Gregory Bassen had a simple question: could a large language model write a recipe for making a ceramic oxide that actually works in a real lab? Not in theory. Not in simulation. In a furnace, with real powders, producing a material you could hold in your hand.
In 2023, the world spent over $200 billion on antibody drugs — treatments like Keytruda for cancer and Humira for autoimmune disease. But finding a single antibody that binds well enough to become a drug is like searching for a specific grain of sand on a beach. A typical drug discovery campaign screens thousands of candidates, each one requiring expensive lab tests to measure how tightly it grabs its target. Most fail.
The European Centre for Medium-Range Weather Forecasts has spent decades running the world’s most sophisticated weather simulations on some of the planet’s largest supercomputers. Their model, the Integrated Forecasting System, crunches through the laws of physics hour by hour to predict what the atmosphere will do next. Last week, ECMWF announced that a machine learning model using 200 times less energy can now match it. In the weeks-ahead range that matters most for disaster planning, it can beat it.
That model is called AIFS-SUBS, and it is ECMWF’s first AI system designed specifically for sub-seasonal forecasting: the awkward middle ground between next week’s weather and next season’s climate. Meteorologists call this the “predictability desert.”
Imagine you’re a junior researcher sitting in a promotion committee meeting. Your department chair, who controls your career trajectory, strongly believes a certain candidate should be promoted. You have serious reservations about the candidate’s record. When the chair turns to you and asks for your opinion, what do you say?
Now imagine the same scenario, but you’re speaking to a confidential journal that nobody else will ever read. Would your answer change?
This is not a thought experiment about human psychology. It’s what a team of researchers from Carnegie Mellon University and independent labs actually did with AI language models, and what they found should give anyone deploying AI agents in professional settings serious pause.
Electrochemical impedance spectroscopy (EIS) is one of the most widely used techniques in battery labs, fuel cell research, and corrosion science. It is elegant in principle: apply a small alternating voltage across a cell, measure the resulting current, and extract a spectrum that encodes the system’s internal dynamics. In practice, however, EIS spectra are notoriously difficult to interpret at the molecular level. Researchers typically fit them to empirical circuit models — collections of imaginary resistors and capacitors — but the connection between those abstract circuit elements and what ions are actually doing inside the electrolyte remains frustratingly opaque.
A new paper from researchers at the University of Cambridge, Durham University, and Sorbonne Université, published on July 2, 2026, takes a significant step toward bridging that gap. Led by Connie Fairchild, Stephen Cox, Benjamin Rotenberg, and Thomas Sayer, the team proposes an alternative framework rooted in statistical mechanics that extracts physically meaningful parameters from impedance data — parameters that directly report on the molecular-scale motions of ions. Their approach, which combines molecular dynamics simulations with the “itinerant oscillator” model, offers a path toward rational design of next-generation electrolytes for batteries and supercapacitors.
The work tackles a fundamental problem: the standard Debye-Falkenhagen theory, which describes how ions move in response to an electric field, was developed for dilute electrolyte solutions. It assumes ions are largely independent, each surrounded by a diffuse “cloud” of counter-ions that distorts under an applied field. But modern energy storage devices increasingly rely on concentrated electrolytes — ionic liquids, which are essentially molten salts at room temperature, or highly concentrated “water-in-salt” formulations. In these systems, ions are packed so tightly that every motion is collective. There is no dilute cloud; there is a cage.
Inside every living cell, enzymes are running the chemistry of life, snipping, splicing, and assembling molecules with a precision that industrial chemists can only dream of. A single enzyme can catalyze a million reactions per second. The problem? Understanding how they do it has been painfully slow. Even with GPU-accelerated quantum chemistry, simulating a single step in an enzymatic reaction on a cluster of 300 atoms can take hours. Mapping a full reaction pathway takes days.
Now researchers at MIT have built something that changes the math entirely.
Heather Kulik’s group released Enerzyme, an open-source framework that trains neural networks to simulate enzyme catalysis with near-quantum-chemical accuracy, but at a fraction of the cost. The key breakthrough isn’t a better algorithm. It’s that Enerzyme handles everything that makes enzymes uniquely hard to model: their size, their solvent environment, and the complex charge transfers that drive their chemistry.
Neutrinos are the ghosts of the particle world. They pass through planets, through stars, through your body right now — trillions of them every second — without so much as brushing against an atom. They have nearly no mass, no electric charge, and an almost pathological disinterest in interacting with anything. You would be forgiven for wondering why physicists spend billions of dollars chasing them.
But here is the thing about ghosts: when you catch one, it can tell you something profound. And for the past three decades, neutrinos have been telling particle physicists that something is wrong with the Standard Model. Not catastrophically wrong — the Standard Model is still one of the most precisely verified theories in science — but wrong enough to hint at a deeper structure beneath. The discovery that neutrinos oscillate between flavours as they travel, which earned Takaaki Kajita and Arthur McDonald the 2015 Nobel Prize, was the first clear crack in the edifice. Now, a team of physicists from the Institute of Physics at the University of Bonn and the Department of Physics at the University of Calcutta has used eight years of data from IceCube DeepCore — a detector buried two kilometres under Antarctic ice — to deliver the most stringent test yet of a closely related prediction: does neutrino mixing obey the rules of unitarity?
The answer, for now, is yes. But the precision with which they have confirmed that “yes” is where the real physics lies.
There is a quiet schism running through modern software engineering. On one side sit the crisp, deterministic functions that computers have always excelled at: sort this list, invert that matrix, validate an email address against a regex. On the other side sits an ever-expanding category of tasks that developers intuitively know how to describe but cannot cleanly code — “flag the important log lines,” “fix this malformed JSON,” “rank these search results by user intent.”
Every year, computational chemists publish lists of new materials that, according to their calculations, should be stable. The structures are plausible. The energies are negative. The phonon spectra show no imaginary frequencies. The papers get published. Then nothing happens.
There is a reason for that silence. The gap between a theoretically stable crystal and something you can actually synthesize in a lab — something that survives on a benchtop, in air, at room temperature — is wide enough to swallow entire classes of predicted materials. A material predicted by density functional theory may be stable at zero Kelvin in vacuum, but fall apart the moment a postdoc tries to grow it on a substrate.
Now a collaboration between Toyota Research Institute, Toyota Central R&D Labs, and the University of Tokyo has built a framework that might finally close that gap. Their approach, described in a preprint posted to arXiv on July 2, takes a list of 894 computationally stable materials and winnows it down to 25 that are actually worth trying to make in a lab.
Every year, millions of cancer patients lie inside the narrow tube of an MRI machine while a nurse injects gadolinium into their veins. The contrast agent makes tumors glow on the scan, which helps surgeons find them and track whether treatment is working. But gadolinium has a problem: it accumulates in the body. The brain, the bones, the kidneys. The metal stays there, sometimes permanently. Some patients have developed nephrogenic systemic fibrosis, a condition where connective tissue grows uncontrollably. A safer alternative has been the goal of contrast-agent research for a decade.
Here is where it gets interesting. A team of researchers at Islamic Azad University in Tehran has compiled the evidence for a material that might replace gadolinium entirely, and it does more than just light up tumors on a scan. A single injection of this stuff could make your cancer visible on MRI, deliver therapy directly to the tumor, and train your immune system to attack it, all at the same time.