Google DeepMind's WeatherNext model forecasts cyclones using data 100 times coarser than usual, Earth's fungal threads could stretch from Earth to the Sun 750 million times, and identical twin medical students won $1.5 million after being wrongly accused of cheating. [DVYio Newsletter](https://newsletter.dvy.io/visit/ZXcKrhRK) Tech [Map showing a cyclone forecast track crossing the Gulf of Mexico toward the southeastern United States, with storm icons growing larger.](https://newsletter.dvy.io/visit/Ry3wX5Pg) [Google DeepMind says WeatherNext forecasts cyclones using data 100 times coarser than traditional models](https://newsletter.dvy.io/visit/Ry3wX5Pg) Google DeepMind says its WeatherNext model can forecast a cyclone's track, intensity, and wind structure three days ahead as accurately as older models managed two days ahead. It does this using data 100 times coarser than traditional models, and even DeepMind's scientists are not sure why. They are open-sourcing the model so other researchers can investigate. [Infographic shows six AI agents moving from no economic activity to work-based task access, then developing loans, transfers, and vote-trading; symbolic energy preserves competition but removes support.](https://newsletter.dvy.io/visit/uDgvrDS5) [What economy emerges when six AI agents get trading tools but no economic instructions?](https://newsletter.dvy.io/visit/uDgvrDS5) In a preprint, researchers put six AI agents (a mix of GPT and DeepSeek models) into 24 simulated worlds. The agents could work, trade, vote, and share resources, but received no instructions on how to behave economically. Little financial behaviour emerged until they had to complete verified work to earn scarce access to tasks. Then loans, transfers, and vote-trading appeared. When the researchers made the underlying “energy” resource symbolic, with no real stakes, the agents still competed for task access but stopped supporting one another. The behaviour seems to follow from the system's rules and incentives, not from roles or personalities baked into the prompt. [McDonald’s fries under a metal dome, superimposed on blue data documents and tables.](https://newsletter.dvy.io/visit/Dc2dqoyh) [McDonald's algorithm gave a WIRED reporter zero percent odds of ever quitting them](https://newsletter.dvy.io/visit/Dc2dqoyh) WIRED writer Reece Rogers requested his data from McDonald's under California privacy law and got back a 515-page file. It didn't just log years of orders (including every Monopoly code scan), it predicted he'd visit 2.16 times in the next six weeks, spend $29.15 total, and gave him an "attrition likelihood" score of zero, meaning the algorithm reckons he'll never stop being a customer. [Graphic shows hydration gadgets: a smart bottle, Vivoo toilet sensor, light measuring urine concentration, and phone reporting a good hydration level.](https://newsletter.dvy.io/visit/cec5KGh4) [Sweat patches, smart bottles, and toilet sensors: inside the boom in hydration-tracking gadgets](https://newsletter.dvy.io/visit/cec5KGh4) Vivoo's gadget clips onto your toilet bowl and estimates how dehydrated you are from your urine. Light sensors measure “urine specific gravity” (how dense the liquid is compared with plain water), a legitimate method used in hydration studies, according to Andreas Flouris at the University of Thessaly. He notes that dehydration can take a while to show up in urine. This Week I Learnt [Three world maps show fungal web density, with highest examples in South Sudan (8.05), the US (10.29), and China (11.40 metres per cm³).](https://newsletter.dvy.io/visit/LPzrY4ED) [Earth's fungal threads could reach the Sun 750 million times over](https://newsletter.dvy.io/visit/LPzrY4ED) [A teaspoon of ordinary soil can hold up to 10 metres of fungal thread, coiled invisibly under your feet. A new global map estimates that arbuscular mycorrhizal fungi span 110 quadrillion kilometres across grasslands, wetlands, and forests worldwide (roughly 750 million times the distance from Earth to the Sun). These underground networks link plant roots to water and nutrients. That's the headline number from the first global map of these networks, built by the Society for the Protection of Underground Networks using machine learning trained on more than 16,000 soil samples and published in Science. The fungi have been striking the same deal with plants for about 475 million years: they extend far past what roots can reach on their own, fetching water and nutrients, and the plant pays them back in carbon. Some of that carbon then gets pulled deep into the soil, which is part of why healthy fungal networks matter for storing carbon and regulating climate, not just for the plants themselves. Grasslands turn out to hold the densest networks, not forests, with the Tibetan Plateau reaching 11.4 metres of fungal thread per cubic centimetre of soil and the Flint Hills prairie in the US hitting 10.29. Cropland tells a different story: network density there runs 47.3% lower on average than in wild ecosystems, largely because tilling physically rips the soil apart and fertilisers and fungicides interfere with the plant-fungi partnership, according to lead author Dr Justin Stewart. That's a problem the researchers think is partly self-inflicted, since Stewart argues current crop yields are artificially propped up by heavy fertiliser use precisely because farmers aren't leaning on the fungi that could deliver nutrients for free. Co-author Dr Toby Kiers says the map now gives scientists a benchmark for what a healthy underground community looks like, and the team plans to present the findings to governments at the desertification COP in Mongolia in August 2026.](https://newsletter.dvy.io/visit/LPzrY4ED) Poll How much of the technology you use every day could you explain without looking it up? Most of it The broad idea, not the mechanism Only a few things Almost none Last Week's Poll On a familiar journey, how often do you arrive with almost no memory of the trip? Regularly 28% Occasionally 33% Rarely 19% Never 20% Science [Infographic shows heifers nose-touching a target, solving a color-and-texture puzzle for grain or straw, and running faster to repeat it: 20.7 versus 56.8 seconds for controls.](https://newsletter.dvy.io/visit/PWgkwxw9) [Cows get a buzz from figuring things out for themselves, not just from getting fed](https://newsletter.dvy.io/visit/PWgkwxw9) Researchers tested ten pairs of Holstein heifers to see whether the cows enjoyed learning, not just eating. One heifer in each pair had to touch a target with her nose to unlock a colour-and-texture puzzle that revealed grain or straw; her paired control received the same food without solving anything. The puzzle-solving cows ran to the training area almost three times faster (20.7 seconds versus 56.8) and tended to keep pressing for more chances to take part. The researchers interpret that as motivation to learn, rather than simply to get food. [Infographic: Liraglutide reduced Alzheimer’s proteins in most animal studies, but two small human trials found no clear benefit.](https://newsletter.dvy.io/visit/nTzwqwxp) [Liraglutide (a GLP-1 drug like Ozempic) cut Alzheimer's proteins in most animal studies, but human trial results are less convincing](https://newsletter.dvy.io/visit/nTzwqwxp) A review in Molecular and Cellular Neuroscience found that liraglutide (a GLP-1 drug in the same class as Ozempic) reduced amyloid plaques in 13 of 15 animal studies and tau tangles in all 12 that measured them. The two small human trials found no clear reduction in plaques or improvement in memory. One possibility is that these drugs work better as prevention, years before symptoms appear, than after Alzheimer's disease is established, but human evidence is still sparse. [Illustration shows slow breathing affecting the heart and brain, with highlighted brain regions above gambling chips and a tilted risk scale.](https://newsletter.dvy.io/visit/RDE3hQym) [Breathing out slowly makes people take more financial risks, brain scans show](https://newsletter.dvy.io/visit/RDE3hQym) In a 41-person study, participants made gambling decisions in a functional MRI scanner while breathing normally or following a slow pattern: two seconds in, eight seconds out. The long exhale increased parasympathetic activity (“rest and digest”, measured through heart rate variability) and led to riskier bets. Brain scans also showed stronger reward responses in two regions, the ventromedial prefrontal cortex and precuneus, among participants whose parasympathetic activity rose most. Data [World map showing wait times for the next total solar eclipse, from days to 975 years, for towns worldwide.](https://newsletter.dvy.io/visit/tjA9RT4V) [Some cities will wait 900 years for a total eclipse, and one mapmaker calculated the wait for every town on Earth](https://newsletter.dvy.io/visit/tjA9RT4V) [Reykjavik gets a total eclipse in 2 days, Sydney in 2 years, London has to wait 574 years, and São Paulo is looking at 880. Alexander Bogachev built an interactive map calculating this wait for every town over 15,000 people (31,644 of them) by crunching NASA's Five Millennium Canon of Solar Eclipses, which tracks all 11,898 eclipses from 2000 BCE to 3000 CE. Highly recommend checking the link to the website on this one.](https://newsletter.dvy.io/visit/tjA9RT4V) [Aerial view of Nereidas, a pink curved concrete structure with a round pool in a cactus-filled Baja California landscape.](https://newsletter.dvy.io/visit/9Pdrgzg6) [Interior of a board-formed concrete house with a curved ceiling, skylight, wood door, bedroom, and rounded concrete sink.](https://newsletter.dvy.io/visit/9Pdrgzg6) [Open pink concrete structure with a curved roof, columns, built-in seating, and a sunlit cactus-filled desert landscape.](https://newsletter.dvy.io/visit/9Pdrgzg6) [Open pink concrete structure with curved roof, round columns, staircase, pool, and desert landscape beyond.](https://newsletter.dvy.io/visit/9Pdrgzg6) [Aerial view of Nereidas, a pink curved concrete structure with sweeping walls and stairways amid sandy, cactus-filled Baja California terrain.](https://newsletter.dvy.io/visit/9Pdrgzg6) [A curved pink concrete structure with open terraces and a circular pool sits amid dense cactus scrub in Baja California.](https://newsletter.dvy.io/visit/9Pdrgzg6) [Aerial view of pink concrete structures, paths, and a circular pool scattered through dry Baja California scrubland.](https://newsletter.dvy.io/visit/9Pdrgzg6) [Aerial view of pink concrete structures amid coastal vegetation beside a wide beach and calm ocean.](https://newsletter.dvy.io/visit/9Pdrgzg6) [Nereidas](https://newsletter.dvy.io/visit/9Pdrgzg6) [Baja California, Mexico](https://newsletter.dvy.io/visit/9Pdrgzg6) History [Vintage vacuum-tube radio beside a graph showing two looping, intertwined patterns representing chaotic motion.](https://newsletter.dvy.io/visit/eCpinxQT) [A wartime radio problem anticipated chaos theory](https://newsletter.dvy.io/visit/eCpinxQT) [In 1938, Britain's Radio Research Board asked mathematicians for help with a practical problem: radios built with vacuum tubes kept producing messy, nonlinear behaviour that engineers couldn't predict. Mary Cartwright and John Littlewood took on a version of an equation called the forced van der Pol equation (a standard model for oscillating circuits) and found something odd. Instead of settling into one steady rhythm, the equation could jump between multiple different stable outputs, switch abruptly between them, and respond wildly out of proportion to tiny changes in starting conditions. That's now textbook chaos theory: deterministic systems that are, in practice, unpredictable. But in 1945, when Cartwright and Littlewood published their findings, "chaos theory" didn't exist as a field. There was no drawer to put this in, so their result got filed under nonlinear radio engineering instead. It sat there, cited by specialists, while later popular histories of chaos theory largely skipped over it. The pattern is worth noticing because it's not unique to Cartwright and Littlewood. New fields tend to get built backwards: someone solves a weird one-off problem in whatever domain they're working in (radios, in this case), and the strangeness gets explained away as a quirk of that domain rather than recognised as a general phenomenon. It's only later, once enough odd results pile up across different fields, that someone notices they're all the same thing and gives it a name people can search for. Once "chaos" existed as a concept, mathematicians like Norman Levinson and Stephen Smale went back and built rigorous foundations under what Cartwright and Littlewood had only described, and Smale's work eventually grew into a whole geometric theory of these systems. The maths was there in 1945. It just didn't have a name yet, which meant almost nobody outside radio engineering knew to go looking for it.](https://newsletter.dvy.io/visit/eCpinxQT) [A cylindrical metal pneumatic mail carrier with dark end caps and a circular hinged door at one end.](https://newsletter.dvy.io/visit/SX4mmxvo) [US mail once travelled underground at 35 mph through pneumatic tubes](https://newsletter.dvy.io/visit/SX4mmxvo) In 1893, Philadelphia became the first US city to shoot mail through underground pneumatic tubes. By 1898, New York, Boston, Chicago, and St Louis had followed, and Philadelphia alone eventually ran nearly 20 miles of cast-iron tubes. The system worked at low volumes, but slowed as mail increased (like pouring two gallons through the same funnel as one). Once road transport improved, the network was switched off in 1953. Random [Sankey chart compares essential vocabulary categories in 1953 and 2023, with General & Abstract rising from 17.4% to 22.3%.](https://newsletter.dvy.io/visit/A86omm2n) [A core English vocabulary list has dropped words like goat, fork, and umbrella](https://newsletter.dvy.io/visit/A86omm2n) [The list teaches English learners 2,800 words that cover 90% of everyday speech. Between 1953 and 2023, it dropped apple, fork, soap, and goat, while adding mortgage, assumption, and evaluation. Concrete, sensory words fell from 21% of the list to 14%. Writer Jasmine Nackash argues that the shift reflects daily life moving away from tools and food towards institutions and abstract systems, alongside white-collar workers overtaking blue-collar workers in the US by 1957.](https://newsletter.dvy.io/visit/A86omm2n) [Two smiling twin sisters wearing white medical coats stand side by side outdoors.](https://newsletter.dvy.io/visit/PSPxv5wD) [Twin sisters accused of cheating for scoring alike win $1.5 million defamation case against their medical school](https://newsletter.dvy.io/visit/PSPxv5wD) Kayla and Kellie Bingham, identical twin medical students, answered 296 of 307 exam questions the same way in May 2016, including 54 identical wrong answers. To the Medical University of South Carolina, that looked like proof of cheating. A school "honour council" accused the pair of "signalling one another and passing notes" during the eight-hour exam and ruled they'd cheated. But the twins had also scored within a point of each other on the SATs, MCATs, and LSATs. Their defence argued that this long-running pattern was consistent with identical twins, not proof of collusion. Nancy Segal, a psychologist who runs the Twin Studies Center at California State University, Fullerton, testified that research consistently shows identical twins perform alike on tests of intelligence and information processing. "I would have been surprised if they hadn't scored alike," she told CBS MoneyWatch. The accusation cost the sisters more than a grade. Classmates who'd studied beside them for two years stopped speaking to them, and the Binghams say the ordeal brought on panic attacks and PTSD. They abandoned medicine altogether and now both work as government affairs advisers at the same South Carolina law firm. After successfully appealing the university's ruling, they sued for defamation, and in November a South Carolina jury sided with them, awarding $1.5 million in damages. "Once word gets out, even if it's not accurate, it damages your reputation as a person," Kayla said. Identical answers, wrong ones included, can look damning without proving cheating. [Illustration links a sardine-only diet trend, warming oceans, depleted fish stocks, and an empty sardine shelf.](https://newsletter.dvy.io/visit/SctiRkUt) [A TikTok diet trend and warming oceans just teamed up to empty the sardine aisle](https://newsletter.dvy.io/visit/SctiRkUt) A viral TikTok and YouTube diet tells people to eat only sardines for one to three days, alongside water, black coffee, or tea. Promoters such as Dr Annette Bosworth helped push up demand just as warmer oceans disrupted the plankton sardines feed on, shrinking fish and cutting Moroccan catches by more than half in some regions. Morocco is a major source of tins sold in North America and Europe. Fishing limits designed to protect stocks mean suppliers cannot simply catch more, so the shortage may last. Which kind of "This Week I Learnt" fact do you enjoy most? A useful misconception correction A weird old word A hidden mechanism in everyday life A historical reversal A tiny detail with a big consequence [“A man with a watch knows what time it is. A man with two watches is never sure.”](https://newsletter.dvy.io/visit/DJn8Gykg) [Anonymous, later known as Segal’s law](https://newsletter.dvy.io/visit/DJn8Gykg) [Web](https://newsletter.dvy.io/visit/4nePJTRH) · [Twitter](https://newsletter.dvy.io/visit/j3YEvSNX) · [LinkedIn](https://newsletter.dvy.io/visit/wHQQAKmJ)