🩻 AI that reads a CT scan in 3D—and explains its findings

NVIDIA, the NIH’s National Cancer Institute, and the University of Zurich have released NV-Reason-CT, an open model that analyzes full 3D CT volumes and generates reports with explanations.

⚙️ How it works
The model pairs the Qwen3.5-4B language model with Primus, a 3D visual encoder. Each scan becomes 13,824 visual tokens, passed to the language model without further compression. Three-dimensional positional encoding preserves spatial information, helping it distinguish, for example, a finding in the right kidney from one in the left.

📚 How it was trained
Training used 550,000 examples from 70,111 CT volumes. Supervised fine-tuning on radiologists’ analyses was followed by reinforcement learning, with rewards for correctly identifying abnormalities and following the required report structure.

📊 What the results show
On CT-RATE, NV-Reason-CT achieved an average precision of 0.614 across 18 abnormality categories, compared with 0.581 for VoxelFM and **0.398 for CT-CLIP**—without a separate classification head.

In a pilot study with radiologists, scan review and reporting time fell from 26.25 to 13.13 minutes: roughly half.

🧩 Part of a broader medical AI toolkit
NVIDIA’s open medical model family also includes:
• NV-Generate-CTMR — generates synthetic 3D CT and MRI volumes.
• NV-Segment-CTMR — segments organs and lesions.
• NV-Reason-CXR — analyzes chest X-rays.
• NV-Reason-CT — analyzes full 3D CT scans.

🔓 Weights and code are available under OpenMDW-1.1, alongside fine-tuning and reinforcement-learning examples and a web demo.

Promising early results for AI-assisted radiology—with the time savings demonstrated so far in a pilot study.

@science

⚛️ Scientists Watched Matter “Appear” Inside a Quantum Computer

Pull two quarks apart and something deeply strange happens.

You never actually get two isolated quarks.

Instead, the energy binding them grows — almost as if an invisible string were being stretched between them. Eventually, storing more energy in that string becomes so expensive that nature takes another option:

it creates a new particle–antiparticle pair.

The string breaks.

This process, called string breaking, is fundamental to the strong nuclear force and probably played an important role in how matter evolved in the extremely hot early universe.

But calculating its real-time quantum dynamics is extraordinarily difficult.

So researchers from Duke University, the University of Maryland, Oxford, Caltech, Cornell and KU Leuven built a miniature analogue of the problem inside a quantum machine.

They programmed a chain of 13 trapped ytterbium ions to behave according to a simplified lattice gauge theory.

The ions were not literally turned into quarks.

Instead, their quantum states encoded the particles, fields and “string” connecting them — allowing researchers to watch the simulated system evolve with both spatial and temporal resolution.

And then the string broke.

New effective particle pairs appeared and propagated through the simulated system, reproducing the essential quantum dynamics physicists wanted to study.

But the experiment also produced a surprise.

The conventional expectation was that particle pairs would spontaneously appear throughout the string through a process related to the Schwinger mechanism.

Instead, the researchers observed pairs forming preferentially near the two ends of the string, then spreading inward.

Their calculations indicate this is a distinct, previously unobserved mechanism for dynamical string breaking.

This is not a simulation of the full Standard Model, and no real matter was created inside the computer. The experiment used a simplified 1+1-dimensional Z₂ gauge theory, and today’s classical computers can still reproduce a system this small.

The real prize comes later.

As quantum simulators grow, they could attack versions of these problems that conventional supercomputers cannot efficiently calculate — potentially letting physicists experimentally explore the quantum dynamics of particle collisions and conditions resembling the universe shortly after the Big Bang.

We built computers out of quantum mechanics.

Now we’re beginning to use them to ask quantum mechanics how the universe built matter.

#QuantumComputing #QuantumPhysics #ParticlePhysics #BigBang #Physics #Science

https://doi.org/10.1038/s41567-026-03422-0

🌌 The Ingredients for Planets Were Spreading Through Space Just 500 Million Years After the Big Bang

The newborn universe started simple.

After the Big Bang, almost everything was hydrogen and helium. Carbon, oxygen, silicon and nearly every other element needed to build planets — and eventually us — had to be manufactured later inside stars.

Astronomers expected that process to take time.

JWST has now shown that it happened remarkably fast.

Researchers analyzed nearly 30 hours of Webb observations of three galaxies seen as they existed roughly 500–700 million years after the Big Bang.

They found unmistakable chemical fingerprints of carbon, oxygen and silicon in gas associated with the galaxies.

But the really interesting part was where that gas was going.

The absorption signatures were blueshifted by roughly 50–250 km/s, indicating that metal-enriched material was moving outward from the galaxies — consistent with powerful galactic winds carrying newly forged elements into surrounding space.

That means an entire cosmic recycling system was already operating while the universe was only about 3% of its present age.

Stars formed.

They forged heavier elements.

Stellar winds and explosions returned those elements to their galaxies.

And galaxies began spraying them outward, chemically transforming the surrounding universe.

Remarkably, the chemical fingerprints look similar to those seen around galaxies billions of years later.

The result may also help solve another mystery.

Astronomers have spent decades searching for Population III stars — the hypothetical first generation of stars, made almost entirely from pristine hydrogen and helium.

None has ever been conclusively found.

If early galaxies contaminated their surroundings with heavier elements this quickly, the window in which truly pristine stars could form may simply have been much shorter than expected.

Important caveat: the result comes from only three unusually bright early galaxies. We don’t yet know whether such rapid enrichment was universal across the young cosmos.

Still, the implication is striking.

Only half a billion years after the Big Bang, the universe had already started distributing the carbon in our bodies, the oxygen in our water and the silicon beneath our feet.

Cosmic chemistry apparently wasted very little time.

#JWST #Space #Astronomy #Cosmology #BigBang #EarlyUniverse #Science

https://www.nature.com/articles/s41550-026-02988-2

⚡️ GPT-6 Astra deciphered a letter to a Napoleonic marshal that nobody had been able to read for 217 years

Researcher Carter Church used GPT-6 Astra to decipher an encrypted letter to Marshal Auguste de Marmont, one of Napoleon's generals. The letter is dated March 1809, and the model took about six hours to do the whole job.

The first step was simply to extract readable text from a poor scan of the manuscript. Astra recognised 1,300 cipher characters, among which there turned out to be 155 distinct signs.

The model then found a published partial key by French cryptology historian Daniel Tant: 33 letters covering roughly 435 characters. For the remaining signs it wrote a simulated-annealing solver, and then checked the result against historical correspondence and corrected the document's date.

Inside was a military briefing from Eugène de Beauharnais's headquarters: troop positions and Austrian movements on the eve of Austria's invasion in April 1809. Marmont is told not to fear "a few detachments or a gathering of rabble." It also turned up the ending of a sentence that breaks off in Napoleon's memoirs, published in 1865.

The solution was reviewed by Satoshi Tomokiyo, who runs the historical ciphers site Cryptiana, and the cipher is now listed there as solved.

https://runtimewire.com/article/gpt-6-astra-marmont-cipher-carter-church

https://x.com/Machinelearrn/status/2105593763582107849

Meanwhile, something interesting is happening on Telegram: @gadget is officially up for auction.

Yes — the actual @gadget username.

Telegram usernames can be turned into blockchain-based collectibles and traded through Fragment. Whoever wins the auction gets control of the handle and can assign it to a Telegram account, channel, group or bot.

And @gadget is exactly the kind of digital property that could be valuable: short, memorable, universally understandable and sitting right in the middle of the global tech industry.

It’s a strange new category of internet real estate — not a domain name, not quite an NFT, but a piece of identity infrastructure inside a platform used by more than a billion people.

Let’s see what the market thinks @gadget is worth.

https://fragment.com/username/gadget

🔬 An AI Scientist Just Made a Discovery by Running Its Own Lab Experiments

We’ve seen AI write scientific papers.

We’ve seen it predict proteins and search enormous biological databases.

This is different.

Researchers built a closed-loop AI scientist connected to a physical laboratory. It can generate a hypothesis, design an experiment, turn that experiment into instructions for laboratory automation, analyze the resulting data — and then decide what to investigate next.

The system was given knowledge about Saccharomyces cerevisiae — ordinary baker’s yeast — including roughly 60,000 known biological relationships involving its metabolism, physiology and phenotype.

From these, it generated 1,933 testable hypotheses about how different compounds might affect yeast growth under stress.

Then came the important part:

the hypotheses met reality.

The system selected experiments and controls, converted them into machine-readable laboratory procedures and analyzed the resulting biological data. Some predictions worked.

Others failed.

And one failure produced the most interesting result.

The AI initially predicted that glutamate might protect yeast from formic-acid stress.

The experiment contradicted it.

Instead of simply recording “wrong,” the system analyzed the new metabolomic data, searched for another explanation and identified aminoadipate, a molecule involved in lysine metabolism, as a candidate.

It formulated a new hypothesis.

The lab tested it.

And aminoadipate did improve yeast growth under formic-acid stress — by about 7% for each millimolar increase in the experiment. The researchers report this as a previously unknown protective interaction.

There is an important caveat.

This was not a completely autonomous robot scientist. Humans defined the research domain and safety boundaries, moved some physical samples between instruments and supplied the overall experimental infrastructure. The biological questions were also relatively narrow yeast-metabolism problems — not Nobel-level discoveries.

But something important has happened.

AI has already become very good at generating hypotheses from existing information.

Now the loop can close:

Hypothesis → physical experiment → unexpected result → new hypothesis → new experiment.

That is no longer just AI analyzing science.

It is AI participating in the scientific method.

What happens when systems like this can run 10,000 experiments while a human scientist sleeps?

#AI #Science #Biology #Robotics #Biotechnology #Automation #Research

https://doi.org/10.1098/rsif.2026.0043