Sceye plans to test 5G from a helium aircraft 18 km above the Pacific, Ford has rehired more than 300 inspectors and engineers after its AI cameras fell short, and employment has fallen about 12% for young US workers in AI-exposed roles. Plus there may be as many as 30 million insect species, and a report about America in 1926 reads like today’s news. [DVYio Newsletter](https://newsletter.dvy.io/visit/GDWqDLkx) Tech [A shiny blimp-like aircraft with fins floats against a blue sky and clouds.](https://newsletter.dvy.io/visit/GYeujNau) [Sceye plans to hover a 200-foot-long aircraft 18 km above the Pacific to extend SoftBank’s 5G network](https://newsletter.dvy.io/visit/GYeujNau) Sceye plans to send a 200-foot-long, helium-filled aircraft across the Pacific as soon as August, then hold it 18 km above the sea near Japan. From there, a custom antenna will extend SoftBank’s 5G network and test sending data straight to phones. The New Mexico company calls the craft a high-altitude platform station, or HAPS. In plain terms, it is a solar-powered mobile phone mast floating in the stratosphere. That puts it above commercial aircraft and most weather, but far below orbit. It can cover a large area like a satellite, while its shorter distance from the ground means signals need much less power. Keeping a giant balloon-like craft parked in thin air is difficult. Solar panels charge batteries during the day, powering an electric fan that corrects its position when winds push it away. Sceye tested that system in 2024. This spring, another flight reached the coast of Brazil, stayed airborne for 12 days, and spent more than 88 hours holding position. Airbus subsidiary Aalto is developing similar platforms, which could also provide internet after disasters or observe the ground below. The goal is satellite-like coverage without putting anything into orbit (still not exactly simple, though). [A silver Ford pickup truck moves along an indoor factory assembly line, with another truck behind it.](https://newsletter.dvy.io/visit/cWbMnPgS) [Ford rehired more than 300 quality inspectors and engineers after its AI checks fell short](https://newsletter.dvy.io/visit/cWbMnPgS) Ford installed 900 AI-powered cameras across its factories to catch faults as cars moved through production. The company fed its design rules into the software and expected it to judge quality much like an experienced inspector. According to Charles Poon, Ford’s vice president of vehicle hardware engineering, that assumption was wrong. A design rule can tell a system which measurements a part should meet. It cannot easily capture the judgement built from seeing thousands of cars over many years (rather like learning carpentry from the instruction manual alone). Many veteran inspectors had already left before Ford used their knowledge to train its automated tools. Ford has since rehired more than 300 experienced quality inspectors and engineers. They are checking vehicles, identifying which defects matter, and mentoring younger staff. The cameras remain, but the people who know what good looks like are now teaching the system. [Chart showing employment change by age and AI exposure since ChatGPT launched: ages 22–25 in the most exposed jobs are down 12%, while older groups are flat or up.](https://newsletter.dvy.io/visit/N8VMarr2) [Employment fell about 12% for young US workers in jobs most exposed to AI](https://newsletter.dvy.io/visit/N8VMarr2) Anonymised payroll records from about 25,000 US firms show a sharp split for workers aged 22 to 25. Since ChatGPT launched in November 2022, their employment has fallen about 12% in jobs where AI can handle many tasks, including software development, customer service, and accounting. In hands-on jobs that AI barely touches, such as home care and skilled trades, employment rose about 7%. Stanford’s Digital Economy Lab and ADP Research suggest junior roles are more exposed because much of the work follows rules that can be written down. Senior staff rely more on judgement built through experience, which AI still handles poorly. The sharpest drop is at the first rung of the career ladder: employment among young software developers has fallen nearly 20% from its late-2022 peak. Separate Dallas Fed research suggests this is mostly a drop in hiring, rather than a wave of redundancies. The evidence is not settled. Stanford’s own follow-up found the pattern did not become statistically clear until 2024, and the researchers do not think AI is the sole cause. Studies from Yale’s Budget Lab and the New York Fed have found little clear evidence of an AI-driven fall in jobs or job adverts. For a recent graduate, though, the concrete problem is already familiar: fewer employers are opening the junior roles that used to provide a way in. This Week I Learnt [Isometric road diagram with a red traffic light, a car, and labels for detection methods: no detection, inductive loop, video, radar, and lidar, plus a fixed cycle icon.](https://newsletter.dvy.io/visit/hLkdbUmj) [Traffic lights inherited red, yellow, and green from Victorian railway signals](https://newsletter.dvy.io/visit/hLkdbUmj) [Traffic lights inherited their colour system from Victorian railway semaphores (large mechanical arms that told train drivers whether the track ahead was clear). Red became the stop colour partly because of physics: its longer wavelength scatters less in fog, so it can be seen from further away than green or yellow. The railways needed signals that worked in bad weather, and the colours stuck around long enough to end up in traffic signals around the world. From there, the piece explains how modern traffic lights work. It covers the inductive loops buried beneath the road (metal coils that detect a waiting car through changes in a magnetic field) and networked controllers that coordinate signals across a city rather than treating each junction in isolation. The piece uses more than 40 custom visualisations to show the mechanisms rather than just describe them. It is part of a broader series that gives the same treatment to QR codes, GPS, and Shazam.](https://newsletter.dvy.io/visit/hLkdbUmj) Poll What makes a technology feel genuinely futuristic to you? Changes how people communicate Makes something invisible or effortless Gives people a new sense or ability Shrinks something huge into something tiny Looks slightly ridiculous at first Last Week's Poll If a safe non-surgical brain-computer interface existed today, would you get one? Yes, straight away 26% Only if medically necessary 19% Not yet, but maybe later 28% No, never 27% Science [Concept artwork of a six-wheeled lunar vehicle marked “3” collecting material on the Moon, with satellite dishes in the background.](https://newsletter.dvy.io/visit/sam5ZPQP) [Most of today’s helium-3 supply comes from tritium in nuclear warheads, but companies want to mine it on the moon](https://newsletter.dvy.io/visit/sam5ZPQP) Lancaster University stores one of its most valuable assets in beer kegs: helium-3, a gas that currently costs around £1,500 ($2,000) a litre. Senior lecturer Dima Zmeev credits “very wise predecessors” who bought in decades ago, when the price was still low. Helium-3 is the rarer sibling of the helium in party balloons, differing by just one neutron. That small difference matters enormously: mixed with ordinary helium and cooled to near absolute zero (-273°C), it creates the ultra-cold conditions quantum computers need to function. It’s also a candidate fuel for nuclear fusion reactors. The problem is supply. Almost all of it comes from tritium decaying inside nuclear warheads, a tightly controlled process that Oak Ridge scientist David McCollum estimates produces only tens of thousands of litres a year worldwide. Quantum computing alone could eventually need thousands of litres per machine. Two US companies think the answer is on the moon. Apollo mission samples suggest lunar dust contains helium-3 at higher concentrations than anything on Earth, so Interlune (co-founded by Apollo 17 astronaut Harrison Schmitt, now in his 90s) wants to send autonomous excavators to scoop, crush, and process moon dust to release the gas. A deal is already signed: Finnish quantum computing firm Bluefors has committed $300 million for 10,000 litres a year from 2028. There are real doubts, though. The Apollo samples may have lost some helium-3 on the way back to Earth, so concentrations could be lower than hoped. Paul Burke at Johns Hopkins estimates you might need to excavate and process hundreds of thousands of tonnes of lunar dust to extract a single kilogram of the gas. Interlune says it has worked through the costs, but has declined to share the numbers. The more practical alternative is Minnesota. Pulsar Helium is investigating a site there with concentrations around 12 parts per billion, and geochemist Peter Barry puts it plainly: “Minnesota is a lot easier to get to than the moon.” [Starry night sky over dark hills, crisscrossed by many bright satellite trails and a band of the Milky Way.](https://newsletter.dvy.io/visit/HbvnCmxN) [A single mirror-satellite could be as bright as Venus and ruin a Rubin Observatory image](https://newsletter.dvy.io/visit/HbvnCmxN) ESO astronomer Olivier Hainaut modelled the effects of proposals for more than 1.7 million new satellites, on top of the roughly 14,000 already in orbit (mostly SpaceX’s Starlink fleet). His peer-reviewed study concludes that fleets on this scale would be “beyond the limit” of what astronomy can cope with. The worst offender in his simulations is Reflect Orbital, a US start-up planning a 50,000-strong fleet of mirror satellites that would bounce sunlight down to Earth at night. Even outside the reflected beam, each satellite would shine as brightly as Venus. A single one crossing an image from the Vera C. Rubin Observatory could ruin it: one bright streak can trigger ghost trails across the whole frame, not just along the streak. Hainaut found that, with the entire fleet in orbit, every image from a camera like Rubin’s would be lost whenever the satellites caught sunlight. The damage is not limited to streaks across individual images. Even satellites too dim to spot with the naked eye add up to a hazy veil of light that brightens the whole sky, making faint targets like distant galaxies or hazardous asteroids harder to see everywhere, not just where a satellite happens to pass. [A black-coated face-shaped object stands beside a reflective gold sculpted face in a laboratory.](https://newsletter.dvy.io/visit/AUdHiJAQ) [The world’s darkest coating could reduce satellite glare for astronomers](https://newsletter.dvy.io/visit/AUdHiJAQ) As a complement to the previous story… University of Surrey researchers tested a possible fix for bright satellites: coat them in Vantablack® 310. The material, made by Surrey NanoSystems, a University of Surrey spinout, reflects only about 2% of the light that hits it. That small amount is scattered in many directions rather than sent back as a sharp glint. In a new study in Monthly Notices of the Royal Astronomical Society, the team measured how the coating behaves under different lighting angles, then simulated what a coated satellite would look like from the ground. The simulated brightness came close to the limit recommended by the International Astronomical Union for protecting stargazing and telescope surveys. The result comes from lab measurements and simulation, not an orbital test. The team plans to test the coating on Jovian-1, a student-led CubeSat involving the universities of Surrey, Portsmouth, and Southampton. The flight will show how the coating copes with space and whether telescopes on Earth can measure the change in brightness. [Suspension bridge through dense tropical rainforest, with trees, vines, and epiphytes surrounding the walkway.](https://newsletter.dvy.io/visit/vJz7PkM5) [Up to 97% of Earth’s insect species may still be unnamed](https://newsletter.dvy.io/visit/vJz7PkM5) More than 1.6 million insects collected in a protected Costa Rican rainforest have helped researchers revise the estimated number of insect species worldwide. The new estimate is 14 million to 30 million species, well above the previous figure of six million. Scientists at Cornell University, the University of Colorado, and the University of Kentucky led the international team, which included Dr Robert Puschendorf from the University of Plymouth. They used DNA barcoding, a method that identifies species from a short genetic sequence (rather like scanning a supermarket barcode). They combined this with field observations and statistics, then checked their estimates against data for trees, amphibians, and moths from Área de Conservación Guanacaste, a UNESCO World Heritage Site. The analysis repeatedly found cryptic species: insects that look the same but have enough genetic differences to count as separate species. Those hidden divisions are easy to miss when identification relies mainly on appearance. The researchers conservatively estimate that 93% to 97% of insect species remain unnamed. Puschendorf says the same methods could be used elsewhere, including the UK, to learn what is living there, what has already disappeared, and what still needs protecting. It is rather difficult to manage an ecosystem when most of its inhabitants are missing from the list. [A low, minimalist pavilion with floor-to-ceiling curved glass walls and a thin, white flat roof.](https://newsletter.dvy.io/visit/P4zwVteD) [A modern building with a curved glass exterior wall reflecting bare winter trees on a grassy lawn.](https://newsletter.dvy.io/visit/P4zwVteD) [Curved, floor-to-ceiling glass walls intersect a flat white ceiling with recessed lighting and minimal track fixtures.](https://newsletter.dvy.io/visit/P4zwVteD) [A low, modern building with floor-to-ceiling glass walls and a flat white roof sits behind a grassy lawn.](https://newsletter.dvy.io/visit/P4zwVteD) [An aerial view of the Toledo Glass Pavilion, a modernist rectangular building with glass walls and a flat roof, set in a park.](https://newsletter.dvy.io/visit/P4zwVteD) [A low, modern building with a flat white roof and curved glass walls that reveal inner layers of glass and white curtains.](https://newsletter.dvy.io/visit/P4zwVteD) [A low minimalist building with curved glass walls and a thin white roof reflecting the surrounding park in Ohio.](https://newsletter.dvy.io/visit/P4zwVteD) [Toledo Glass Pavilion](https://newsletter.dvy.io/visit/P4zwVteD) [Toledo, Ohio](https://newsletter.dvy.io/visit/P4zwVteD) Data [US map showing the share of households with two or more refrigerators in 2024, ranging from 19% in DC to 48% in South Dakota.](https://newsletter.dvy.io/visit/WdJE2Cau) [America has a ‘second-fridge belt’ running through the Great Plains](https://newsletter.dvy.io/visit/WdJE2Cau) [The US Energy Information Administration’s 2024 survey reveals a neat stripe on the map: households with multiple fridges are most common across the Great Plains and Upper Midwest. Nationwide, about 34% of households have at least two fridges. South Dakota leads at 48%, followed by North Dakota and Nebraska (both 46%), Kansas (43%), and Minnesota (42%). The pattern fades sharply elsewhere. Just 19% of households in Washington, DC, have two or more fridges, compared with 27% in New York and 30% in California.](https://newsletter.dvy.io/visit/WdJE2Cau) History [Black-and-white photo of a rainy city street at night with a line of early cars and bright shop signs reflected on wet pavement.](https://newsletter.dvy.io/visit/Az2p7Tzu) [A 1926 report on America's anxieties about tech, immigration, and work reads like this week's news](https://newsletter.dvy.io/visit/Az2p7Tzu) [Herbert Hoover commissioned Recent Social Trends in 1929. The 1,500-page report took four years to complete and examined American life in the mid-1920s. It has since been almost entirely forgotten. The parallels are hard to ignore. In 1926, America was celebrating a big birthday (its 150th) during a booming stock market, worried about machines stealing jobs (tractors, not algorithms), watching a new communication technology rewire how people got their news (radio, not social media), and clamping down on immigration after years of record arrivals. Swap a few nouns and it reads like this week. But 1926 was also genuinely alien. Half the country was rural. Millions of homes had no electricity or indoor plumbing. Women had voted in just two presidential elections. Child labour was common. Only 11 million of the 27 million US households owned a phonograph or a car, and the first talking picture was still a year away. The article imagines a composite “typical American”: a 26-year-old man named John, raised on a farm and moving to a city, earning $100 a month with no unemployment insurance or Social Security to fall back on. The same anxieties in a completely different material world.](https://newsletter.dvy.io/visit/Az2p7Tzu) [A woman and two US servicemen play small ocarinas for a crowd at an indoor wartime gathering.](https://newsletter.dvy.io/visit/APhVk7Jf) [How the US Army shipped thousands of plastic ocarinas to its troops in World War II](https://newsletter.dvy.io/visit/APhVk7Jf) The US Army shipped two models of plastic ocarina to its troops during World War II: the Gretsch Alto C “sweet potato” (a stubby, oval instrument made by a Brooklyn company founded by a German immigrant) and the tonette, a straight design that looked a bit like a recorder. Both came with two filled-in finger holes, so beginners could play a simplified scale right away, then carve the holes open with a penknife once they got confident, adding two extra notes to the range. The ocarina won out over other options for practical reasons. The plastic recorder had not been invented yet (that came in 1946), and wooden recorders could warp or crack in the wet, rough conditions of the field. The harmonica was pocketable and plenty of soldiers carried their own, but the military did not hand them out. The ocarina, meanwhile, could be pressed from two plastic moulds and glued together cheaply and at scale, and its fingering was simpler than a recorder’s. Each instrument came with a 25-page method booklet that started with tab notation and worked up to reading music, covering folk songs and Christmas tunes. The Army also produced a companion booklet for ocarina quartets. The box the Gretsch came in read: ANYONE! EVERYONE CAN PLAY OCARINA WITH THE ARMY METHOD! By most accounts, that was roughly true. If you got a tonette rather than a Gretsch, you were the lucky one. The sweet potatoes weren’t tunable; the tonettes were, and they sounded noticeably better. The old war-issue tonettes were made from a heavier, shinier plastic than the airy kids’ versions sold from the 1960s onwards, and that difference in material is apparently audible. Random [Amazon drill-bit search results with one DeWalt and one Amazon Basics listing shown normally while most others are faded and labelled “No brand”.](https://newsletter.dvy.io/visit/rAJWgFCJ) [Browser extension hides Amazon listings from junk brands](https://newsletter.dvy.io/visit/rAJWgFCJ) [WNPETHOME, EHEYCIGA, and GODONLIF are among the Amazon brands flagged by Knockoff, a browser extension that scores unfamiliar names using heuristics and community-informed lists. The extension is meant to clear Amazon’s results pages of near-identical phone cases, pet beds, and other products sold under unfamiliar names. Rather than checking each listing yourself, you can make flagged products less visible or remove them from the results entirely. Knockoff says its verdicts come from heuristics and community lists, so a flag is not proof that a brand or product is fake.](https://newsletter.dvy.io/visit/rAJWgFCJ) [Colour diagram of a four-cylinder internal combustion engine showing pistons, crankshaft, camshafts, valves, and timing belt.](https://newsletter.dvy.io/visit/KxpdZ45v) [Crank, cannon, piston: building an engine from first principles](https://newsletter.dvy.io/visit/KxpdZ45v) This piece opens with a coffee grinder, not a car. Ciechanowski’s interactive walkthrough builds the internal combustion engine from a simple hand crank, the kind you’d find on an old coffee grinder or bicycle pedal, before ever mentioning petrol. The logic runs like this: pushing a handle at a distance from a shaft creates torque. Swap your arm for something stronger, like an exploding cannon, and you have a way to generate serious force. The problem is that a cannonball delivers one push and then flies out of the barrel. So the article turns it into a piston trapped inside a cylinder and connects it to a crankshaft through a connecting rod, turning each push into rotation. From there, it walks through the four-stroke cycle: intake, compression, power, and exhaust, each shown as a 3D animation you can drag and rotate. Once the single-cylinder toy engine works, it scales up to a real four-cylinder car engine block, explaining why cars use several cylinders instead of one (smoother power delivery) and how the crankshaft's counterweights stop the whole thing from shaking itself apart. It is a slow build from “handle on a wheel” to “the thing under your car’s bonnet”, with every part rendered so you can spin it around and watch it move. [Two cooks prepare food at a busy street stall, with a large pan of stew and bowls of ingredients.](https://newsletter.dvy.io/visit/ea856jTF) [Bangkok’s Wattana Panich has kept a beef broth going since 1974](https://newsletter.dvy.io/visit/ea856jTF) Nattapong and Wisa Kaweenuntawong look after the pot now, and by their own account they barely take holidays. That’s the practical cost of keeping a broth going for over fifty years: someone has to be there every day, topping it up and never letting it fully empty. At Wattana Panich, the pot is never washed out between batches. Each day’s fresh beef, bones, and herbs go in on top of the previous day’s broth, so the stock has continued through constant dilution and renewal since 1974. Thai diners call this a “mother stock”, and the family have kept it going through three generations. Someone’s grandchildren are eating soup seasoned by a pot their grandparents started before they were born. Some Italian cheese is aged for up to 18 years, and one French butcher sells steaks more than a decade old. Wattana Panich’s broth has been going much longer than either. Poll What’s the best feeling after reading a newsletter? "I understand something better now" "I had no idea that existed" "I want to send this to someone" "I've got a new rabbit hole" "That was pleasingly weird" [Web](https://newsletter.dvy.io/visit/h8FPQ3mL) · [Twitter](https://newsletter.dvy.io/visit/3cPTPMe4) · [LinkedIn](https://newsletter.dvy.io/visit/Z8L6P74c)