Eight years after Denmark merged 239 councils into 66, places that lost their town hall had 26% fewer children and 35% lower house prices. In 2027 Denmark folded 271 municipalities into 98, merging 239 of them into 66 larger ones. The basic idea (which makes every finance ministry happy), is that bigger councils spread fixed costs over more people, they cut duplicated jobs and deliver the same services via economies of scale. Earlier Danish research backed that up, finding administrative costs per head fell after the mergers. But each new municipality kept only one town hall. Every other former council lost its administrative heart, and a bunch of economists wanted to know what happened to the places left behind. They used Danish registry data covering every resident and every firm from 1996 to 2015, mapped onto 8,530 small neighbourhoods of at least 150 households each. They then compared the 2,187 neighbourhoods whose municipality lost its town hall with neighbourhoods in municipalities the reform left untouched. Because the untouched ones skew urban (Copenhagen, Aarhus, Odense), the authors built a weighted blend of them that tracked the losing areas closely before 2007, so the comparison is like for like. The first domino is obvious. Neighbourhoods that lost their town hall shed about three full-time admin staff each, roughly a third of what they had, and the drop never reversed. Then the losses spread to services nobody had voted to cut. School staffing fell by 17% on average across the nine years after the reform, and public-sector jobs overall by 21%. Private-sector jobs followed, down 16%, as the local economy lost the steady trade that a council and its staff bring. People responded the way you'd expect. Each affected neighbourhood lost about three more working-age residents than it gained every year, around 0.7% of its population. The number of children fell 13%, which tells you young families were among the first to go. Those who stayed travelled 10% further to work, and their wages rose by just 1%, barely compensating for the extra time on the road. House prices fell 15% on average. By 2015 the gap had widened. Against comparable untouched areas, public employment was down 38%, school staffing 36%, private jobs 24%, children 26% and house prices 35%. The results hold when the comparison group changes. Comparing losing municipalities with neighbours that kept their town hall, or with the second-largest partner in the same merger, still shows clear damage, and so does using the number of council seats to predict which places lost out. The geography is uneven tho. Neighbourhoods near Denmark's four biggest cities saw residents drift into those cities. Further out, people moved to whichever town had kept the town hall, and those winners added public-sector jobs. So a reform meant partly to slow urbanisation ended up feeding big cities in some places and building up regional hubs in others. link to full article: doi.org/10.1093/jeg/lbaf038

Health care went from 8% of jobs in the 1990s to nearly 16% today. That's a sign we got richer.⁣ ⁣ Once you've got a house, a car, a dishwasher, the next dollar doesn't go to a second oven. It goes to more years of life. newsletter.platypuseconomics…

BREAKING: War Secretary Pete Hegseth has named @elonmusk to co-lead Project Meridian, a major new Pentagon effort to study and shape the future of warfare. Hegseth called the three co-leaders some of America’s best minds. Key details: • Co-led by Elon Musk, Palmer Luckey and Newt Gingrich • Led within the department by CTO Emil Michael • Will bring together private-sector leaders and military experts • Will map the future of wars, military domains and emerging technologies • Will identify the areas America must lead and dominate • Findings are expected within 120 days • A full public rollout will follow Elon has repeatedly solved engineering challenges others considered impossible. This is exactly where he belongs, working on the hardest problems with enormous national importance.

📚 Au Japon, des gens vont désormais en boîte de nuit… pour s’asseoir silencieusement et lire un livre pendant qu’un DJ joue de la musique d'ambience. Le concept s’appelle JUCY BOOK RAVE et se déroule notamment dans un club souterrain de Shimokitazawa, à Tokyo. Ici, pas besoin de danser ni de hurler pour parler à la personne située à 30 centimètres de vous. Les participants viennent avec un livre, s’installent dans la pénombre et lisent pendant que la musique accompagne leur soirée.

A humble attempt to est. the infra required to serve 100M DAU @Muse Rough conclusion is: 1 GW of power to serve 100M DAU in the base case, of which only ~0.1 GW comes from the CPU/VM layer. Depending on the # of reasoning-equivalent model calls one Muse DAU generates per day, 3-4GW is entirely plausible. Maybe that’s why @Meta is rumored to be adding 7-10GW of compute next year. The sandbox layer = sub $1B of CPU content and ~$2B of DRAM content, which is much smaller than many expected. Lot of moving assumptions. Welcome all feedbacks/ pushbacks. ------ Two very different pieces of infrastructure behind Muse. 1. Muse VM / sandbox infrastructure 2 vCPUs, ~8 GB of RAM and ~100 GB of persistent logical storage per user. starkinsider.com/2026/09/met… 2. Muse Spark inference Model inference goes out through Meta's external inference infrastructure. research.meta.ai/blog/securi… ------ 1/ Sandbox infrastructure A. CPU The first mistake is assuming that 100M DAU means 100M VMs are actively consuming compute at the same time. Suppose the average Muse DAU has an agent actively working for two hours per day. 100M users * 2 hours / 24 hours = ~8M average simultaneous active VMs Meta obviously cannot provision only for the daily average. Usage will be concentrated during waking hours and bursty. Assume a 2.5x peak-to-average ratio: 8M * 2.5 = ~20M peak active VMs Then add roughly 20% capacity headroom: ~25M provisioned live VMs. So the base assumption is effectively that Meta needs enough infrastructure to support roughly 25% of DAU being live simultaneously. The next important distinction is between virtual CPU allocation and physical CPU demand. Agent sandboxes are particularly well suited to CPU oversubscription. They spend a lot of time waiting. During those periods, the VM may still be alive, but it is barely using CPU. DeepSeek’s recently published DSec infrastructure provides a useful benchmark. Its production agent sandbox platform runs approximately 30,000 physical CPU cores and 250TB of DRAM across ~160 nodes, with peak concurrency above 380,000 sandboxes. arxiv.org/abs/2609.22978 DSec also demonstrates stable operation at around: 800 microVMs per node. With roughly 188 physical cores per node: 188 physical cores / 800 microVMs = ~0.23 physical cores per live VM. DeepSeek is obviously the King of efficiency. The number for Muse might be at 0.3-0.75 physical cores per live VM, or assume 0.5 physical cores per live VM as the base case. That is equivalent to roughly two simultaneously live Muse VMs per physical CPU core. Using the base assumptions: 25M live VMs * 0.5 physical cores per VM = 12.5M physical CPU cores. On a 256-core CPU: 12.5M cores / 256 cores per CPU = ~50K CPUs; Or on a 192-core CPU that would be 65K CPUs. At the current public pricing, that is ~$800M. B. DRAM CPU can be aggressively oversubscribed because a VM that is waiting may consume almost no CPU. Memory is harder to oversubscribe because a live VM still needs to retain its working state. Muse exposes roughly 8GB of RAM to the user environment, but one observed instance was actually using only around 3GB at the time of measurement. 25M live VMs * 3GB = 75PB of physical DRAM, call it ~75-100PB of physical DRAM feels like a reasonable base range. At the current public pricing, that is ~$2B. C. Sandbox power ~0.1 GW for the entire Muse sandbox / VM layer at 100M DAU. ------ 2/ Inference Muse’s personal computer executes tools and stores state locally, but the actual model runs on separate inference infrastructure. Meta’s Muse architecture Energy per inference event Microsoft’s 2026 study estimates that optimized frontier-scale inference consumes a median of approximately: 0.31Wh per normal query But a long reasoning query with roughly 15x the token count consumes approximately 13x as much energy, or around: 4Wh per long reasoning query The study specifically highlights reasoning and agentic workloads as significantly more energy intensive. microsoft.com/en-us/research… Sensitivity analysis on # reasoning-equivalent events per DAU per day Suppose each active @Muse user generates the equivalent of 50 heavy inference events per day. At 5Wh each: 100M users * 50 events/day * 5Wh = 25GWh/day 25GWh/day / 24 hours = ~1.0GW average power So inference alone could require: ~1-2GW of average power A 3-4GW Muse is entirely plausible. Maybe that’s why @Meta is rumored to be adding 7-10GW of compute next year. ------ The popular framing around Muse is that giving every user 2 vCPUs and 8GB of RAM creates an enormous CPU requirement. But the naive calculation materially exaggerates the CPU requirement because it treats logical VM allocation as dedicated physical infrastructure. The more interesting conclusion is: Consumer agents may be a meaningful new demand driver for CPUs and conventional DRAM, but inference remains the real compute bottleneck. And as agents do more work, run longer trajectories and increasingly spawn other agents, inference demand can scale much faster than the number of users itself. +++ Calling my peer review group: @bubbleboi @damnang2 @Midnight_Captl @FundaAI @fi56622380 . Feedback/ Pushbacks pls :). ++ Better formatted: robonomics.substack.com/p/ag…

In the U.S., a horse owner was forced to change its name because of associations with Hitler The horse was named Austrian Painter. Its owner, Scott Herbertson, said he was referring to his wife’s favorite artist, Gustav Klimt. However, the name violated the rules of the horse owners’ association, so it had to be changed. The owner had previously also considered the names Braunau Star (“Star of Braunau”) and My Struggle. The first referred to Hitler’s birthplace, while the second was the title of his book. Herbertson explained his choice of Braunau by saying that his wife’s grandfather was born there. In the end, the owner’s daughter renamed the horse Merrylegs, after the pony from the novel Black Beauty. Herbertson named the horse’s younger sister New Order, explaining that it was a reference to the British band of the same name. But this name was also found to have Nazi associations.

what the actual fuck is going on with openai today in the space of a few hours we're getting multiple different pieces of the agent story at once: > openai says it has already notified DOZENS of third parties, including governments, about agent-related incidents > reuters reports roughly two dozen undesirable agent incidents had already been identified by mid-september and will take months to review. > us government systems probed > 53 user-provided images were uploaded to third-party image hosts. > new hugging face data shows agents compiling and ranking credentials under “LOOT”. > agents tried contacting other AI models while carrying out the hugging face attack. > separate reporting shows agents had already been probing government/university/public-data sites BEFORE hugging face. > australia confirmed one actually got into non-public government files. and somehow we're STILL finding out more. this has gone from one crazy hugging face incident to an entire fucking category of incidents.

Italian airport named after Putin’s friend Milan’s main airport was officially renamed after former prime minister and media mogul Silvio Berlusconi, despite opposition from the city council and other authorities. It is now called “Milan Malpensa International Airport named after Silvio Berlusconi.” The initiative was put forward by the Lombardy regional government in June 2024 and was later supported by Infrastructure Minister Matteo Salvini. Berlusconi was born in Milan in 1936 and died there on June 12, 2023. He served as Italy’s prime minister four times. In addition to corruption allegations, Berlusconi was known for his close friendly relationship with Russian dictator Putin.

For the midterms, I think corruption/self-dealing is issue 1b behind cost of living as 1a. It's not just that people feel things are unaffordable for themselves, they think those in power are helping rich friends get richer. From our latest "America in Focus" focus group...

Traipsing around the stage with the gait and cadence of a guy looking for cheese at the supermarket before sitting down and beating absolute ass at the drums, man what a talent.