DuckDuckGo traffic jumped 28% after Google's CEO insisted everyone loves AI search. Bats are crossing a Danish road lit red, and French newborns cry in a rising pitch while German ones don't. [DVYio Newsletter](https://newsletter.dvy.io/visit/iVzjbrbz) Is there someone you think would enjoy the newsletter? You'll find a personalised link at the bottom of every email. Send that to them and they'll be recorded as signing up via you. As always, a huge thanks for being subscribed. Any questions or comments? Reply to this email and I'll see them. Tech [A laptop screen displays the DuckDuckGo search homepage, featuring its duck mascot and the slogan Protection. Privacy. Peace of mind.](https://newsletter.dvy.io/visit/PGQQbNxw) [After Sundar Pichai said people love AI search, DuckDuckGo visits jumped 28%](https://newsletter.dvy.io/visit/PGQQbNxw) DuckDuckGo's AI-free search page ([noai.duckduckgo.com](https://newsletter.dvy.io/visit/gHMVRd3d)) saw visits jump nearly 28% in the week after Google CEO Sundar Pichai told the world that "people love Search's AI Mode." iOS installs of the DuckDuckGo app peaked at almost 70% above the previous week's average. Google has been pushing AI overviews to the top of results with no opt-out, which means searching for something like "upcoming PC games 2026" now means scrolling past a chunky AI summary before you reach any actual links. DuckDuckGo CEO Gabriel Weinberg put it plainly: "Google is force-feeding AI with no way to opt out." To be fair about the scale here: DuckDuckGo holds roughly 2% of the US search market. Google still has about 85%. A 28% spike on 2% of the market is a ripple, but it's a meaningful signal about user sentiment, even if it won't dent Google's numbers in any measurable way. [The TikTok logo and wordmark displayed on a smartphone screen angled diagonally against a dark, blurred background.](https://newsletter.dvy.io/visit/fJXXdcsL) [TikTok's AI described dancer Charli D'Amelio as a bowl of blueberries](https://newsletter.dvy.io/visit/fJXXdcsL) TikTok launched "AI overviews" beneath videos, similar to the summaries Google shows at the top of search results. It went badly. Users in the US and the Philippines (the only markets it reached before being pulled back) started screenshotting descriptions that had nothing to do with the actual content. A video of dancer Charli D'Amelio was described as "a collection of various blueberries with different toppings." Shakira and Olivia Rodrigo videos got similarly confused treatment. TikTok has now narrowed the feature considerably. Instead of summarising video content generally, it will only surface information about products shown in videos, which is a much more constrained task and presumably one where being wrong is less embarrassing. [A graphic of a smartphone screen showing a notification card being fed by diverging arrows from two separate circles.](https://newsletter.dvy.io/visit/pPUEZ56A) [Push notifications are now filtered, reordered, and rewritten before they reach you](https://newsletter.dvy.io/visit/pPUEZ56A) Push notifications have always felt direct: your app sends a message, it appears on someone's lock screen. That was true for a while, but it was never quite the full picture. The infrastructure was intermediated from day one. Every notification sent to an iPhone has always passed through Apple's servers; every Android notification through Google's. The platforms could throttle, reorder, or drop messages whenever they wanted. For most of the channel's history, they mostly didn't, but that's changing. It started with a battery problem Scott Forstall described at WWDC in 2009. Running a background polling connection per app would drain any phone inside a day, so Apple built a single persistent connection from each device to its own servers, through which any registered app could deliver alerts. Google followed with its own equivalent in 2010. Two pipes, controlling every notification on almost every smartphone on the planet. Now both platforms have models sitting between delivery and the lock screen, summarising, reordering, and in some cases rewriting notifications before the user sees them. It's the same pattern the article describes happening to email, just with fewer players. Worth reading if you send push notifications professionally, or if you've ever wondered why a message arrived late, out of order, or subtly different from what you sent. [Valentina Palmiotti smiling behind her laptop at the Pwn2Own competition, set against a backdrop of digital beetle icons.](https://newsletter.dvy.io/visit/KUEoNrVq) [AI hacking tools are getting good enough to put champion hackers out of work](https://newsletter.dvy.io/visit/KUEoNrVq) Valentina Palmiotti, known in the hacking world as Chompie, just won $70,000 at Pwn2Own Berlin, the annual competition where ethical hackers race to find security flaws in real software before criminals do. She pulled an all-nighter between rounds and took home more prize money than anyone else at the event. Then she told the BBC that AI tools like Claude Mythos could make people like her obsolete. Mythos is a model from Anthropic that, according to the company, has already found 1,600 vulnerabilities across hundreds of software programs. That's the kind of work Chompie does manually, under pressure, on no sleep. Anthropic considers Mythos too dangerous to release publicly and is only making it available to a small number of governments and security institutions. For now, Chompie says AI is helping her win bug bounties (the money companies pay to hackers who spot holes before criminals do). But "helping" and "replacing" have a way of swapping places quickly in this field. This Week I Learnt [A diagram connecting a human profile to three rows of icons representing light, noses, and slippery motion.](https://newsletter.dvy.io/visit/ZnJQGmro) [Why do "glitter", "gleam", and "glimmer" all involve light?](https://newsletter.dvy.io/visit/ZnJQGmro) [In English, certain sound combinations quietly predict what a word means. "Gl-" tends to signal light or vision: gleam, glint, glimmer, glare, glisten. "Sn-" clusters around the nose: sniff, snout, sneeze, snore, snot. "Sl-" often implies low-friction movement: slide, slick, slither, sled. These patterns are called phonesthemes (foh-NESS-theems), a term coined by British linguist J. R. Firth in 1930. The idea is that a sound cluster carries a loose but consistent meaning, even though it isn't technically a morpheme (a proper unit of meaning like "un-" or "-ness"). The "gl-" in "glitter" isn't a prefix you could detach and reuse the way you'd reuse "re-" or "pre-". It just... tends to show up near light-related words, more than you'd expect by chance. Phonesthemes aren't rules, and they're not universal across languages. They're statistical tendencies, patterns that emerged over centuries of language evolution rather than anything deliberately designed. English speakers absorb them without noticing.](https://newsletter.dvy.io/visit/ZnJQGmro) Poll How much would you pay for a home robot that handles basic chores (washing, cleaning, cooking, etc.) perfectly? Nothing, not interested Under $1,000 $1,001–$5,000 $5,001–$10,000 Over $10,000 Last Week's Poll What do you do when you can't remember if you locked the front door? Trust myself and carry on 51% Go back to check 42% Check via a camera or app 7% Science [A dark country road and forest at night illuminated by a series of low-level streetlights emitting a deep red glow.](https://newsletter.dvy.io/visit/4KQqbNUh) [Denmark is lighting a road red so bats can cross it](https://newsletter.dvy.io/visit/4KQqbNUh) [Most bats won't fly through bright light. It's not a preference thing, it's almost like a wall. Bright white LEDs expose them mid-flight, making them easy prey, so many species simply stop at the edge of the lit zone and turn back. A streetlight can quietly cut a bat colony off from its feeding grounds without anyone noticing. In Gladsaxe, just outside Copenhagen, seven bat species have been recorded along a stretch of road called Frederiksborgvej. In February 2026, the town swapped the white LED streetlights along that stretch for red ones. The road now glows a deep, steady red at night. Red light sits at the far end of the visible spectrum, and bats appear far less bothered by it than white or blue-white light. The road stays lit for drivers, but the bats can cross without triggering their avoid-the-light reflex. It's a small, specific fix for a problem most people never think about. None of the seven species recorded there are currently classified as threatened, but some are considered more vulnerable to road disruption. The coloured road surface is part of the same project, helping mark the corridor visually.](https://newsletter.dvy.io/visit/4KQqbNUh) [A massive, bright fireball and plume of smoke completely engulf a rocket launch pad at night.](https://newsletter.dvy.io/visit/GXfu2TpG) [Blue Origin rocket explodes on the launch pad during a hotfire test in Florida](https://newsletter.dvy.io/visit/GXfu2TpG) Blue Origin's New Glenn rocket exploded on its launch pad at Cape Canaveral during a pre-launch hotfire test on Thursday evening. The footage is striking: the rocket disappears into a fireball that swallows the surrounding pad. Jeff Bezos confirmed all personnel are accounted for and said the team is already investigating the root cause. This is the second significant setback in about a month. Last month, a New Glenn launch failed to place an AST SpaceMobile satellite into the correct orbit, prompting an FAA investigation. The FAA confirmed this latest explosion falls outside its licensed activity scope, so there is no separate regulatory action triggered by the test itself. [A technician in a white lab coat stands beside a large, spherical stainless steel reactor tank in an industrial facility.](https://newsletter.dvy.io/visit/zgzJHgJj) [Turning plastic waste into feedstock via depolymerisation](https://newsletter.dvy.io/visit/zgzJHgJj) Plastic is made by stringing small molecules called monomers into long chains called polymers. That chain structure is what makes plastic tough and durable, which is exactly what makes it a problem once you're done with it. Most recycling today works by melting or grinding plastic into a lower-grade material, and each time you do that, the chains degrade a little more. After a few cycles, the material is too compromised to use and ends up in landfill anyway. Depolymerisation goes the other way: instead of melting the chain, you break it apart at the joins and recover the original monomers. The Lego analogy holds well here. Traditional recycling squishes the bricks together into a lump. Depolymerisation unsnaps them one by one, so you can build something new with the same quality pieces. A company called Denovia says it has a process that does this quickly, by shredding PET plastic (the kind used in drinks bottles) to increase surface area, then immersing it in a chemical solution that splits the polymer chains without high heat. The output is purified monomer that it claims approaches virgin quality, reusable to make new plastic of the same grade, in theory indefinitely. Virgin plastic is cheap because oil is cheap, and any recycling route has to compete with just extracting more. The energy cost of breaking polymer chains can easily exceed the value of what you recover, which is why most chemical recycling processes have stayed at lab scale. Denovia says its process is fast (minutes rather than hours) and low-temperature, which are the two things that tend to kill unit economics. [A glass and copper walkway connects the arched brick walls of the historic Meeresmuseum to a modern copper-clad annex.](https://newsletter.dvy.io/visit/3u7Wfo3H) [A narrow cobblestone street in Stralsund flanked by a classical white building and a modern wall of vertical copper panels.](https://newsletter.dvy.io/visit/3u7Wfo3H) [A massive whale skeleton suspended from the vaulted ceiling of a sunlit hall with red brick walls.](https://newsletter.dvy.io/visit/3u7Wfo3H) [A towering aquarium window illuminates a dark, modern museum hall with tiered wooden seating and concrete floors.](https://newsletter.dvy.io/visit/3u7Wfo3H) [A vast museum interior with high grey ceilings, industrial beams, and large doorways looking out onto green trees.](https://newsletter.dvy.io/visit/3u7Wfo3H) [A modern copper-clad roof in the foreground looks out over the gabled clay tile roofs and brickwork of the old town.](https://newsletter.dvy.io/visit/3u7Wfo3H) [Meeresmuseum](https://newsletter.dvy.io/visit/3u7Wfo3H) [Stralsund, Germany](https://newsletter.dvy.io/visit/3u7Wfo3H) Random [Infographic showing French babies cry with a rising pitch like French speech, while German babies cry with a falling pitch like German speech, using line graphs to compare the intonation patterns of infants just days old.](https://newsletter.dvy.io/visit/CUvWPTKK) [French and German newborns cry in their native language's melody](https://newsletter.dvy.io/visit/CUvWPTKK) [French newborns cry with a rising pitch. German newborns cry with a falling pitch. And that difference maps almost perfectly onto the melody of each language. The study recorded 60 healthy newborns, split evenly between French and German monolingual families, and analysed the shape of their cries. French cries tend to rise toward the end, the way a French phrase does. German cries peak earlier and fall away, the way German intonation does. The effect was statistically strong enough that researchers could tell the two groups apart just from cry melody. Important to note that these babies are hours or days old. They haven't said a word, haven't smiled, haven't done much of anything. But they've already absorbed something about how their language sounds. The leading explanation is that foetuses start picking up the melody and rhythm of speech in the womb during the final trimester, when the auditory system is developed enough to process sound. Not words, not grammar, just the musical shape of it: whether sentences rise or fall, where the peaks tend to land. Crying, it turns out, isn't just noise. It seems to be the first attempt at imitation.](https://newsletter.dvy.io/visit/CUvWPTKK) [A line drawing of a melting wax figure and a devil in a suit drinking tea, illustrating common similes.](https://newsletter.dvy.io/visit/q42w642V) [Fiction writers keep reaching for the same similes](https://newsletter.dvy.io/visit/q42w642V) Russell Samora pulled 200,000 similes from popular fiction, restricting himself to the "as ___ as ___" form because its rigid structure makes it easy to extract programmatically. "Eyes like daggers" is harder to catch automatically; "as sharp as a knife" practically tags itself. Each simile gets broken into three parts: the tenor (what's being described), the ground (the shared quality, usually an adjective), and the vehicle (the comparison noun). So in "as dry as sawdust", "dry" is the ground and "sawdust" is the vehicle. The patterns that emerge are odd. Most adjectives pair with a small cluster of familiar nouns, with a sharp drop-off after the top few choices. "Dry" reaches for "bone", "dust", "sawdust", "stone". But nouns vary wildly in how many adjectives they attract. "Hell" turns out to be a generalist, showing up across dozens of adjectives. "Cucumber" is almost exclusively "cool", a dedicated specialist so strongly paired with one adjective that it's become idiomatic. Samora uses Simpson's index (a measure of diversity borrowed from ecology, where it's used to quantify species variety in a habitat) to quantify this, which is a neat methodological touch. Turns out that figurative language isn't as inventive as it might feel while writing it. Most "as ___ as ___" constructions land on the same small set of vehicles, suggesting that similes calcify into stock phrases faster than writers probably realise. Design [The One World Flag, featuring a solid blue circle on a translucent white field, flies in the wind against a mountain.](https://newsletter.dvy.io/visit/wRMSVAk6) [A flag for Earth: a blue dot on a transparent background](https://newsletter.dvy.io/visit/wRMSVAk6) [Thomas Mandl, a Munich photographer, spent two years on this problem and landed on something genuinely simple: a blue dot (representing Earth) centred on a transparent rectangle. The transparency means the flag takes on whatever is behind it, your street, your sky, your crowd, so every version of it looks different depending on where it's flown. It's a neat bit of design thinking. Every colour carries baggage, so he removed colour from the equation entirely.](https://newsletter.dvy.io/visit/wRMSVAk6) [The title 'STORIED COLORS' above the text 'Named colors with documented histories' and a row of 30 different vertical colour swatches.](https://newsletter.dvy.io/visit/zc3XPwEP) [Storied Colors — a catalog of named pigments with their histories, chemistry, and citations](https://newsletter.dvy.io/visit/zc3XPwEP) Colour history online has a provenance problem: the same facts get rewritten endlessly, and the sources quietly disappear along the way. Storied Colors is a catalogue built around putting those sources back. The launch covers 250 named colours across roughly five thousand years, from Egyptian blue (c. 2500 BCE) to YInMn blue (2009). Each entry runs four to six paragraphs and covers the chemistry, the origin, the trade routes, the patent disputes, and the bans, all on one page, sourced and dated. About 40% are mineral pigments, a quarter are plant or insect dyes, and the rest are synthetics. There's also a small reserved section for digital colours like Klein blue and Baker-Miller pink that have no physical form but earned their place by other means. The entries are explicitly designed as a starting point, not a definitive reference. Every one ends with a citation list, which is where the longer story lives. New entries go up each Sunday. Older ones get corrected as conservation labs publish new findings. Data [Line chart showing US remote work rising slowly from 5m in 2005 to 9m in 2019, leaping to 27m in 2021, and dipping to 22m by 2024.](https://newsletter.dvy.io/visit/aKDGVMCQ) [Remote work is still 2.5× its pre-Covid level, but falling](https://newsletter.dvy.io/visit/aKDGVMCQ) [Remote work in the US peaked at nearly 28 million people in 2021, more than triple the pre-pandemic figure, and has been slowly falling since. By 2024 it sat at around 22 million, still roughly 2.5 times the 2019 level. The headline number suggests a stable "new normal", but the city-level data is more interesting. Sunnyvale, California saw the sharpest swing of any location in the dataset. Remote workers there increased almost 11x between 2019 and 2021 (from around 3,200 to 38,000), then fell 67% between 2021 and 2024. The same city holds both records. That probably reflects how concentrated tech employment is there, and how quickly the big tech firms reversed their remote-first policies once the acute phase of the pandemic ended. For context, most locations clustered between a 150% and 300% increase during the 2019-2021 period. Sunnyvale's 1,100% jump was anomalous. At the other extreme, only one location in the whole dataset saw a decrease in remote work during that period: Rice County, Minnesota, down 7.5%. The author has no explanation for it either. The underlying data comes from the American Community Survey, built into an interactive explorer that lets you generate the same charts for individual states, counties, and cities over 65,000 people.](https://newsletter.dvy.io/visit/aKDGVMCQ) History [A conical, symmetrical mountain peak illuminated by low light under a hazy sky with wisps of cloud.](https://newsletter.dvy.io/visit/hkCpmy9U) [How a Scottish mountain helped scientists work out the mass of the Earth](https://newsletter.dvy.io/visit/hkCpmy9U) [In the summer of 1774, a team of astronomers hauled telescopes, pendulums, and barometers into the Scottish Highlands with one goal: work out how much the Earth weighs. Their tool was a mountain. Schiehallion, a neatly symmetrical peak in Perthshire, was chosen precisely because its regular shape made the maths tractable. The idea came from Newton's law of universal gravitation, published in 1687, which says every object pulls on every other object in proportion to its mass. If you could measure how strongly a mountain pulled on a plumb bob (a hanging weight on a string), and you knew the mountain's approximate mass from its size and likely rock density, you could work backwards to figure out Earth's total mass by comparing the two gravitational forces. It sounds straightforward, but the practicalities were hard. The team spent months measuring tiny deflections in pendulum angles on both the north and south faces of the mountain, trying to detect the subtle sideways tug of the rock. The differences they were looking for were fractions of a degree. The experiment worked. Astronomer Royal Nevil Maskelyne ran the fieldwork, and mathematician Charles Hutton later crunched the results, arriving at a figure for Earth's density that was in the right ballpark. Hutton also invented contour lines to help with the analysis, which turned out to be a more enduring contribution to cartography than anyone expected.](https://newsletter.dvy.io/visit/hkCpmy9U) [A smiling man with a large block of ice on his shoulder, flanked by two women in 1920s coats and hats.](https://newsletter.dvy.io/visit/gNSJuKqc) [Frederic Tudor built a global ice empire from a Massachusetts pond](https://newsletter.dvy.io/visit/gNSJuKqc) Before refrigerators, keeping food cold meant waiting for a knock at the door. The ice man (and during wartime, ice women) made daily rounds through cities and towns, hauling blocks cut from frozen ponds and sliding them into household iceboxes. Arthur Miller remembered them in their leather vests and wet sackcloth, standing dripping at the door while waiting to be paid. The person who built that trade into a global business was Frederic Tudor, a 23-year-old from Massachusetts who loaded his first ship with pond ice and sent it to Martinique. It mostly melted. He kept going anyway. The early numbers were brutal: in the 1820s, only about 10% of harvested ice actually reached the end buyer. Tudor's eventual solution was sawdust, which turned out to be a surprisingly effective insulator, and obsessive attention to storage and shipping logistics. Eventually he was sending ice on 114-day voyages to Australia. By the peak of the trade, American ice was the country's second-largest export after cotton. Tudor had built what was effectively a monopoly, and became one of the wealthiest people in the world. Ice went from something only the rich could afford (harvested from private estate ponds, stored in underground ice houses) to a daily household staple delivered to your door. It also quietly reshaped what people ate and drank. Refrigerated transport opened up long-distance meat, fish, fruit, and vegetable markets. Chilled cocktails became a thing. Food that previously spoiled within miles of where it was produced could now travel. The electric refrigerator eventually killed the ice man's job in a single generation. But the habit of expecting cold food and cold drinks, any time, anywhere, is his legacy. [Book cover showing two families posed outdoors with all their household possessions, contrasting an American family on a driveway with a Bhutanese family on a hillside temple terrace.](https://newsletter.dvy.io/visit/HR37LNrR) [What does a family own? Portraits from 30 countries, front door open, everything dragged outside.](https://newsletter.dvy.io/visit/HR37LNrR) Families from 30 countries, photographed outside their homes with every possession they own pulled into the street. A Malian family's pile looks very different from a Kuwaiti one, or an Icelandic one. Peter Menzel and Faith D'Aluisio spent years making this book, and the images do something statistics rarely manage: they make material inequality feel immediate and personal rather than abstract. [Web](https://newsletter.dvy.io/visit/4uctiAx2) · [Twitter](https://newsletter.dvy.io/visit/zvpdqSE2) · [LinkedIn](https://newsletter.dvy.io/visit/gxVn6vL2)