Low sugar intake in the first years of life has long-term health benefits, research shows. CC-licensed photo by Rosmarie Voegtli on Flickr.
You can sign up to receive each day’s Start Up post by email. You’ll need to click a confirmation link, so no spam.
A selection of 9 links for you. Sweet. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
Ukraine’s drones knock out AI data center belonging to “Russia’s Google” • Ars Technica
Jeremy Hsu:
»
Ukrainian drone strikes have knocked out two of five data centers belonging to the Russian tech giant Yandex. The damage has left Yandex, known as “Russia’s Google” because it operates the country’s most popular search engine, struggling to maintain its AI chatbot and other online services.
The Ukrainian drones struck Yandex data centers in the Sasovo and Kaluga regions of Russia on October 8 and October 9, according to Reuters. Those actions came as a direct response to a Russian drone strike campaign targeting Ukraine’s data centers, which has disrupted mobile phone and Internet services for thousands of Ukrainians since late September.
“We continue to respond to Russia’s attacks on Ukraine, and we’re doing so with high precision,” said Ukrainian President Volodymyr Zelenskyy in a post on X. “Another data center was hit as well—our symmetrical response.”
The Ukrainian company Firepoint said its FP-1 one-way attack drones were involved in the latest strike on the Kaluga data center that is Yandex’s largest such facility, Reuters reported. The first Ukrainian attack targeted Yandex’s Sasovo data center that houses two of the company’s three supercomputers used for training the company’s YandexGPT AI model.
Yandex said no employees were hurt but acknowledged disruptions to some services. Russian users reported problems accessing media websites and popular apps for the Ivi video streaming platform, the digital banking service T-Bank and the state-owned Russian Railways, according to the independent Russian news publication Meduza.
Other impacted Russian services included a dating app, a real estate aggregator, a self-publishing book platform and the website of the most successful professional soccer club in the Russian Premier League, according to Reuters.
«
It also had the effect of significantly reducing the activity of anonymous politically inflammatory right-wing accounts on various social media platforms. Odd, that. Keep going, Ukraine.
unique link to this extract
OpenAI withdraws three preprints a day after releasing 722 manuscripts on unsolved math problems • Retraction Watch
Alicia Gallegos:
»
Less than 24 hours after OpenAI posted over 700 preprints describing major math results by an internal AI model, the company has withdrawn three of the manuscripts for errors.
On Oct. 6, OpenAI published 722 preprints in a GitHub repository about purported progress in solving 372 math problems in geometry, computer science, algebra and other subjects. The company called the release an effort “to push the frontier of human knowledge and enable further progress in mathematics.”
But on Oct. 7, OpenAI announced the withdrawal of three manuscripts because of a sign error. The error invalidated an argument in one manuscript and the construction used by two dependent papers, according to the announcement. Each of the withdrawn papers carry notices explaining “the gap,” and that OpenAI has “revised 14 other manuscripts with proof repairs, corrected statements, clearer hypotheses and dependencies, and one correction to an obsolete citation.”
That an error cascaded into three of the OpenAI manuscripts is not surprising, said Alex Townsend, an associate professor of mathematics at Cornell University in Ithaca. He suspects there will be more errors found, he said.
“Given the skepticism of AI in the mathematics community, I think that OpenAI should have announced the manuscripts that were lean verified first,” Townsend told us, referring to Lean, the open-source programming language that checks math proofs. “They could have made a separate announcement for the other ones and ask for help verifying them. This would have been more clear to the community and general public.”
A spokesperson for OpenAI told us the company identified the errors during an audit. As with “other research manuscripts,” the process is iterative, the spokesperson said.
«
They identified the errors during an audit? But didn’t the LLMs work through their proofs, testing each step? Well, no. Not that humans are above making similar mistakes – it’s why proof checking exists – but you might have thought LLMs had reached, well, superhuman status on this.
Meanwhile, you can read multiple (human) mathematicians’ reactions to the paper dump.
unique link to this extract
Satya Nadella says we should assume all AI models are “compromised” • The Verge
Terrence O’Brien:
»
In a lengthy post on X, Microsoft’s CEO laid out his views on the dangers posed by highly advanced AI models and how to confront those risks. Nadella says we can no longer accept a world where AI is treated as a “set of nested black boxes” whose advice and actions we simply accept or reject. He calls for building a more transparent system where models can be contained, observed, and leaves behind “tamper-proof human readable evidence.”
Many of his recommendations align with what we’ve heard from others in the industry: timely incident disclosure, independent audits, verifiable data, and containment. It’s on that last point that he appears to go slightly farther than some others in the field, saying:
»
We must assume a model is compromised and contain it from the start. Think of it like an emergency brake. An authorized person should always be able to pause or shut down a model mid-task. More advanced models will require more advanced containment technologies that we need to standardize on.
«
«
I think this is what we call a “growing chorus” on this point.
unique link to this extract
The generation raised on sugar rations, and what their health reveals 70 years later • The Guardian
Linda Geddes:
»
They were born into a Britain still recovering from the hardships of war. Though the air-raid sirens had fallen silent, many everyday foods remained rationed, including sugar. Adults were restricted to about 3 tablespoons (40g) a day to stretch between cups of tea, baking and desserts; children were limited to less than a tablespoon. Babies and very young children were expected to have little or no added sugar.
Then, in 1953, the brakes came off. Separate restrictions on sweets and chocolate were lifted in February, followed by sugar in September. By the second half of 1954, average sugar consumption had almost doubled.
The lifting of postwar sugar rationing inadvertently created a natural experiment, whose consequences scientists are only beginning to unpick. A flurry of recent studies suggests that spending the first 1,000 days of life under sugar rationing may have left traces on people’s health more than seven decades later.
“What makes the end of sugar rationing particularly interesting is the nature of the exposure,” said Dr Jiazhen Zheng at Hong Kong University of Science and Technology in Guangzhou, China, who worked on several of the recent studies. “In September 1953, sugar availability changed quite abruptly at the population level, largely because of policy rather than individual choice. In a sense, history created an experiment that we could never deliberately run over a human lifetime.”
The first of this recent wave of sugar rationing studies was published in the journal Science two years ago. Dr Tadeja Gračner at the University of Southern California and her colleagues used data from the UK Biobank to compare the health of people born shortly before and after sugar rationing ended in the UK. Some had been exposed to this low sugar environment only during their time in the womb; others remained exposed for months or years, while those conceived once rationing ended were not exposed at all.
Gračner found that the longer people had spent under rationing, the lower their risk of developing type 2 diabetes and high blood pressure in mid- to later life: those exposed throughout their first 1,000 days had a 35% lower risk of type 2 diabetes and a 20% lower risk of high blood pressure than those who had never experienced rationing.
«
Perhaps those annoying parents who deny their children anything sugary for their first few years are actually onto something.
unique link to this extract
Slovenian .si domain registrations outpace .ai after Trump speech • Netcraft
Harry Everett:
»
Three years ago, we reported on a rise in web servers using .ai (Anguilla) domains driven by global interest in artificial intelligence. With Trump’s new “super intelligence” announcement, and .si being a registrable ccTLD (country code TLD) representing the country of Slovenia, we investigated domain registration patterns surrounding the announcement.
Slovenia is a Central European country that sits between Italy, Austria, Hungary, and Croatia. Though unrelated, there is a second connection between Trump and Slovenia. Both Melania and Barron Trump have Slovenian citizenship.
Interest in the .si TLD has risen sharply since the announcement on September 22. At time of writing, the website for Slovenia’s registry, register.si, shows more new domains registered in the “last 24 hours” than in the “last month” (presumably meaning the previous month).
Despite the ccTLD being Slovenian, registration is not limited to Slovenian people or entities. Anyone can register a .si domain. Currently, .si domains are cheaper than .ai, at around $11 for one year compared to around $180 for the minimum 2-year registration period for .ai.
While registrations accelerated after the speech, the initial spike appears to have begun roughly three days earlier. This suggests that some market participants may have been speculating on the possibility of a rename before Trump’s remarks were delivered, though we cannot determine whether registrants were acting on anticipation of the announcement or unrelated factors. Trump shared a poll about this topic on September 19, which appears to line up roughly with the increased registrations of .si domains.
Splitting the registrations by domain registrar, we can see that these registrations are largely at Dynadot, which has an aftermarket that allows users to sell their domains.
«
Would be incredible if there wasn’t some insider trading going on there, of course.
unique link to this extract
AI is throwing a Roadside Picnic • metedata digest
Mete Polat:
»
This whole situation [of AI-generated proofs and “discoveries”] then posits a question: will the rate of scientific discoveries get so rapid that we’ll completely lose the ability to wrap our heads around it, turning AI output into the proverbial alien artefacts? Artefacts that sow chaos in the wake of their arrival and by their mere existence (when discussing the most recent drop of AI scientific papers, the NYU math professor Tristan Buckmaster said that entire research programs were wiped out overnight). In other words, will the rate of discovering answers split off from the rate of our understanding?
I’m certainly not a science historian nor a philosopher, so I’m not well-placed to speculate about the answers or the exact implications. But it’s nonetheless an interesting lens for examining AI’s ongoing impact on science. This may also be one of the few valid anti-accelerationist arguments: if AI vastly outruns our ability to understand the significance of what it’s doing, the collateral damage may outweigh any pure scientific value.
Poignantly enough, the “Roadside Picnic” book explores human irrelevance and the resulting emotional toll behind this alien visit. You could speculate that, at best, aliens didn’t even notice the humans and, at worst, just didn’t care. The big difference is that an alien visit is entirely out of our control, unlike AI development.
I’m fairly skeptical that any coordinated action is even possible in frontier AI development simply due to its game theory dynamics. But the labs have a choice in how they engage with scientific communities to make sure the understanding keeps up with the rate at which answers arrive and that their models act to augment scientists instead of speed-running them to history-making discoveries. Otherwise, we may become the insects watching in horror as everything we know becomes a site of one big alien picnic.
«
The “Roadside Picnic” mention in the title is a reference to Strugatsky’s famous SF novel, where an alien visit leaves a strew of mysterious objects with incomprehensible workings; one scientists suggests the aliens were having the equivalent of a roadside picnic and the objects are the detritus, and we’re the uncomprehending field mice and foxes trying to make sense of spark plugs, penknives and coins. (Thanks Joe S for the link.)
unique link to this extract
What ArtCraft’s vibe-coded apps say about Adobe • Tedium
Ernie Smith:
»
The names aren’t the ones you’re familiar with, but you nonetheless know what each one is. One is a Photoshop stand-in. Another is Premiere Pro. Another is Acrobat. And yes, there’s even an InDesign riff.
Do they work particularly well? On the surface, yes. Opening up these apps in AppImage shows that these apps, programmed in the Rust programming language, are not only fast, but relatively small. Rather than hundreds of megabytes, these apps load in tens of megabytes. They detected my discrete GPU right away. And within a couple of minutes, I was manipulating an image.
A lot of finicky, below the surface stuff did not work so well. Colour pickers and secondary menu boxes were rough, not elegant. It did not detect my fonts in DesignCraft until I downloaded an update published just a few minutes before I downloaded it. And while I could set a colour with a hex code in PhotoCraft, I could not do the same in DesignCraft.
But it was familiar. I could see myself building a zine with this tool, or using it to access old files that I cannot otherwise open without a Creative Cloud subscription.
There are real reasons to cheer on something like this, even if you otherwise hate AI. The spirit of this evokes a Robin Hood-like mindset, in that it exists because of real ethical and economic issues around Adobe that many end users have long been stuck with. In one sense, it is “fighting AI slop with AI slop.” And even if you don’t end up using these tools, the techniques that these projects land on will ultimately inspire open-source projects that actually rely on craft. There are other apps that have emerged in the past couple of years that hope to bring design to Linux; this open-source tool may have opened up a shortcut for these projects.
«
PC market woes continue: shipments fall 20.1% in Q3 2026 • IDC
»
Worldwide PC shipments fell 20.1% year over year in Q3 2026 to 62.7m units, according to preliminary results from IDC Research (IDC), the second consecutive decline and a deeper one than the 3.8% drop in Q2. Shipments also slipped 9.1% sequentially, upending the usual pattern in which Q3 outpaces Q2. Continued supply constraints and the inventory pull-in earlier in the year, as vendors and channels rushed to stock up ahead of memory-driven price hikes, left the quarter with little demand to draw on.
During the quarter the channel worked to move through the large volume of PCs shipped in during the previous quarter. As a result, we didn’t see the typical third-quarter seasonal lift in shipment volumes.
“What we’re seeing is the result of the strong first half pull-in. Vendors and channels loaded up on inventory early in the year to get ahead of price hikes, and that has thrown off the usual seasonality, where Q3 is typically larger than Q2,” said Jitesh Ubrani, research director for consumer devices at IDC. “Channels are now worried about carrying too much inventory into a market where high prices are suppressing demand. That could translate into promotions and some short-term relief for consumers, but we don’t expect pricing anywhere near what it was a year ago. Prices will remain elevated.”
«
No good news in sight. The three biggest vendors – Lenovo, HP and Dell – all fell faster than the market (-23%, -31% and -25% respectively). Apple and ASUS saw their share increase very slightly, even with a fall in shipments. The decline since the start of the year is accelerating; the final quarter could be a bloodbath.
unique link to this extract
AI surveillance startup Flock to cut several hundred jobs amid backlash, sources say • Reuters via The Guardian
»
Flock Safety plans to shed about 18% of its employees, people with direct knowledge of the plans said on Thursday, as the maker of AI-powered surveillance cameras and license-plate readers faces mounting opposition to its products from communities and lawmakers.
The people said the job cuts came after a voluntary buyout program and were expected to affect roughly 270 employees at Flock, a surveillance technology startup that has seen rapid growth in recent years. Employees will leave the company at the end of the month.
Sources declined to be identified as the company has yet to make the plan public. In September, Flock announced a voluntary separation program, asking employees to apply if they wanted to leave the company, as Wired reported. The company has about 1,500 employees.
Flock faces growing scrutiny from regulators, lawmakers and privacy advocates, despite drawing strong investor interest among venture capitalists. A recent Reuters/Ipsos poll showed Americans are divided on whether Flock cameras are a good idea. About 38% of the country supports the use of Flock cameras in their community, compared with 47% who oppose their use.
Across 49 states, Flock’s network of about 120,000 AI-powered cameras, affixed to streets and highways, automatically read license plates and record vehicles as they pass.
The company, based in Atlanta and founded in 2017, has raised more than $950m from venture capital investors, with its latest $275m round in March led by Andreessen Horowitz valuing the company at $7.5bn. The funds have supported a new US manufacturing facility and an upcoming push into drone products, according to the company.
Flock provides cameras and software to more than 4,800 law enforcement agencies and nearly 1,000 businesses. The company says its technology helps investigate and solve crimes, but it has come under increasing scrutiny over alleged privacy violations and data-sharing practices.
«
| • Why do social networks drive us a little mad? • Why does angry content seem to dominate what we see? • How much of a role do algorithms play in affecting what we see and do online? • What can we do about it? • Did Facebook have any inkling of what was coming in Myanmar in 2016? Read Social Warming, my latest book, and find answers – and more. |
Errata, corrigenda and ai no corrida: none notified
