page @nytimes today, by @RoniNYTimes gift link -
RCT: Patient-facing chatbot improves lung diagnosis accuracy
By
–
A randomized trial of a patient-facing chatbot vs web based search enhanced accuracy of diagnosis of lung conditions This is one of the first RCTs for public chatbot assessment in healthcare https://
nature.com/articles/s4436
0-026-00189-9
… -
AI stress testing finds vulnerabilities in EPIC health records
By
–
EPIC maintains ~325 million electronic records and used AI stress testing to detect privacy and security vulnerabilities to AI. The risks are substantial amidst efforts to amp up defense.
gift link, by @CFRowland https://
nytimes.com/2026/09/30/bus
iness/ai-epic-health-records-privacy.html?unlocked_article_code=1.FlE.DnDS.AyT48cZFU7Jm&smid=url-share
… -
Frontier models improve but insufficient for reliable medical use
By
–
Thanks for running our open-source work on current frontier models “The results are: the most capable models today (GPT-5.5 Pro) did outperform the best models from before (79/100 vs 69/100), but did not improve enough to be considered sufficient for reliable medical use.” Read
-
Generative AI Designs Optimal Burger for Taste, Nutrition, and Environment
By
–
Generative AI design of the best burger for its taste, nutrition, and for the environment
-
Burden of proof to show current models are not susceptible
By
–
That’s incorrect. The burden of proof lies in demonstrating current models are not susceptible to the same issues we identified. To assume otherwise is blind faith in untested models
-
Frontier AI models fail medical reasoning stress test, study finds
By
–
We stress tested many frontier AI models for multimodal medical reasoning (including GPT-5, Claude 3.5, Gemini 2.5 Pro). They’re not ready. Faulty reasoning, use of inappropriate shortcuts, hallucinations. Published today @NatureMedicine https://
nature.com/articles/s4159
1-026-04501-8
…