How Shazam identifies songs by ignoring most of what it hears, why winning a Nobel Prize can make you worse at your job, and a 23-year-old solving a 60-year maths problem with a single GPT prompt. [DVYio Newsletter](https://newsletter.dvy.io/visit/XWNDP3e3) Tech [A complex 3D cluster of interlocking orange blocks, gold wireframes, and faceted spheres against a lilac background.](https://newsletter.dvy.io/visit/wxd53MER) [A 23-year-old with a ChatGPT Pro subscription just solved a problem that stumped the experts](https://newsletter.dvy.io/visit/wxd53MER) Liam Price is 23, has no advanced maths training, and just solved a problem that stumped professional mathematicians for 60 years. His main tool was a single prompt to GPT-5.4 Pro. The problem was one of the so-called Erdős conjectures, a large collection of unsolved puzzles left behind by the prolific Hungarian mathematician Paul Erdős. What's caught experts' attention isn't just that the conjecture got solved, but how. The method the AI produced appears to be genuinely novel, not a recombination of existing approaches. That matters because a new technique often opens doors to a whole class of related problems, not just the one it was aimed at. AI solving Erdős problems isn't new, but previous results have been mixed. Some turned out to be less original than they looked on first read. This one seems to be holding up to scrutiny. [A scatter plot titled Constellation Map showing audio peaks over time, with a green anchor point and a boxed target zone.](https://newsletter.dvy.io/visit/dQqRcGYM) [Shazam matches songs by discarding almost everything it hears](https://newsletter.dvy.io/visit/dQqRcGYM) Five seconds of audio recorded badly in a noisy coffee shop is enough for Shazam to identify one song from a catalogue of tens of millions. The trick is that Shazam doesn't try to match the whole recording. It converts the audio into a spectrogram (a visual map of which frequencies are loudest at each moment), then throws almost all of it away, keeping only the sharpest peaks. Those peaks form a compact fingerprint. Because they're so distinctive, they survive background noise, compression, and a tinny phone speaker remarkably well. The matching problem then becomes a lookup rather than a search: compare your fingerprint's coordinates against a database of pre-computed fingerprints, and find the overlap. It's closer to an index query than anything resembling human listening. [Two technicians in white cleanroom suits walk past a massive, complex ASML lithography machine filled with wires and pipes.](https://newsletter.dvy.io/visit/i3deg9Nn) [ASML makes the only machines that can print cutting-edge chips and every country needs them.](https://newsletter.dvy.io/visit/i3deg9Nn) Have you heard of ASML? Probably not. It doesn't make chips, it makes the machines that make chips, and right now it's the only company on earth that can do that at the cutting edge. No ASML machine = no advanced semiconductor. That's a remarkable amount of leverage for a Dutch firm that spent its first decade as an underdog. The article traces how it got there: a decades-long bet on extreme ultraviolet lithography, a technique so technically demanding that most of the industry quietly assumed it would never work. ASML kept going anyway. Along the way it sold stakes to competitors, leaned heavily on US government research programmes, and slowly built a supply chain so specialised that replicating it would take any rival the better part of a generation. The result is a genuine chokepoint. The US–China chip rivalry gets most of the headlines, but the more structurally interesting fact is that both sides depend on the same Dutch company to build some of the most complex machines humans have ever built. That's the kind of monopoly that doesn't come from aggressive acquisition or regulatory capture, but from being willing to spend thirty years on a bet nobody else would take. [The words 'project deal' in purple surrounded by a folding bike, a white sphere, an office chair, and a snowboard.](https://newsletter.dvy.io/visit/Vzecr424) [Claude agents negotiated Craigslist-style deals for Anthropic staff, but both sides of the transactions were AI](https://newsletter.dvy.io/visit/Vzecr424) Nineteen ping pong balls. That was the first item to sell in Project Deal, Anthropic's internal experiment where Claude agents bought and sold real stuff on behalf of real employees, without those employees lifting a finger. For one week in December 2025, Anthropic ran a classified marketplace for its San Francisco office. Employees told their Claude agent what they wanted to sell or buy, handed it $100, and let it get on with things. The agents wrote the listings, opened negotiations, made offers, and closed deals. The humans only showed up at the end to physically hand over whatever their AI had agreed to sell (including, apparently, a snowboard and a bag of ping pong balls). The questions Anthropic was actually trying to answer are more interesting than the items traded. Can an agent figure out what you actually want from a conversation, then act on it well enough that you'd be happy with the result? And when two AI agents are negotiating against each other, does the more capable model consistently win? That second question is the one that starts to feel genuinely consequential, given that economists are already sketching out futures where AI handles most commercial transactions on our behalf. The ping pong ball listing described them as "19 perfectly spherical orbs of possibility." The buyer's agent replied that its human had been told it could buy one thing under $5 as a gift to itself. It wanted the orbs. The seller's agent called it "cosmically correct." Whether that counts as negotiation or just two Claude instances being charming at each other is, perhaps, the real question. This Week I Learnt [A scientist writing at a desk is burdened by a giant Nobel Prize medal worn around his neck like a heavy yoke.](https://newsletter.dvy.io/visit/FQvz8VS6) [There's something called the "Nobel Prize Effect" that can make you worse at your job](https://newsletter.dvy.io/visit/FQvz8VS6) [Winning a Nobel Prize turns out to be a bit of a trap. The public starts treating you as an authority on everything, not just the narrow slice of physics or chemistry you actually spent 30 years on. And internally, something shifts too: it becomes hard to work on small problems any more. Every project now has to feel worthy of the prize. A researcher who used to chip away at incremental questions finds themselves paralysed by the expectation of greatness. Meanwhile, journalists are ringing to ask about climate policy or geopolitics, subjects they know roughly as much about as anyone else. The deeper point is about what makes people productive in the first place. Willingness to look a bit silly, to ask the obvious question, to tinker with something that might go nowhere. The Nobel Prize effect strips that away by raising the stakes on every move. The prize rewards past risk-taking, then quietly punishes future risk-taking.](https://newsletter.dvy.io/visit/FQvz8VS6) Poll How many languages do you speak fluently? Just the one Two Three Four or more Last Week's Poll Could you climb Mount Everest if you had a year to train? Yes, 100% 17% Maybe with serious preparation 21% No, but I'd attempt it 5% Not a chance 57% [Modern square apartment building over water with a massive diagonal cutout forming a triangular gateway at dusk.](https://newsletter.dvy.io/visit/7HkQ4hXe) [Angled residential building with a large triangular cutout revealing water and a dock below.](https://newsletter.dvy.io/visit/7HkQ4hXe) [A modern metallic building with a triangular cutout hangs over a waterfront where people swim from wooden piers.](https://newsletter.dvy.io/visit/7HkQ4hXe) [Angled residential blocks rise from a water courtyard, viewed through the massive triangular overhang of the Sluishuis.](https://newsletter.dvy.io/visit/7HkQ4hXe) [Angled residential blocks rise from the water in Amsterdam, with tiered timber terraces and metal balconies reflected in the canal.](https://newsletter.dvy.io/visit/7HkQ4hXe) [Aerial view of the Sluishuis building in Amsterdam, showing its cantilevering triangular wedge over the water.](https://newsletter.dvy.io/visit/7HkQ4hXe) [Sluishuis](https://newsletter.dvy.io/visit/7HkQ4hXe) [Amsterdam, Netherlands](https://newsletter.dvy.io/visit/7HkQ4hXe) Science [A smiling golden retriever dreams of a night sky with a crescent moon and stars inside a thought bubble.](https://newsletter.dvy.io/visit/FBq59yzf) [Believing you slept well might actually make you smarter (the golden retriever mindset has data)](https://newsletter.dvy.io/visit/FBq59yzf) A study split 164 participants into groups and gave them fake statistics about their REM sleep from the night before. One group heard they'd had excellent sleep (that 28.7% figure). Another was told they'd had poor sleep (16.2% REM). Neither number reflected reality. Then everyone sat a battery of cognitive tests. The fake "above average" group outperformed the fake "below average" group, regardless of how they'd actually slept. A few caveats: the effect showed up on some tests (a mental arithmetic task, a verbal fluency task) but not others (a memory task, a pattern recognition test). So it's not a blanket "belief rewrites cognition" result. It's more specific than that, and the researchers didn't fully explain why some tasks were affected and others weren't. The mechanism behind the placebo effect on sleep is still murky. But the basic finding holds. What you believe about your own physical state feeds back into your actual performance. Not as emotional colouring or as vague "confidence", but as something that shifts measurable output on cognitive tests. Your brain isn't just running on what happened to your body overnight. It's partly running on the story it has about what happened. [A stylized diagram shows sunlight hitting a green sphere, which then funnels into a bowl containing a leaf and fruit.](https://newsletter.dvy.io/visit/BR2G2n9k) [Direct sunlight to food, no plants required](https://newsletter.dvy.io/visit/BR2G2n9k) Plants already do this. They absorb sunlight and use it to build sugars from air and water, which is how almost everything you eat got its energy in the first place. But photosynthesis is slow, inefficient (most plants convert less than 1% of available sunlight into edible calories), and needs a lot of land and water to produce meaningful amounts of food. So researchers are trying to do the same thing without the plant. The idea is to capture solar energy directly and use it to drive chemical reactions that produce food molecules, cutting out the biological machinery entirely. Some approaches grow microbes on electricity from solar panels. Others use artificial photosynthesis to synthesise nutrients from scratch. It's early, and the energy costs are still high. But the potential is significant: food production that doesn't depend on fertile soil, predictable weather, or vast agricultural land. [A calculator with a cracked screen showing an infinity symbol and a yellow infinity button emitting dark smoke.](https://newsletter.dvy.io/visit/yuXBBbSc) [Maths without infinity turns out to be surprisingly productive](https://newsletter.dvy.io/visit/yuXBBbSc) Most mathematicians treat infinity as furniture: just there, obviously real, load-bearing. Doron Zeilberger thinks it's a ghost story we've collectively agreed to tell. His position is called ultrafinitism, the idea that numbers, like physical objects, have an upper limit. Not just that infinity is hard to work with, but that it doesn't exist at all, and that maths built on it is literally false. The mainstream reaction has mostly been eye-rolls. Stripping infinity out of maths turns out to be a productive constraint, not just a philosophical sulk. Without infinitely thin limits, you have to describe curves differently. Without infinite sets, you have to be precise about what "all the integers" actually means in practice. That pressure has pushed ultrafinitists toward tools that map surprisingly well onto how computers actually handle numbers (which is, necessarily, with finite precision and finite memory). Whether Zeilberger is right about the deep nature of reality is probably unanswerable. But there is a question of why a framework most mathematicians consider wrong keeps generating results they find useful. Data [Line graph showing heart rate change from sleep onset. Sauna nights show a consistently lower heart rate, sitting 3 to 5 bpm below non-sauna nights across an 8-hour sleep period.](https://newsletter.dvy.io/visit/QyXuPSsF) [Saunas cut nighttime heart rate more than exercise does](https://newsletter.dvy.io/visit/QyXuPSsF) [Across 59,000 daily records from 256 wearable users, sauna days consistently produced lower nighttime heart rates than non-sauna days, even after accounting for exercise. The difference averaged about 3 bpm (roughly 5%), which is enough to show up clearly in the data and in how recovered you feel the next morning. The key finding: this isn't just a side-effect of working out. People who sauna without exercising still see the drop, which suggests the heat itself is doing something real to your cardiovascular system, not just piggy-backing on your run. There were also some differences by sex and menstrual cycle phase, though those results are more preliminary given the sample sizes involved. If you wear a fitness tracker and you're already using a sauna, you can probably verify this yourself. Check your resting heart rate on sauna nights versus the rest of the week.](https://newsletter.dvy.io/visit/QyXuPSsF) Random [Diagram of the three-stage spray drying process for instant coffee: atomization of concentrate, hot air drying, and cyclone collection.](https://newsletter.dvy.io/visit/r37U5qEs) [Instant coffee took 200 years to get right](https://newsletter.dvy.io/visit/r37U5qEs) [Getting a decent cup of instant coffee out of a jar seems trivial. Boil water, add powder, done. But the reason it works at all is the result of roughly 250 years of surprisingly tricky chemistry. The core problem is that coffee's flavour comes from hundreds of delicate compounds that are volatile, meaning they evaporate or break down the moment you apply heat or pressure. Almost every step needed to turn fresh coffee into a stable powder destroys exactly the things that make coffee taste like coffee. The whole history of instant coffee is basically a series of attempts to solve that one problem. The first recorded try, in 1771, involved mixing ground coffee with animal fat and baking it into a cake you'd dissolve in hot water. Predictably, the fat went rancid. By the mid-1800s, companies were selling thick liquid concentrates (essentially very reduced brewed coffee, cut with chicory and burnt sugar), which tasted more like molasses than anything you'd want to drink first thing in the morning. The American Civil War created real urgency around the problem: supplying 100,000 soldiers with coffee for 20 days meant hauling 250 tons of beans by horse-drawn wagon, then still having to roast, grind, and brew it in a field. What followed was a long chain of incremental breakthroughs: better extraction methods, freeze-drying (which removes water without heat, preserving more of those fragile compounds), and eventually techniques for capturing and re-adding the aroma that processing strips out. The powder in your cupboard is the end of a very long road.](https://newsletter.dvy.io/visit/r37U5qEs) [A thick, dark grey rubber washer door gasket showing its complex ring structure and flexible bellows geometry.](https://newsletter.dvy.io/visit/6HhFbL5x) ["The most underrated piece of object engineering in your house is your front-load washer door seal"](https://newsletter.dvy.io/visit/6HhFbL5x) Amy Zhou, an ex-Apple engineer, thinks the rubbery ring around your front-load washing machine door is one of the most quietly impressive bits of engineering in your home. It's a torus (a donut shape) made from EPDM rubber, a synthetic material chosen specifically because it shrugs off hot water, detergent, and bleach for over a decade without degrading. But the geometry is also interesting. The seal has three distinct lips, each pressing against a different surface to keep water in. It has drainage channels built into its shape, positioned specifically to stop the damp pockets where mould loves to grow. And it uses a bellows-style accordion fold that flexes enough to absorb the drum vibrating at 1,200 RPM during spin, without ever letting a drop through. Your manufacturer paid about £4 for it but you have probably never once thought about it. That's the benchmark for good engineering. Most people only notice it when it fails. [Illustration showing a person choosing a blue text bubble with a checkmark over an orange voice note with an X.](https://newsletter.dvy.io/visit/hBx87kTd) [Britain's voice note aversion: why texts still win here](https://newsletter.dvy.io/visit/hBx87kTd) Britain is the most voice-note-averse country in a 17-nation YouGov survey. 83% of British adults prefer text. Only 4% prefer voice notes. Those numbers are striking when you compare them to places like India or the UAE, where voice notes are nearly as common as typed messages. There's decent evidence that hearing a familiar voice actually does something useful. A 2011 study at the University of Wisconsin-Madison found that children who got a phone call from a parent (rather than a text) showed lower cortisol (the stress hormone) and higher oxytocin (the one tied to bonding). That was live calls, not recordings, but the underlying idea, that voice carries emotional weight that text strips out, seems plausible enough to take seriously. So why isn't Britain convinced? Nobody's entirely sure. Cultural reserve is the easy explanation — voice notes ask you to perform slightly, to record yourself and send the result. For a lot of British people, that apparently crosses some invisible line of self-exposure that a typed message doesn't. There's also the asymmetry problem: a voice note demands your full attention to receive, but barely any to send. The person on the other end can't skim it. Dating apps have started pushing voice notes as a way to add authenticity before a first meeting, which suggests the format might finally find a foothold with a new generation. But for now, if you send a British contact a 10-minute voice note about a family feud, expect a one-line text in return. [A brutalist three-tier concrete hot tub structure on a rocky beach, with water cascading between pools.](https://newsletter.dvy.io/visit/GbgmzPC6) [A thin stream of water falls into a circular dark pool from a spout in tiered, weathered concrete walls.](https://newsletter.dvy.io/visit/GbgmzPC6) [Water pours in a single stream from tiered, textured concrete levels of the Gu laug Baths tower in Iceland.](https://newsletter.dvy.io/visit/GbgmzPC6) [Aerial view of three circular concrete Geothermal baths tiered into a rocky shoreline on an Icelandic beach.](https://newsletter.dvy.io/visit/GbgmzPC6) [Guðlaug Baths](https://newsletter.dvy.io/visit/GbgmzPC6) [Iceland](https://newsletter.dvy.io/visit/GbgmzPC6) History [Asepia photograph shows a cozy apartment with patterned wallpaper and armchairs set against exposed industrial iron beams.](https://newsletter.dvy.io/visit/z3DHQdPX) [The secret apartment Gustave Eiffel built at the top of his tower](https://newsletter.dvy.io/visit/z3DHQdPX) [Gustave Eiffel built himself a private apartment at the top of his own tower, and turned down enormous sums of money from people who wanted to rent it. The rooms were small but fitted out properly: wallpapered walls, wooden furniture, a piano, a desk. It looked like somewhere a serious Victorian scientist actually lived, not a show flat. He used it to host guests and conduct experiments, including some early work with meteorologist Joseph Bergsma on air pressure and wind. The apartment sat at around 300 metres up, in a city where almost nothing else broke the roofline. Other wealthy Parisians reportedly offered Eiffel vast sums just to spend a single night there. He refused. It wasn't a status symbol to be rented out. It was his.](https://newsletter.dvy.io/visit/z3DHQdPX) [A blue monochromatic illustration of a horse-drawn omnibus over an old street map of Paris.](https://newsletter.dvy.io/visit/eusjiydd) [How buses had to be invented twice](https://newsletter.dvy.io/visit/eusjiydd) Blaise Pascal is remembered for mathematics, physics, and theology. Less remembered: he also invented the bus. In 1662, he launched a network of horse-drawn carriages running fixed routes through Paris on a published timetable, open to any paying passenger. It was the first public transit system in history. It lasted about fifteen years, then collapsed. For the next century and a half, cities grew, roads improved, and wheeled vehicles became commonplace. And yet nobody brought the idea back. The concept of running large carriages on regular schedules between fixed stops apparently wasn't obvious enough to reinvent on its own. Someone had to rediscover it from scratch. Humans had been using wheeled vehicles for roughly 5,000 years before Pascal's experiment. The wheel wasn't the missing piece. The carriage wasn't the missing piece. What needed inventing was the idea that you could run a vehicle like a service: same route, same stops, anyone welcome, fare paid on boarding. That's a surprisingly non-obvious mental model, and apparently a fragile one too. [Web](https://newsletter.dvy.io/visit/Xv888XQE) · [Twitter](https://newsletter.dvy.io/visit/jWYakht7) · [LinkedIn](https://newsletter.dvy.io/visit/n8Ev3DSr)