ืชืืื ืื ืื ืืืืงืจืื ืืกืื ืืช ืืชืืืื ืฉื ืืืื ืืขืื ืฉื ืชืืื ืืืืจ ืืืื ืฉื ืืืจืืื ืืฆืืืข ืฉืืืื ืืืฉืืื ืฉืฆืืืข ืืืืื
๐๏ธ Uri Goren retweeted
Researchers handed an AI agent a 40-year-old Fortran climate simulation and said โport it to modern GPUsโ
And it did it.
It solved one of the most expensive bottlenecks in modern science.
For decades, the world's most critical scientific applications, weather models, fluid dynamics, nuclear simulations, have been trapped in the past.
They were written in ancient Fortran. They were built for basic CPUs. They were never designed for the massive, parallel GPU supercomputers we use today.
To manually rewrite a 250,000-line legacy codebase into modern C++ takes teams of specialized HPC engineers years of agonizing, line-by-line translation.
It is a multi-million-dollar nightmare.
So researchers built an AI to do it instead.
They didn't just ask an LLM to translate text. They built a fully autonomous agentic workflow.
The AI acted like a senior systems engineer.
It translated the ancient Fortran into Kokkos (a modern C++ performance framework). It compiled the new code. It ran the simulation. It profiled the execution time. It found its own performance bottlenecks.
It iteratively rewrote its own code to fix them.
It looped this process until the 40-year-old simulation was screaming across modern hardware with a 5x speedup.
Without a human touching the keyboard.
We have spent the last two years obsessing over AI writing new apps.
But the real superpower is an AI that can reach into the past, rip out decades of dead legacy infrastructure, and autonomously drag it into the exascale era.
The greatest breakthroughs of the next decade won't just come from new AI models.
They will come from AI resurrecting the science we didn't have the compute to finish.
๐๏ธ Uri Goren retweeted
ืืขืื ืฉืืืขืืื ืืืื ืืืืงืจ ืฉืื ื ืืงืื ืคืจืก ืืงืฆืื ืืจืคืืื ืืจืืฉื ืฉื ืฆื"ื. ืื ืืื ืืคืจืกืื ืฉืืื ืจืืืฉื ืืืชื ืืืื. ืืืืืงื ืื ืืืื ืืคืจืก ืขืฆืื.
โ
ืื ื ืจืืฆื ืืกืคืจ ืืื.
โ
ืืคื ื ืืฉืืข ืฉื ืื ืืงืื ื ืืช KI Research Institute - ืืืื ืืืงืจ ืขืฆืืื, ืืื ืืืจืืช ืจืืื, ืฉืขืืกืง ืืืืงืจ ืจืคืืื ืืืืฆืขืืช ืืืื ืืืื ื ืืืืืืชืืช.
โ
ืืื ืืกืคืงื ื ืื ืืขื. ืคืจืกืื ื ืืืงืจืื ืืืชืื ืืขืช ืืืืขืืื ืืืืงืจืชืืื ืืขืืื, ืืืื ื ืืคืจืกืื, ืืืืื ื ืืืงืจืื ืืื ืืืืืืื ืืืืืื, ืืืขืืงืจ - ืืฆืืื ื ืืืฉืคืืข ืืืจืืื ืจืืืช ืขื ืืจืืืืชื ืฉื ืื ืฉืื.
โ
ืืื ืืืืื ืชื, ืื ืืื ืื ืืืฆืขืื ืืืืจื ืืืช ืืืืืื:
โ
ืืขืฉืืช ืืื ืฉืืืชืจ ืืื. ืืฉืคืจ ืืช ืืืจืืืืช ืฉื ืืื ืฉืืืชืจ ืื ืฉืื.
โ
ืื ืื ืืกืืื ืฉืื ืื ื ืคืืขืืื ืืืืคื ืงืฆืช ืืจืื.
โ
ืื ืื ื ืคืื ืื ืืืฉืจืื ืืืฉืื, ืืืชื ืืืืื, ืืงืืคืืช ืืืืื ืืืืืกืืืช ืืืงืจ ืืืฆืืขืื ืืื ืืฉืื ืคืฉืื:
โ
ืืฉ ืื ื ืืืงืจืื ืืขืืืื, ืืืข, ื ืืกืืื, ืืืืืืช ืืชืงืืืืช ืื ืืชืื ื ืชืื ืื ืืืื ื ืืืืืืชืืช.
โ
ืืื ืื ืื ื ืืืืืื ืืขืืืจ ืืื ืืฉืคืจ ืืช ืืืจืืืืช ืฉื ืื ืฉืื?
โ
ืืื ืืฉืืื ืืืช. ืืื ืืกืฃ ืฉืขืืืจ ืืฆื ืืฆื. ืืื ืืื ืืจืกืื ืืกืืจืืื. ืคืฉืื ืฉืืชืคืืช ืืืขื ืืืจื ืืฉืืชืคืช.
โ
ืืืืืฉื ืืืืช ืขืืืืช.
โ
ืืื ืืคื ื ืืฉื ืชืืื ืืื ื ืฉืืฉ ืื ื ืืขืื.
โ
ืื ืื ื ืืฆืืืืื. ืืื ืื ืื ื ืืืืื ืืฆืืจื ืืื ืืจืืช.
โ
ืขืื ืืืงืจืื, ืขืื ืืืงืจืื, ืขืื ืืืืคืงื.
โ
ืืื ืืืืืื ืืช ืืืืืคืงื ืคื ืขืฉืจื, ื ืฆืืจื ืืืืืจื ืืืืื ืื ืื ืื ื ืคื ืขืฉืจื. ืืื ืืืืื ืืื ืขืืื ืืฉื ืืช ืืช ืื ืื ืฉืืืืื ืืืืื - ืืช ืืืืืืจื, ืืชืจืืืช, ืืืคืืื ืืช ืืืจื ืฉืื ืื ืื ื ืืืืจืื ืืช ืืืืงืจืื ืฉืื ื.
โ
ืืื ืขืื ืจืขืืื ืืืจ.
โ
ืืืงืื ืืืืื ืคื ืขืฉืจื - ืืื ืฉืื ื ืขืืืจ ืืืงืื ืขืื ืขืฉืจื ืืืื ืื ืืืื ื?
โ
ืืืื ืื ืขืฆืืืืื, ืขื ืืืงืจืื ืืฉืืื, ืื ืืื ืืฉืืื ืืชืจืืืช ืืฉืืื. ืืื ืขื ืืืชื DNA - ืืืงืจ ืืืขื ืืขืืื, ืฉืื ืชืืืืชื ืืฉืคืจ ืืช ืืจืืืืชื ืฉื ืื ืฉืื.
โ
ืืืื ื ืื ืก ืืชืืื ื ืืื ืืจืคืืื ืฉื ืฆื"ื.
โ
ื ืคืืฉื ื ืขื ืืื ืฉืื ืฉื. ืกืืคืจื ื ืขื ืื ืฉืื ืื ื ืขืืฉืื ืืขื ืืคืืืืกืืคืื ืฉืืืืืจื ืืืืื. ืืฆืขื ื ืืช ืืืืข, ืื ืืกืืื ืืืืืืืืช ืฉืื ื ืืื ืืขืืืจ ืืืงืื ืืืื ืืืงืจ ืืืื ืืชืื ืืื ืืจืคืืื.
โ
ืืื ืืื.
โ
ืืืชืืื ืืืืืื ื ืืช ืืืงืื ืืงืจืื. ืืืืฉื, ืืืืื ืืชืืื ืืขืืื ืขื ืืจืืืืื ืืขืฆืื.
โ
ืืืืื?
โ
ืืืื ืืจืคืืื ืืฉ ืืืื ืืืงืจ ืขืฆืืื, ืฉืขืืฉื ืขืืืื ืืืืืื.
โ
ืื ืื ื ืขืืืื ืืฉืชืคืื ืคืขืืื ืืืืืืื ืืืงืจืื, ืืื ืื ืืืจ ืื ืืืืื ืฉืื ื. ืืื ืฉืืื. ืขื ืื ืฉืื ืืฆืืื ืื ืืฉืืื, ืจืขืืื ืืช ืืฉืืื ืืืืฉืืื ืืฉืืื.
โ
ืืื ืืืืืง ืื ืฉืจืฆืื ื ืฉืืงืจื.
โ
ืืื, ืืคื ื ืืื ืืืื, ืืืืข ืืืื ื ืืืชื ืืคืชืืข.
โ
ืืืฉืื ืืืื ืืจืคืืื ืืืืื, ืืืืืืชื, ืืืขื ืืง ืืืืื ืฉืื ื ืืืช ืืืงืจื ืืืขื ืงืฆืื ืืจืคืืื ืืจืืฉื, ืขื ืชืจืืื ืืืฆืืช ืืืคื ืืืืงืจ ืืจืคืืื ืฆืืืืช.
โ
ืื ื ืจืฉืื ื ืืชืืจืืช. ืื ืืืงืฉื ื ืคืจืก. ืื ืืืฉื ื ืืืขืืืืช.
โ
ืืืฉืื ืคืฉืื ืืืืื ืืืืื ืชืืื.
โ
ืืฉ ืืื ืืืืื ืืคื ืืขืืจืืช - ืืืืช ืืื ื.
โ
ืื ืืจืืฉ ืืืชื ืืคืืื ืืืชืจ ืืืคืจืก ืขืฆืื.
โ
ืื ืืขืื ืื, ืืฉ ืืื ืืืืื ืืืฉืื ืฉืื ืื ื ืืืืื ืื ืื ืืืื:
โ
ืืืจื ืืขืฉืืช ืืืืคืงื ืขืฆืื ืืื ืื ืืืืจื ืืื ืืช ืืช ืืืจืืื ืืื ืืืื, ืืื ืืขืืืจ ืืืืจืื ืืืฆืืื.
โ
ืืืฉืื ืืฆืืืืื, ืืืืืคืงื ืืืฉืื ืืืืื, ืื ืืื ืฉืื ืื ื ืฆืจืืืื ืืืืืช ืืขืืจืืื ืืื ืืืจ.
โ
ืื ื ืจืืฆื ืืืืืืช ืืขืืืง ืืื ืืื ืืื ืฉืื ืืืืคืืืื ืืืืื KI, ืืืืืืื ืืื ืฉืืืืืื ืืช ืฉืืชืืฃ ืืคืขืืื ืืื, ืืืื ืฉืื ืืืฆืืื ืื ืืืื ืืจืคืืื ืฉืืคืื ืืช ืืืืื ืืืฆืืืืช.
โ
ืืื ื ืจืืฆื ืืกืืื ืืืืฉืื ืืืช:
โ
ืืืื ืืืื ืืื ืืคื ืืืฆืืื ืฉืื ื ืืืืื ืืื ืื ืจืง ืื ืฉืืฆืืื ื ืืขืฉืืช ืืขืฆืื ื, ืืื ืืื ืืืจืื ืืืืื ืืืฉืืื ืืงืจืืช ืืขืืื ืืืืืชื ื - ืื ืืืขืืื ื.
๐๏ธ Uri Goren retweeted
What if your annual physical became a continuous physical?
Wearables + AI could turn everyday health data into insights your care team can act on, between visits.
An 80-second glimpse from my conversation with Moira Gunn on @BiotechNation ๐
ืืฆื ืืื ืื ืืืฉื ืืืชื ื ืชืคืก. ืืฆื ืฉื ื ืื ืจืืฆื ืืืืืช 150 ืฉื ื
Prepare to live to 150, I told Brian Yennis on @FoxNews @SatAmericaFNC. Our IPF drug rentosertib showed age reversal of 3-6 years on aging-clock biomarkers in a 12-week Phase 2a trial of 71 patients. Early data, Phase 3 ongoing, nothing approved. Insilico, AI-designed. #ARDD2026
ืืชืืจ ืจืืคื ืื ื ืืืืืฅ ืืืืื:
1. ืืืชืืกื ืืฉืคืขืช (ืืื ืงืฉื ืืฉื ื ืืืืกืื ืืคืืืช ืกืืืื ืืืืืฉื ืืื ืื ื ืืืงืื ืืคืืืช ืกืืืื ืืืืื ืงืฉื)
2. ืืืืช ืืืืฉ ืืช ืืืืชืื ืืืืืงืจืืืช ืืืืฆืืืข
ืจืง ืืจืืืืช ืืืืื
Vibing advocacy and support
We get to see many beautiful use cases at @Base44 that make us wake up every day with a feeling that weโre doing something that is very net positive in the world.
For me - this is one of the top ones -
Dotan Haas Hershkovitz built โCheck it outโ - after losing her mother to cancer - to help women get checked for breast cancer.
Some great stuff inside:
- Find a doctor that fits you, by location and how they approach care
- WhatsApp or email reminders to get checked once or twice a year
- Matches you with another woman - if you don't want to go to the appointment alone
- Info on self exams and a risk questionnaire
Since publishing the app she has had thousands of women using it - getting checked, helping each other with info and going to checkups together.
Some of them found breast cancer early because of it. Which is hard to overstate how important that is for effective treatment.
I love every part of this story.
We will do our best to support Dotan, and other builders building for good.
ืื ืคืฉืื ืืืืจืฃ ืื ืฉืงืืจื ืืชืืื ืืืื ื ืืืืืืืชืืช, ืืืืจืื ืื ืืื ืืฉืื ืืืฉ
ืืณืง ืืืจืกื, ืืืืกื ืืืืืืจ, ืืฆืื ืืช Dessa, ืืืื AI ืืืฉ ืฉื ืืืจืช Block, ืืื ื ืืืฉื ืฉืื ืืื ืืคืืชืืืื ืืืขื ืืื ืื ืฉืจืืืชื ืืืืจืื ื
ืื ืฉืืืืื ืืื ืืื ืฉืืืืื ืืืื ืื ืืืื ืขื ืืงืกื, ืืื ืื ืืจืง ืขื ืฉืืืืช ืื ืืฉืืืช
ืืื ืืกืืื ืืืงืฉืื ืืืืืจ ืื ืืื ืืช, ืืืืื ืืื ืืืืืจ, ืืืืื ืืฆืืืง, ืืืื ืืก ืจืืฉ ืืฉืืื ืืืคืืื ืืขืฆืืจ ืืืืฆืข ืืฉืคื ืืฆืืจื ืืืขืืช (ืงืจืืคื ืฉื ืืืืื)
ืื ืื ื ืขืืืจืื ืืืืื ื ืฉืืืื ืฉื ืกืืื ืืืืจ ืืื ืืขื ืืช ืื ื, ืืืืื ื ืฉืืกืืืืช ืื ืื ืืืชื ื ืฉืืื ืืื ืื ืืื
ืื ืื ื ืืชืงืจืืื ืื ืงืืื ืฉืื ืืืจ ืืืื ืงืฉื ืืืืืื ืื ืื ืื ื ืืืืจืื ืขื ืืื ืื ืขื AI
ืื ื ืืืฉืืช ืื ืืื ืชื ืืืชืืื ืื ืื ืืืืื ื ืืื ืืืื
ืืื ืกืคืง ืฉืื ื ืืืจ ืจืืื ืืื ืฆืจืื ืืคืชื ืืืจืืช ืชืืื ื ืฉืืืขื ืืืืืช ืื ืืืืืจ ืืืื-ืืื
ืื ืื ืื ืืืืช ืืกืงืืืจืื ืืืื ืฉืื ืฉืืืคืื ืืืืื ืชืืื ืื ื ืืจืืืืช
We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a personโฆfull of emotion, nuance, pauses, and a natural cadence.
block.xyz/dessa
๐๏ธ Uri Goren retweeted
A PCORI-funded study looked at whether web-based tools could help families manage their childโs asthma. Researchers compared outcomes for children whose families used a standard web-based tool to track asthma symptoms and attacks, an enhanced version of the tool or no tracking tool.
The study found children whose families used either version of the online tool had better asthma symptom control and quality of life. They also missed fewer school days and had fewer ER or hospital visits.
Read more: pcori.me/4yWv9Zt
๐๏ธ Uri Goren retweeted
Your doctor cuts you off about 11 seconds after you start talking. Researchers literally timed it in a 2018 study. That number is why Google's new Lancet result is bigger than it looks.
In the study, 98 real urgent care patients texted with Google's AI days before their appointment. No scans, no exam, no bloodwork, just an unhurried conversation about what was wrong. That alone put the correct final diagnosis on the AI's list 90% of the time.
Sounds like magic until you learn what medicine has known for 50 years.
Back in 1975, a British team published a study in the BMJ tracking which part of a visit actually produced the right diagnosis. The physical exam? The labs? Nope. In 82% of cases, the patient's story alone got the doctor there. Everything else mostly confirmed what listening had already found.
So the playbook for accurate diagnosis has existed since 1975. Let the patient talk. Medicine just built a system where nobody has time to run it, and an 18-minute appointment with 11 seconds of uninterrupted patient talk was the result.
Google didn't build a superhuman doctor. It built the first clinician in 50 years with enough time to use the method we already proved works.
And my favorite detail buried in the study... physicians who read the AI's pre-visit summary said it changed how they treated the patient in over half of cases. Hand a doctor the full story before you walk in the room and you get different medicine.
Turns out the most advanced medical instrument Google ever shipped is a conversation that doesn't get interrupted.
Today @TheLancet published our AMIE study with @BIDMC_Medicine, the first prospective study of patient-facing conversational diagnostics in a real clinic.
AMIE is a research system people can chat with before doctors appointments. Testing AMIE outside the lab with real urgent care patients showed how technology can help gather medical histories safely. Clinicians reported that the summaries helped them prepare for visits in 75% of cases and influenced their approach to care in more than half. AMIEโs differential diagnoses also matched the doctorsโ final diagnoses 90% of the time giving clinicians more time to focus on patient care.
Itโs Googleโs first-ever paper in The Lancetโs main journal!
Cost and speed are the small prizes here , yet sovereignty is the big one. A pharma renting frontier intelligence is strategically exposed: access, pricing, and roadmap decided elsewhere, with its most sensitive data crossing a boundary it doesn't control. Meanwhile every major pharma already owns the one asset frontier labs can't replicate : decades of proprietary screens, assay readouts, and failed trials. Fine-tune open weights on that data and you own the whole stack: weights, inference, IP. That's AI sovereignty at company scale. The first pharma to internalize this compresses discovery timelines by years. The rest end up renting intelligence about their own molecules.
We were able to fine tune an open weight model (Qwen3.5-9B) on proprietary internal data at @coinbase, and it out performed frontier models, at much lower cost and speed, for a specialized task (fraud prevention in this case).
I think we'll see more of this pattern inside companies. The frontier models are amazing generalists. But increasingly you'll see specialist models out perform on more narrow tasks, either working on teams, or by themselves.
We're making it easy to fine tune models internally for our employees. We even have a skill for it in our internal harness that makes it easy to spin these up and start training them:
"/fine-tune-and-serve-llm I want to fine-tune Qwen for payments risk review. help me get started"
I imagine we'll have some backend services that could be replaced entirely by this new architecture.
๐๏ธ Uri Goren retweeted
Is it over for human doctors?
No. Airliners can largely take off, fly, and land themselves. We still put two pilots up front, for judgment, edge cases, and trust. Same in the exam room.
My take on Washington's AI-in-medicine bet, on @NewsNation with @jessecordweber
๐๏ธ Uri Goren retweeted
$1.8 billion committed for predictive models of AI biology
@biohub @GoogleDeepMind, @ENERGY , @NIH, @Meta @IsomorphicLabs all came together
reuters.com/business/healthcโฆ
Congrats!
Karl Deisseroth awarded 2026 Nobel Prize in physiology or medicine ๐ฌ
The professor of bioengineering and of psychiatry and behavioral sciences shares the prize with two other researchers for their work in optogenetics.
news.stanford.edu/stories/20โฆ
I joined OpenEvidence!
At Anthropic, I created Claude Science, a product built so that scientists' curiosity, not their capacity to execute, sets the limit on what they can discover. It takes on the heavy lifting across the early stages of the discovery and translational pipeline and scales what a single researcher can carry out. Building Claude Science has been the most fulfilling work of my career so far.
Now I'm excited to help build the future of AI in medicine at OpenEvidence, bringing that same leverage to doctors at the point of care. OpenEvidence already reaches the majority of patients in America. It doesn't get more translational than that.
๐๏ธ Uri Goren retweeted
The company Everlywell just announced the nationwide launch and commercial availability of its AI-driven software that can assess a woman's five-year risk of developing breast cancer from a routine screening mammogram!
I started wondering how much time it took them to get here. According to my research, here it is:
August, 2017 โ Earliest related breast-AI patentย
November, 2018 โ Mammography-based 5-year breast cancer risk prediction study manuscript received.
May, 2019 โ Study published online, demonstrating deep learning from mammograms to predict future breast cancer risk.
2019 โ Company later known as Clairity incorporated as Boston Health AI.
2020 โ Clairity describes itself as founded.
September, 2024 โ FDA De Novo submission for Allix5 / Clairity Breast.
May, 2025 โ FDA authorizes Clairity Breast.
October 1, 2026 โ Everlywell announces nationwide access to the FDA-authorized Clairity breast cancer risk assessment.
Incredible how much work goes into every single AI-based technology used in practice!
Source: businesswire.com/news/home/2โฆ
๐๏ธ Uri Goren retweeted
Thanks @Forbes for featuring the great work weโre doing @ChronicleBioAI.
There are exactly 0 FDA-approved therapies for the hundreds of millions of people suffering from diseases like POTS, Long COVID, MECFS or hEDS. We created ChronicleBio to change this.
By building the deepest dataset on these conditions and leveraging AI, we will understand what the biology behind the symptoms is. From there, we can map the right drugs to the right patients.
Sufferers of complex, chronic diseases like POTS and long COVID-19 often spend years searching for relief. This former OpenAI exec started ChronicleBio to find treatments using AI and lots of biological data.
Read more: forbes.com/sites/amyfeldman/โฆ
๐ธ: Joel Saget/AFP via Getty Images






























