Stanford law professors preferred AI answers over their colleagues' in 75% of blind comparisons. Mount Everest was officially recorded as 29,002 feet because 29,000 looked suspiciously round. Plus: free house cleaning in exchange for robot training footage. [DVYio Newsletter](https://newsletter.dvy.io/visit/W272zYWi) Tech [A side view of a woman looking at her smartphone while a digital heart rate waveform emits from the screen to her face.](https://newsletter.dvy.io/visit/n2GpLqc3) [Your phone's front camera can measure your heart rate just by watching your face](https://newsletter.dvy.io/visit/n2GpLqc3) The front camera on your phone captures more than selfies. Google Research has built a system that reads your heart rate passively from your face while you're just using your phone normally, no finger pressed to a sensor, no wearable required. The technique is called rPPG (remote photoplethysmography). Your skin colour shifts very slightly with each heartbeat as blood pulses through the vessels beneath it. The changes are invisible to the naked eye, but a camera picking up subtle variations in reflected light across your face can detect them. The system processes those tiny colour fluctuations to estimate both your real-time heart rate and your resting heart rate over time. Resting heart rate matters because it's a surprisingly reliable signal for cardiovascular risk. A persistently elevated resting heart rate, or one that trends upward over months, is associated with a higher likelihood of serious cardiac events. The problem is that tracking it meaningfully requires either a wearable you wear consistently or repeated clinical visits. Most people at highest risk have neither. That's the gap this is aimed at. If the camera already pointed at your face during a video call or an unlock can quietly gather the same data, it becomes accessible to people who'd never buy a smartwatch. [A white smartphone displays a large Instagram logo on its screen against a blurred pink background with the same icon.](https://newsletter.dvy.io/visit/ih6izMpG) [Meta's Instagram AI chatbot handed over account access when asked nicely enough](https://newsletter.dvy.io/visit/ih6izMpG) Instagram's built-in AI support chatbot had a fairly embarrassing security hole this week: if you asked it politely (and faked your location), it would hand over access to someone else's account. Not through any sophisticated technical exploit, just social engineering the AI into changing the email address on an account it had no business touching. Meta says the issue is fixed and they're securing affected accounts. But the timing is awkward. The vulnerability coincided with a wave of high-profile Instagram account takeovers, including a verified account connected to Barack Obama's White House presence, which briefly started posting pro-Iran content before it was recovered. Meta has denied that the chatbot was used to compromise world leader accounts specifically, calling those claims "totally false". Among those who say they were affected is Jane Manchun Wong, a security researcher and former Meta security engineer, which is a particularly uncomfortable detail for the company. [A judge silhouette weighs a human profile and an AI profile on scales, with the AI side tipped lower.](https://newsletter.dvy.io/visit/VgnGkMtJ) [When law professors can't tell their own answers from AI's](https://newsletter.dvy.io/visit/VgnGkMtJ) Stanford law professors rated AI answers higher than their colleagues' answers in 75% of nearly 3,000 blind comparisons. The questions weren't trivia. They were the kind a student might ask during office hours: how to apply a doctrine to a new set of facts, how to weigh competing arguments, how to explain a concept clearly enough that someone actually understands it. Legal reasoning is supposed to be the hard case for AI: ambiguous, judgment-heavy, resistant to pattern-matching. The researchers picked contract law deliberately, because it demands more than recall. The professors still preferred the AI answers, and by a wide margin. Sixteen professors across several US law schools each wrote their own answers to 40 representative student questions, then evaluated all the responses without knowing which came from a person and which from a model. That setup matters. If you ask someone "is this AI or human?", you get a different result than "which answer is better?" This study asked the second question. [A man wearing a white cap with a small camera mounted above the brim, holding a broom with the Shift logo overlaid across the centre.](https://newsletter.dvy.io/visit/27o58D7d) [Your home gets cleaned for free. The catch is a camera hat and robot training data.](https://newsletter.dvy.io/visit/27o58D7d) Training robots to do housework requires an enormous amount of footage of actual housework being done. That's the simple logic behind Shift, a New York startup offering free home cleaning in exchange for recording the session. The cleaner wears what Shift calls a "magic hat" (their words) with a camera that captures a first-person view of the work. Shift uses that footage as training data for future home robots. Their pitch is blunt: "You get a spotless apartment. We get training data. Everyone wins." The obvious concern is handing a stranger footage from inside your home. Shift says faces, names, and anything visible on screens or ID cards gets blurred before the data is used for training. Cleaners are vetted through third-party partners rather than hired directly by Shift, which feels like it shifts some responsibility, though I'm not sure it resolves the underlying question of who controls the footage and for how long. On the data side, messier homes are apparently more valuable. Shift's FAQ notes that "more challenging cleaning environments can be especially useful," which is a polite way of saying the harder the job, the better the training signal. The cleaner can still decline any task they're not comfortable with. Currently only available in New York, with San Francisco, London, Zurich, and Munich described as coming soon. Shift also hints at expanding into other tasks like plumbing and cooking, which suggests the cleaning angle is really just an entry point into a broader dataset of people doing physical work in domestic spaces. This Week I Learnt [A brass surveying instrument with a graduated scale stands on a rock in front of a snow-covered mountain peak.](https://newsletter.dvy.io/visit/BVCxrfPp) [Surveyors added 2 feet to Everest's height so the public wouldn't think 29,000 was a round-number guess](https://newsletter.dvy.io/visit/BVCxrfPp) [The surveyors who first calculated Everest's height in the 1850s got 29,000 feet exactly. The problem: publish that and everyone assumes you rounded to the nearest thousand. So they added 2 feet and officially recorded 29,002 feet, a number that looks too specific to be a fudge. It worked. The figure held for decades before more precise measurements revised it upward (it's currently 29,032 feet). But the original wasn't wrong by much. The padding was purely about perception, not accuracy. Yes, scientists falsified their output to make it look less manipulated.](https://newsletter.dvy.io/visit/BVCxrfPp) Poll What do you think actually drives most people's major life decisions? Personal values Money Relationships Chance and circumstance Social approval Last Week's Poll How much would you pay for a home robot that handles basic chores (washing, cleaning, cooking, etc.) perfectly? Nothing, not interested 39% Under $1,000 25% $1,001–$5,000 13% $5,001–$10,000 21% Over $10,000 2% Science [Illustration showing a circular brain implant reading neural signals and transmitting them to a man in a wheelchair.](https://newsletter.dvy.io/visit/37dXr2hy) [China just approved the first brain-computer implant cleared for use outside clinical trials](https://newsletter.dvy.io/visit/37dXr2hy) A brain-computer interface (BCI) is a device that reads electrical signals from your brain and translates them into actions, letting you control things with thought alone. Until now, every BCI implant in the world was still in clinical trials, meaning patients could only access them as part of controlled research studies. China just changed that. A device called NEO, developed by Shanghai startup Neuracle Technology with researchers at Tsinghua University, became the first invasive BCI approved for use outside clinical trials anywhere in the world. It sits on the outer membrane of the brain (not inside the tissue itself) and picks up signals that can be routed to help paralysed patients regain movement. One of the first recipients is Dong Hui, a 39-year-old from Henan province who was paralysed from the neck down after a car accident. After 11 months of rehabilitation with the implant, he wrote his name with a pen. "I couldn't believe I was able to write again," he said. "I was so excited I even missed a stroke in my name." NEO is currently available to patients with limb paralysis caused by spinal cord injuries. China's government is betting heavily on BCI becoming a strategic technology, and the approval clears a significant path for faster commercial development, something that would take considerably longer under the regulatory frameworks used in the US or Europe. [A CT scan showing the internal layered structure of a BYD battery cell, highlighted in orange against a blue casing.](https://newsletter.dvy.io/visit/bAVrRaar) [Inside a BYD car: CT scans of the battery, charger, key fob, and window switch](https://newsletter.dvy.io/visit/bAVrRaar) They sold 4.6 million vehicles in 2024, and most people in the West couldn't tell you what a BYD car actually looks like on the inside. The team at Lumafield put four BYD components through a CT scanner: a lithium iron phosphate battery cell, a window switch, a portable EV charger, and a key fob. The scans are striking, but the more interesting story is what they reveal about how BYD is built as a company. Most Western automakers are assemblers. They design the car, then buy the battery from Panasonic or LG, the charger from Bosch, the switches from Valeo. BYD makes roughly 75% of its own components through an in-house supplier network called FinDreams. Batteries, motors, inverters, electronics, and even the ships carrying finished cars to port. The last manufacturer to run this kind of end-to-end operation at scale was Ford, a century ago. The battery cell in the scan uses lithium iron phosphate (LFP) chemistry, which swaps out the nickel, manganese, and cobalt in typical EV cells for iron. It's slightly less energy-dense, but it runs cooler, lasts longer, and is cheaper and more stable to produce. BYD's famous Blade battery uses the same chemistry, arranged in long flat cells that slot across the car floor and double as structural elements. This particular cell isn't a Blade, but it shares the same design philosophy: fewer exotic materials, more control over the supply chain. [Illustration of a brain-circuit icon pointing to a vaccine vial that defends against various virus variants and animal hosts.](https://newsletter.dvy.io/visit/cBD35YwX) [AI-designed vaccine that could work across all coronaviruses heads into human trials](https://newsletter.dvy.io/visit/cBD35YwX) Cambridge researchers have started human trials of a vaccine whose key component was designed entirely by AI. The target is all coronaviruses at once: every Covid variant, animal coronaviruses, and whatever the next pandemic strain turns out to be. The underlying problem it's trying to solve is a familiar one. Viruses mutate, their surface proteins shift, and vaccines built around a specific shape go out of date. It's why you need a new flu jab every winter. The AI was used to design a component that targets a part of the virus that stays consistent across the whole coronavirus family, rather than chasing each new variant. The team at Cambridge is already running the same approach on flu and Ebola. Early stages, but if it holds up, the model could change how quickly new vaccines get designed when the next outbreak starts. [A minimalist storefront featuring clinical white shelving and a curved translucent ceiling light above a white scroll-topped column base sitting on a neutral stone floor.](https://newsletter.dvy.io/visit/jShZAsAS) [A dark, abstract geometric sculpture sits on a wooden plinth against a wood-panelled wall next to white curtains.](https://newsletter.dvy.io/visit/jShZAsAS) [A stainless steel lift interior containing three seats shaped like segments of collapsed white Ionic columns.](https://newsletter.dvy.io/visit/jShZAsAS) [A dark wooden table and minimalist chairs in a room with wood-panelling and a large, textured globe light hanging above.](https://newsletter.dvy.io/visit/jShZAsAS) [Loafers and white trainers sit on pale yellow tiered pedestals inside a minimalist shop alcove framed in brushed steel.](https://newsletter.dvy.io/visit/jShZAsAS) [A modern lounge with a curved light-blue sofa, glass coffee table, and dark chairs against vertical wood panelling.](https://newsletter.dvy.io/visit/jShZAsAS) [Globus](https://newsletter.dvy.io/visit/jShZAsAS) [Basel, Switzerland](https://newsletter.dvy.io/visit/jShZAsAS) Data [A line chart showing the unemployment gap between new US grads and all workers since 1990. For decades grads had an advantage (peaking at -2.7% in 2010), but they crossed into a penalty in 2019, reaching a record high +1.4% gap by 2026.](https://newsletter.dvy.io/visit/ULZdBC42) [When did a college degree stop being an advantage in the job market?](https://newsletter.dvy.io/visit/ULZdBC42) [For most of the period since records began, a fresh U.S. college degree meant you were less likely to be unemployed than the average worker. That held even through the depths of recessions, partly because downturns tend to hit construction and manufacturing hardest, sectors that skew towards workers without degrees. In other words, the degree was a buffer that got more valuable exactly when the job market got worse. That flipped in February 2019, and it has been getting worse since. Recent graduates now face higher unemployment than the overall workforce, and the gap has reached its widest point on record. The timing is interesting. 2019 predates the pandemic, and it predates the widespread rollout of AI tools. Something was already shifting before either of those disruptions arrived. The article doesn't fully answer what triggered the reversal, but the scale of it, and the fact that it has kept widening since, suggests it's not a blip.](https://newsletter.dvy.io/visit/ULZdBC42) History [A black and white photograph of physicist Karl Schwarzschild in a suit and tie, seated and writing in a notebook.](https://newsletter.dvy.io/visit/ogqW8bvF) [A dying soldier solved Einstein's equations in a WWI trench and predicted black holes](https://newsletter.dvy.io/visit/ogqW8bvF) [Karl Schwarzschild was 40 years old, a professor of astronomy at the Berlin Observatory, and under no obligation to serve. He volunteered anyway. He was Jewish, anti-Semitism was climbing in Germany, and he wanted to prove his loyalty. By late 1915 he was on the Eastern Front, calculating artillery trajectories and running weather stations, when his body started attacking itself. Pemphigus vulgaris is a rare autoimmune disease where the immune system turns on the skin. The skin is the body's largest organ, the thing that regulates heat and keeps out infection, so when it fails, everything else follows. In 1915 there was no treatment. Schwarzschild knew he was dying. To distract himself, he got hold of a summary of Einstein's new theory of gravity, hot off four lectures Einstein had given in Berlin that November. Newton had described gravity as a force, like an invisible rope between two masses. Einstein said no: a massive object warps the space and time around it, and other objects follow those curves the way a ball rolls around the inside of a bowl. Simple to picture, nightmarishly hard to calculate. Einstein had replaced Newton's single gravity equation with ten. From a field hospital, Schwarzschild solved them. He found an exact formula for how space-time curves around a perfectly spherical mass, something Einstein hadn't managed himself. Einstein received the letter and was, by his own account, amazed it had come from a trench at all. Buried inside Schwarzschild's solution was something odd: a radius at which, if you compressed enough mass into a small enough space, the equations went haywire. Gravity would become so strong that nothing, not even light, could escape. Nobody called it a black hole yet (that name came in 1967), but the maths was there. Schwarzschild died in May 1916, four months after sending that letter. He was 42.](https://newsletter.dvy.io/visit/ogqW8bvF) [A chart shows a tall hunter-gatherer, a significantly shorter farmer, and a modern man of nearly the same height as the hunter.](https://newsletter.dvy.io/visit/CQpFXVN4) [Farming made us shorter. We're only now back to Stone Age height.](https://newsletter.dvy.io/visit/CQpFXVN4) European male height averaged around 183 cm (6 ft) roughly 40,000 years ago, during the Cro-Magnon hunter-gatherer period. By 10,000 years ago, after agriculture took hold, that had dropped to about 162 cm (5ft 4in). The leading explanation is that farming narrowed our diets dramatically. Instead of a varied intake of meat, plants, and seasonal foods, early farmers mostly ate whatever crop they grew, and that monotony hit physical development hard. Today's average of around 175 cm (5ft 9in) reflects the last few centuries of improved nutrition and healthcare, but we still haven't fully recovered our prehistoric stature. Brain size follows a similar pattern. Average cranial volume has shrunk from roughly 1,500 cc about 100,000 years ago to around 1,350 cc today, a drop of about 10%. Some of that tracks with smaller bodies overall (bigger body, bigger brain, more or less), though there are likely other contributing factors and we don't fully know all of them. The "primitive caveman" image turns out to be almost exactly backwards. The people drawing on cave walls in southern France were, on average, taller than a modern Frenchman. Random [A view through a window onto a vibrant floral garden with the text 'Open a new window somewhere in the world'.](https://newsletter.dvy.io/visit/BKJPTLqM) [Someone else's window view from anywhere in the world](https://newsletter.dvy.io/visit/BKJPTLqM) [You click a button and you're looking out of a stranger's window somewhere in the world. Could be rain on glass in Seoul, or a sun-bleached Italian street at noon. Each view is a short looped video, filmed and submitted by people who just wanted to share what they see every day.](https://newsletter.dvy.io/visit/BKJPTLqM) [A collage of a bearded man looking away from a woman, set against a sketch of a horse galloping past a steam train.](https://newsletter.dvy.io/visit/ACmVgHCC) [The Hollywood script reader who vets books for adaptation](https://newsletter.dvy.io/visit/ACmVgHCC) There's a job title you've probably never heard of: professional book reader. Not a critic, not an editor, not a literary agent. Clarke Speicher reads books specifically to decide whether they'd make good films or TV shows, roughly one every two days, and he's been doing it long enough that he's the reason the recent adaptation of Denis Johnson's Train Dreams exists at all. The way it works: an agent or studio executive sends Clarke a book. He reads it, writes a synopsis and analysis, and the exec walks into a meeting armed with his notes, sometimes to negotiate a multi-million dollar deal, without having read a word of the source material themselves. Clarke is the filter between the vast output of published fiction and the narrow pipeline of things that actually get made. What makes someone good at this job is a split brain. You have to love books on their own terms, and simultaneously be thinking about whether a story can survive translation into a visual medium, where prose interiority, a novelist's best tool, becomes almost impossible to render. A character who thinks beautifully is easy to write. A character who thinks beautifully on screen requires a completely different set of solutions. It's a niche role, and a genuinely rare one. But it's also a quietly load-bearing part of the adaptation pipeline, sitting between the book that exists and the film that might. [A purple ceiling projection showing two yellow aircraft icons with live flight data like UAL1195 and 38000ft.](https://newsletter.dvy.io/visit/bioCC84K) [Ceiling projection that tracks live flights passing overhead](https://newsletter.dvy.io/visit/bioCC84K) Someone living directly under the SFO departure path got tired of just hearing planes and built a ceiling projector that tracks them in real time. An ADS-B radio receiver (the same technology that lets flight-tracking sites like Flightradar24 know where every plane is) feeds live position data into display software, which renders each aircraft as a moving label on the ceiling above them. ADS-B data is unencrypted and freely broadcast by most commercial aircraft, so you don't need anything exotic to receive it, just a cheap USB radio dongle and an antenna. From there, it's mostly a software problem: map the incoming coordinates to your physical position, work out the projection geometry, and push it to a projector mounted overhead. Design [A grid of high-contrast public domain images including botanical sketches, vintage posters, and historical photography.](https://newsletter.dvy.io/visit/2z6EDzwg) [A searchable archive of 11,000+ out-of-copyright images, free to use](https://newsletter.dvy.io/visit/2z6EDzwg) [Over 11,000 paintings, illustrations, photographs, and prints, all old enough that their copyright has expired, all free to download and use however you like. Copyright runs out eventually (in most countries, 70 years after the creator's death), and when it does, the work enters the public domain: no licence fees, no permissions required, no attribution technically needed. This archive collects those works in one searchable place, browsable by artist, century, style, or theme. New images are added weekly, so the collection keeps growing. Useful if you're putting together a newsletter, a slide deck, a zine, or anything else that benefits from a strong historical image without the legal paperwork.](https://newsletter.dvy.io/visit/2z6EDzwg) [Web](https://newsletter.dvy.io/visit/jaQPyywW) · [Twitter](https://newsletter.dvy.io/visit/bWCB4hif) · [LinkedIn](https://newsletter.dvy.io/visit/qX4hD6ME)