AI and Biology

Paris
Joined November 2018
A month ago, I wrote about why we release open-weight models at @bioptimus_ai (to increase impact & get rapid feedback). I promised a follow-up on what actually happened once our models were out in the wild. Today, we’re sharing the results: H-Optimus just hit 2M downloads! 🚀
4
1
134
3️⃣ General-purpose models work. We're seeing researchers across 59 countries reuse the same baseline pathology representations for vastly different diseases and datasets. 🌍
9
2️⃣ Real-world applications are vast. 60% of users start with benchmarking, but quickly move to biomarker discovery, tissue segmentation, and predicting patient survival. (Over 95% of disease-specific use cases are in oncology 🔬).
1
30
What did 2 million downloads teach us about foundation models in biology? 1️⃣ Adoption is completely organic. It rarely happens top-down. One researcher tries it, a neighboring lab sees the results, and it spreads bottom-up through major pharma companies & universities.
15
AI just claimed a proof on a Millennium Problem. Mathematicians feel vertigo. I see a renaissance: machines will explore the forest of mathematical truth, and mathematicians become its interpreters and pathfinders, the secular theologians of the AI era. acsivoluptate675147.substack…
1
3
8
1,754
Huge congrats to Judith Abecassis on her "Woman in Tech" recognition by @AcadTechnolog She shows the next generation that top-tier AI research and deep humility go hand-in-hand. To young women in STEM: aim high, push through the "no's," and own your path. academie-technologies.fr/en/…
2
7
320
AI products move fast; understanding why neural nets behave the way they do moves much slower — and needs open publishing, which frontier labs seem to be doing less of. One small win this week: a conjecture of ours just got proven. Some thoughts about it: linkedin.com/pulse/from-conj…
2
8
628
S:CORT joins STELA—our initiative to build the world's largest clinically linked, multimodal atlas of human biology. The vision is simple: ensure that what our AI learns from one patient today can directly inform the care of the next tomorrow. Onward!
138
To build AI that actually moves the needle for cancer patients, you need exceptional data depth. S:CORT brings exactly that: 1,750+ colorectal cancer patient profiles, up to 6 modalities per patient (imaging, genomics, clinical data), and longitudinal outcomes.
3
184
🎉 Excited to share that we've just hit a $76M funding milestone for @bioptimus_ai , with a fresh $41M round to build the first multiscale foundation model of biology: bioptimus.com/news/bioptimus…
1
1
29
1,650
Back in SF for #JPM25! Eager to explore and discuss the cutting-edge advancements of AI in healthcare. 🧘‍♀️And to beat jet lag next Monday, nothing better than a free charity fitness class hosted by @OwkinScience at the Equinox Sports Club in San Francisco owkin.com/connect/fitness-an…
4
517
I've been eagerly awaiting this book for years! At last, a standalone and meticulous exposition of the current mathematical principles of machine learning, adorned with beautiful proofs. Well done and thank you, @BachFrancis . Discover more here: francisbach.com/my-book-is-o…
1
2
24
2,310
Honored to join @AcadTechnolog (National Academy of Technologies of France)! Grateful to my teachers, students, and collaborators over the years. I’ll strive to advance thoughtful, inclusive progress in tech and help society understand innovations like AI. buff.ly/4gDbvYP
2
1
19
1,423