Hi Reader, China approved what is being described as the world’s first commercially available invasive brain-computer interface for medical use. The coin-sized device is implanted on the brain’s outer covering and decodes neural signals to control a robotic glove, helping some people with spinal cord injuries regain limited hand movement. It’s a small step in scale but a big shift in what’s now possible, a common theme in this week’s top stories.

Coming up this week:
🥧 Calculating 314 trillion digits of 𝜋
🚗 The most dangerous road trip in physics history
🐁 What mice’s brains reveal about our vision
+More

COMPUTER SCIENCE

Engineers Calculated Pi to 314 Trillion Digits for Pi Day

To mark Pi Day, a team at StorageReview released 314 trillion digits of pi. That’s 130 terabytes of data.

For context, NASA says 37 decimal places are enough to calculate the circumference of the observable universe to within the width of a single hydrogen atom. Every digit after that is, for physical measurements, effectively pointless. So why did a team of engineers just spend 110 days computing pi to 314 trillion decimal places?

They weren't "just" computing pi. They were stress-testing the outer edge of what modern hardware can do. The team ran their calculation on a single 2U server - a Dell PowerEdge R7725 with dual 192-core AMD EPYC processors and 40 high-capacity NVMe SSDs - and surpassed a previous 300 trillion digit run by Linus Media Group. They didn’t use a cloud cluster but rather used one box, running flat-out for nearly four months.

What's interesting is where the real bottleneck turned out to be: storage. At this scale, the computation generates so much intermediate data that the limiting factor is simply how fast you can read and write enormous files without the system choking. StorageReview's secret was connecting those 40 drives directly to the processors via high-speed PCIe lanes, reducing many of the usual bottlenecks and capable of up to ~280 GB/s of aggregate bandwidth. The total data generated, including intermediate files and checkpoints, exceeded 2 petabytes.

As StorageReview's Kevin O'Brien wrote: “The storage layer, specifically, is where this record was actually won.”

The same lesson applies beyond calculating pi’s decimal places. Any workload that has to sustain massive, continuous input and output processes - AI training, climate simulations, genomics - hits the same wall. Pi is one of the cleanest possible test cases: a calculation with no noise, no ambiguity, and a result you can verify with certainty.

TL;DR: StorageReview computed pi to a world-record 314 trillion digits on a single server over 110 days. The real breakthrough wasn't the math itself but rather showing that at this scale, computing is a storage problem, with lessons that carry directly into AI, scientific simulation, and any data-heavy workload.

PHYSICS

The Most Dangerous Road Trip in Physics History

Somewhere at CERN, there's a wardrobe-sized device chilled to -269°C, threaded with electric and magnetic fields. It’s designed to keep ~1000 particles from touching anything while riding in the back of a truck.

That device is called BASE-STEP, and this month, physicists plan to load it with antiprotons and drive them around the CERN campus. It's a 20-minute journey. If it works, this will be the first time trapped antiprotons will be transported outside their production area for precision experiments.

The need for the road trip comes down to noise. CERN's Antimatter Factory is the only place that routinely produces and traps low-energy antiprotons for precision study. But it's also one of the most magnetically chaotic environments imaginable. That noise puts a hard ceiling on experimental precision. So BASE collaboration physicists Christian Smorra and Stefan Ulmer came up with a bold solution: just take the antimatter somewhere quieter.

BASE-STEP keeps antiprotons in a near-perfect vacuum, with cryogenic temperatures helping residual gas molecules stick to surfaces before reaching the trap. Electric and magnetic fields suspend the particles at the centre of the chamber, never touching the walls. It's engineered to withstand accelerations of up to 2g while the superconducting magnet stays cold and live. A liquid helium buffer buys four hours of autonomous operation if power fails. The worst case, as Smorra puts it: the antiprotons annihilate and you drive back to refill.

The team already proved the concept in October 2024, transporting 70 ordinary protons on a 4-kilometre loop of the CERN site without losing a single one.

The destination is Heinrich Heine University in Düsseldorf, where a magnetically calm lab could enable measurements up to 100 times more precise than current limits at CERN. Those measurements would probe one of physics' deepest unsolved problems: why the Big Bang produced a universe made almost entirely of matter, when equal amounts of antimatter should have been created alongside it.

For now, the truck just needs to make it around the block.

TL;DR: CERN's BASE team is preparing the first-ever transport of antimatter, loading antiprotons into a cryogenic portable trap for a truck journey around the CERN campus. If successful, the next stop is a lab in Germany, where 100x more precise measurements could help explain why the universe is made of matter at all.

NEUROSCIENCE

What Mice’s Brains Reveal About Our Vision

In a lab at UCL's Sainsbury Wellcome Centre, a mouse watched a video. Scientists tracked the mouse’s brain cells one by one using a microscope sensitive enough to detect tiny spikes in calcium levels as each neuron lit up. Then they fed those signals into an algorithm and “recreated” the new video clip the mouse had been watching - starting from a blank screen, without ever showing the algorithm that clip. The reconstruction wasn't perfect. And that imperfection turned out to be the most interesting result.

The method works by a process of elimination. The team trained seven versions of a neural encoding model using a blank grey screen as a baseline, then calculated the difference between the neuronal activity a mouse would show watching nothing, and what it actually showed watching a real video. That gap - the brain's response to the content itself - became the signal. The algorithm then updated the blank screen pixel by pixel, frame by frame, until the predicted neural activity matched what the mouse had actually produced.

The real test came when the researchers showed five mice a brand-new 10-second clip the model had never seen. Working only from the animals' neuronal activity, the algorithm reconstructed the video from scratch. The best reconstructions achieved a pixel-level correlation of about 0.57 (on a scale from 0 to 1) with the original. For comparison, previous reconstructions of static images from awake mouse V1 reached about 0.24 over a similar retinotopic area.

The best reconstructions achieved a pixel-level correlation of about 0.57 with the original. For comparison, previous reconstructions of static images from awake mouse V1 reached about 0.24 over a similar retinotopic area.

The results are still blurry. But the timing is strikingly accurate, and quality improved steadily as more neurons were included, suggesting that visual information is distributed across neural populations, and that reconstruction quality is strongly limited by how much neural activity can be measured and modeled.

What's more interesting is what the mismatches reveal. The reconstructed videos don't perfectly match the originals and that gap is the point. As lead author Dr. Joel Bauer puts it, the brain's deviation from reality “is not necessarily an error but a feature, reflecting how our minds interpret and augment sensory information.”

The team now wants to use this technique to map exactly where and how the brain's version of reality diverges from the real thing - in mice first, but with longer-term implications for understanding visual processing across species, and potentially for future debates about neural privacy.

TL;DR: Researchers reconstructed 10-second videos that mice had watched using neural activity from the visual cortex, achieving a pixel-level correlation of about 0.57 with the original. The mismatches between original and reconstruction may help reveal how the brain’s internal representation of the visual world differs from the raw input, with longer-term implications for neuroscience and, potentially, neural privacy.

News

MIT’s chip packs 30,000 pixels into two. Engineers built a photonic chip that projects intricate light patterns directly into free space using thousands of microscopic “ski-jump” structures that bend light upward. MIT says the device can pack about 30,000 pixels into the area of two smartphone-display pixels, potentially enabling ultra-compact displays, improved lidar, and new ways to control quantum computers.

A new diamond tougher than diamond? Researchers report the strongest evidence yet that hexagonal diamond, also called lonsdaleite, can be made in the lab. By squeezing graphite to about 20 gigapascals and heating it above 1,300 °C, they produced millimeter-sized crystals whose structure matches the long-debated material. If confirmed in follow-up work, it could prove harder than ordinary diamond, with possible uses in cutting tools, thermal management, and advanced sensing.

Physicists may have broken a 30-year superconductivity record. By briefly compressing the cuprate Hg-1223 and then releasing the pressure, researchers reported “locking in” superconductivity up to 151 K (-122 °C) at ambient pressure - about 18 K warmer than the long-standing 1993 record of 133 K. The result suggests new ways to create higher-temperature superconductors without maintaining extreme pressure.

Stem cells helped repair severe eye injuries. In a small clinical trial at Massachusetts Eye and Ear, surgeons transplanted lab-grown limbal stem cells into 14 patients with severe corneal damage, restoring the corneal surface in most cases over 18 months and improving vision in some patients. The results suggest that rebuilding the eye’s own stem-cell population could become a powerful new treatment for certain otherwise untreatable corneal injuries.

Lab tests challenge a popular theory of life on Titan. Scientists found that vinyl cyanide, also known as acrylonitrile, which had been predicted to form cell-like membranes in Titan’s methane-rich lakes, instead crystallized in lab experiments under Titan-like conditions. That makes Titan-style “azotosomes” less likely, though it doesn’t rule out exotic life entirely.

How your brain finds one voice in noise. MIT researchers built a computational model explaining how the brain solves the “cocktail party problem”: focusing on one voice in a noisy room. The model selectively boosts neurons tuned to features of a target voice, like pitch. That small shift lets the system reproduce both the successes and the mistakes humans make when trying to follow one speaker among many.

Lung disease and deaths set to rise under Trump policies. A generation ago, air pollution in the US started to fall. Laws like the Clean Air Act helped drive a steady decline in US air pollution and the diseases that come with it. But that trend may now be reversing. A new analysis estimates that recent healthcare policy rollbacks could lead to tens of thousands of additional deaths from lung disease in the years ahead.

This Week in History

March 14, 1879. Albert Einstein was born. His theories of special and general relativity shattered the old view of space and time as fixed and separate. Instead, Einstein revealed a universe where space, time, and gravity bend and evolve together.

March 16, 1961. NASA formally dedicated the Goddard Space Flight Center on this day in 1961, naming it after rocket pioneer Robert H. Goddard. Thirty-five years earlier, Goddard launched the world’s first liquid-fuel rocket. This experiment helped turn rocketry from a curiosity into the technology that would eventually launch satellites and spacecraft.

March 18, 1965. Soviet cosmonaut Aleksei Leonov made history by stepping outside Voskhod 2 for the world’s first spacewalk. Floating above Earth for about ten minutes, he proved humans could work in open space - a daring test that made future space stations, satellite repairs, and deep-space missions possible.

This Week’s Puzzle

🧩

PI DAY PUZZLE

A quarter circle of radius r contains two semicircles whose diameters lie along the sides of the square.

Inside the diagram are two shaded regions: a yellow crescent and a green lens

Without measuring anything, show that the yellow area is exactly equal to the green area.

Until next time,

The Ve Team 👋

Know someone who would enjoy this newsletter? Forward it their way.

Research by Pablo
Written by Ines

​Unsubscribe · Preferences​

🧩

​
Solution

Imagine four copies of the diagram arranged in a circle. Each quarter piece fits together to form the symmetric picture shown below.

We now have one large circle of radius r and four smaller circles of radius r/2.

The area of a circle is 𝜋r². So the large circle has area 𝜋r².

Each small circle has radius r/2, so its area is 𝜋(r/2)²=𝜋r²​/4

Since there are four of them, 4 x 𝜋r²​/4 = 𝜋r²​

So the total area of the four small circles is exactly the same as the area of the large circle.

When we place the four small circles inside the big one, the green regions are where the small circles overlap each other and the yellow regions are parts of the big circle not covered by the small circles.

If we look at the white + green region, we can describe it in two ways.

1. Start with the four small circles and subtract their overlaps (the green regions).
2. Start with the big circle and subtract the yellow regions.

Both of these expressions equal the same thing because the big circle area = area of the four small circles

So we get

white + green = 𝜋r² − green

white + green = 𝜋r² − yellow

Since both equal the same quantity,

green = yellow