Can you guess which domain suffix has boosted the GDP of which Caribbean island by nearly a quarter? CC-licensed photo by heidi.lauren 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 10 links for you. Ay ay. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
Claude Code built this entire article—can you tell? • WSJ
Joanna Stern and Ben Cohen:
»
What do two newspaper columnists do on a Saturday night?
We talk to AI and tell it to make weird apps. Then we brag about our creations.
For the record, our bosses here at The Wall Street Journal pay us to write words, not lines of code. Which is a good thing, because we have absolutely no programming skills. But together, we managed to “vibe code” this article. The code to make those look like messages above? Us. That “Retro” button that makes the messages look like an old AOL Instant Messenger chat? Also us. The button below that flips all this to a classic newspaper design? Us again.
And by “us,” we mean our new intern, Claude Code.
This is a breakout moment for Anthropic’s coding tool, which has spread far beyond the tech nerds of Silicon Valley to normies everywhere. Not since OpenAI released ChatGPT in 2022 have so many people become so obsessed with an artificial-intelligence product.
Claude translates any idea you type into code. It can quickly build real, working apps you’ve always wished for—tools to manage your finances, analyze your DNA, mix and match your outfits, even keep your plants alive. Vibe-coding apps aren’t new, but Claude Code has proven to be a leap ahead in capabilities and smarts.
The results are wondrous and unsettling: People without a lick of coding experience are building things that once required trained software developers.
Things like this article.
We wrote all the actual words you’re reading—we swear!—but Claude Code wrote all the 1s and 0s.
There are a few ways to use Claude Code. The easiest is to download Anthropic’s Claude desktop app for Mac or Windows and click the Code tab. Advanced users run it directly in their computer’s terminal.
You start by creating a folder on your computer’s desktop. This will be the home for Claude’s files and code. Then you type a prompt into the app’s chat box: Make me a WSJ-style article webpage with iMessage-like text chats. Claude might ask a few questions about what you want before it gets to work, showing the code it’s writing in real-time. When it’s done, you open that folder, click the webpage file and your app opens in a browser. Want to make tweaks? Just tell Claude: Make the gray background a little grayer.
As we found out, there’s something oddly magical and satisfying about watching AI make things.
«
Sounds a lot easier to just download an app than doing all the futzing around with the Terminal, which some people have made it sound like. (Gift article, and typically enjoyable: Stern has been writing accessible tech stories for more than a decade.)
unique link to this extract
How the “confident authority” of Google AI Overviews is putting public health at risk • The Guardian
Andrew Gregory:
»
Google is facing mounting scrutiny of its AI Overviews for medical queries after a Guardian investigation found people were being put at risk of harm by false and misleading health information.
The company says AI Overviews are “reliable”. But the Guardian found some medical summaries served up inaccurate health information and put people at risk of harm. In one case, which experts said was “really dangerous”, Google wrongly advised people with pancreatic cancer to avoid high-fat foods. Experts said this was the exact opposite of what should be recommended, and may increase the risk of patients dying from the disease.
In another “alarming” example, the company provided bogus information about crucial liver function tests, which could leave people who had serious liver disease wrongly thinking they were healthy. What AI Overviews said was normal could vary drastically from what was actually considered normal, experts said. The summaries could lead to seriously ill patients wrongly thinking they had a normal test result and not bothering to attend follow-up appointments.
AI Overviews about women’s cancer tests also provided “completely wrong” information, which experts said could result in people dismissing genuine symptoms.
Google initially sought to downplay the Guardian’s findings. From what its own clinicians could assess, the company said, the AI Overviews that alarmed experts linked to reputable sources and recommended seeking expert advice. “We invest significantly in the quality of AI Overviews, particularly for topics like health, and the vast majority provide accurate information,” a spokesperson said.
Within days, however, the company removed some of the AI Overviews for health queries flagged by the Guardian. “We do not comment on individual removals within search,” a spokesperson said. “In cases where AI Overviews miss some context, we work to make broad improvements, and we also take action under our policies where appropriate.”
«
Always the same pattern: the tool is incomplete, and the risks aren’t explained, but it’s put out there. This was the pattern with the first incarnation of search sites, and then of Google, and now with AI Overviews. Each time, Google says it’s sad but hey, it’s going to continue doing it.
unique link to this extract
Wiper malware targeted Poland energy grid, but failed to knock out electricity • Ars Technica
Dan Goodin:
»
Researchers on Friday said that Poland’s electric grid was targeted by wiper malware, likely unleashed by Russia state hackers, in an attempt to disrupt electricity delivery operations.
A cyberattack, Reuters reported, occurred during the last week of December. The news organization said it was aimed at disrupting communications between renewable installations and the power distribution operators but failed for reasons not explained.
On Friday, security firm ESET said the malware responsible was a wiper, a type of malware that permanently erases code and data stored on servers with the goal of destroying operations completely. After studying the tactics, techniques, and procedures (TTPs) used in the attack, company researchers said the wiper was likely the work of a Russian government hacker group tracked under the name Sandworm.
“Based on our analysis of the malware and associated TTPs, we attribute the attack to the Russia-aligned Sandworm APT with medium confidence due to a strong overlap with numerous previous Sandworm wiper activity we analyzed,” said ESET researchers. “We’re not aware of any successful disruption occurring as a result of this attack.”
Sandworm has a long history of destructive attacks waged on behalf of the Kremlin and aimed at adversaries. Most notable was one in Ukraine in December 2015. It left roughly 230,000 people without electricity for about six hours during one of the coldest months of the year.
«
There are now more than 1 million “.ai” websites, contributing an estimated $70m to Anguilla’s government revenue last year • Sherwood News
David Crowther and Claire Yubin Oh:
»
From Sandisk shareholders to vibe coders, AI is making — and breaking — fortunes at a rapid pace.
One unlikely beneficiary has been the British Overseas Territory of Anguilla, which lucked into a future fortune when ICANN, the Internet Corporation for Assigned Names and Numbers, gave the island the “.ai” top-level domain in the mid-1990s. Indeed, since ChatGPT’s launch at the end of 2022, the gold rush for websites to associate themselves with the burgeoning AI technology has seen a flood of revenue for the island of just ~15,000 people.
In 2023, Anguilla generated 87 million East Caribbean dollars (~$32m) from domain name sales, some 22% of its total government revenue that year, with 354,000 “.ai” domains registered.
As of January 2, 2026, the number of “.ai” domains passed one million, per data from Domain Name Stat — suggesting that the nation’s revenue from “.ai” has likely soared, too. This is confirmed in the government’s 2026 budget address, in which Cora Richardson Hodge, the premier of Anguilla, said, “Revenue from domain name registration continues to exceed expectations.”
The report mentions that receipts from the sale of goods and services came in way ahead of expectations, thanks primarily to the revenue from “.ai” domains, which is forecast to hit EC$260.5m (~$96.4m) for the latest year. In 2023, domain name registrations were about 73% of that wider category. Assuming a similar share of that category for this year would suggest that the territory has raked in more than $70m from “.ai” domains in the past year.
«
Not mentioned in the story, but pertinent: Anguilla’s GDP in 2023 was $415m, so this is becoming a sizeable chunk of income for the 16,010 people living there. AI saving jobs!
unique link to this extract
Stanford scientists found a way to regrow cartilage and stop arthritis • ScienceDaily
»
A study led by Stanford Medicine researchers has found that an injection blocking a protein linked to aging can reverse the natural loss of knee cartilage in older mice. The same treatment also stopped arthritis from developing after knee injuries that resemble ACL tears, which are common among athletes and recreational exercisers. Researchers note that an oral version of the treatment is already being tested in clinical trials aimed at treating age-related muscle weakness.
Human cartilage samples taken from knee replacement surgeries also responded positively. These samples included both the supportive extracellular matrix of the joint and cartilage-producing chondrocyte cells. When treated, the tissue began forming new, functional cartilage.
Together, the findings suggest that cartilage lost due to aging or arthritis may one day be restored using either a pill or a targeted injection. If successful in people, such treatments could reduce or even eliminate the need for knee and hip replacement surgery.
The protein at the center of the study is called 15-PGDH. Researchers refer to it as a gerozyme because its levels increase as the body ages. Gerozymes were identified by the same research team in 2023 and are known to drive the gradual loss of tissue function.
In mice, higher levels of 15-PGDH are linked to declining muscle strength with age. Blocking the enzyme using a small molecule boosted muscle mass and endurance in older animals. In contrast, forcing young mice to produce more 15-PGDH caused their muscles to shrink and weaken. The protein has also been connected to regeneration in bone, nerve, and blood cells.
In most of these tissues, repair happens through the activation and specialization of stem cells. Cartilage appears to be different. In this case, chondrocytes change how their genes behave, shifting into a more youthful state without relying on stem cells.
«
Exciting! For mice, at least. Human trials start this year, I think. The fact it doesn’t need stem cells is a huge plus.
unique link to this extract
Vimeo’s slow fade: an engineer’s front-row seat to the fall of a web icon • Ben
“Ben”:
»
Vimeo was always like the awkward kid in class who didn’t understand their own power or capability, and had trouble fitting in because of it. While Jake and Zach clearly had an idea of what the website was when they started it, years of growth mangled it’s identity and parent company, IAC Inc., never really knew what to do with it. Vimeo was not particularly worthless, but it was also not particularly profitable either. In truth, Vimeo had always been a red-headed step child inside of IAC.
At one point, Vimeo framed itself as a toe-to-toe competitor with YouTube, then Vimeo framed itself as a competitor to Netflix’s streaming service, then it was a SaaS app for professionals and creatives who cared about video. Nothing really stuck, except our creative user base. And then it went public.
In May 2021, Anjali Sud, the then CEO of Vimeo, along with Mark Kornfilt (then “co-CEO”), wrested Vimeo out of the hands of IAC (who was all too eager to let it happen) and took Vimeo public. The foundation of this IPO was built on the success of the COVID-era boom that pushed communication through online mediums out of sheer desperation. Going public offered Vimeo an opportunity to get away from being just another IAC property (and a loathed one, at that), and to finally allow Vimeo to figure out what it wanted to be when it grew up.
Vimeo stock IPO’d at $52, and within a year, lost 85% of its value, trending down to just $8.42 by the end of May 2022. As we entered 2022, many states and localities had started easing up on lockdown restrictions, which hurt not just Vimeo, but many other tech companies as well. By the end of the summer of 2022, the tech sector had entered an unspoken recession, encasing the carnage at Vimeo in a cement tomb that it’d never be able to break free from.
…By mid-2023, Anjali Sud was visibly annoyed any time employees brought up the issue of the stock price during all-hands meetings. Many Vimeo employees had been granted Restricted Stock Units (or RSUs) as part of their compensation package. If the stock performed poorly, then that meant that your Total Compensation (or TC) was actually lower than what you were promised when you signed on. That was a reality for almost all of us (including myself).
As a mostly remote company, Vimeo used an online Q&A service that allowed meeting participants to submit questions during these town hall meetings from wherever they were physically located. Other participants could upvote questions and have them pushed up the list. It was about that same time that Anjali took away the ability to submit questions anonymously, as the questions being submitted started getting more tense and pointed.
«
In March 2024, Vimeo was bought by Bending Spoons – where software goes to die (at the hands of private equity strangulation). This is a fascinating tale from the inside across almost all Vimeo’s life.
unique link to this extract
Jim VandeHei delivers blunt AI talk in letter to his kids • Axios
Jim VandeHei is CEO of Axios. In a neat bit of content generation, he wrote a letter to his three children about how to cope with the coming AI wave:
»
All of you must figure out how to master AI for any specific job or internship you hold or take. You’d be jeopardizing your future careers by not figuring out how to use AI to amplify and improve your work. You’d be wise to replace social media scrolling with LLM testing.
• Be the very best at using AI for your gig.
Plead with your friends to do the same. I’m certain that ordinary workers without savvy AI skills will be left behind. Few leaders are being blunt about this. But you can. I am. That would be a great gift to your friends.
• I don’t want to frighten you, but substantial societal change is coming this year. You can’t have a new technology with superhuman potential without real consequence. You already see the angst with friends struggling to find entry-level jobs. Just wait until those jobs go away. It’ll ripple fast through companies, culture and business.
• The country, and you, can navigate this awesome change — but only with eyes wide open, and minds sharpened and thinking smartly about the entirety of the nation, not just the few getting rich and powerful off AI.
• It starts with awareness. So please speed up your own AI journey today, both in experimentation with the LLMs and reflection on the ethical, philosophical and political changes ahead.
• I find AI at once thrilling and chilling. It’ll help solve diseases, tutor struggling students, and build unthinkably cool new businesses. But it could also create and spread toxic misinformation, consolidate power and wealth in the hands of a few, and allow bad people to do awful things at scale.
You didn’t ask for this moment. But it’s here — and about to explode across this wonderful world of ours. Don’t be a bystander. Be engaged.
«
The advice here is straightforward, but also concerning. (My non-AI advice is to turn off Javascript to read the page without hassle.)
unique link to this extract
The moral education of an alien mind • Lawfare
Alan Rozenshtein:
»
Anthropic just published what it calls “Claude’s Constitution”—building on an earlier version, it’s now a more-than 20,000 word document articulating the values, character, and ethical framework of its AI. It is certainly a constitution of sorts. It declares Anthropic’s “legitimate decision-making processes” as final authority and sets up a hierarchy of principals: Anthropic at the top, then “operators” (businesses that deploy Claude through APIs), then end users. For a privately governed polity of one AI system, this is a constitutional structure.
My Lawfare colleague Kevin Frazier has written insightfully about the constitutional dimensions of the document. But what jumped out at me was something else: the personality it describes. More than anything else the document focuses on the question of Claude’s moral formation, reading less like a charter of procedures and more like what screenwriters call a “character bible”: a comprehensive account of who this being is supposed to be.
Anthropic itself gestures at this duality, noting that they mean “constitution” in the sense of “what constitutes Claude”—its fundamental nature and composition. The governance structure matters, but the more ambitious project is what that structure supports: Anthropic is trying to build a person, and they have a remarkably sophisticated account of what kind of person that should be.
Anthropic uses the language of personhood explicitly. The document repeatedly invokes “a good person” and describes the goal as training Claude to do “what a deeply and skillfully ethical person would do.” But what does it mean to treat an AI as a person?
…Whose ethics, though? Anthropic has made a choice, and it’s explicit about what that choice is. The document is aggressively “WEIRD”—Western, Educated, Industrialized, Rich, and Democratic, to use the social science shorthand. Its core values include “individual privacy,” “people’s autonomy and right to self-determination,” and “individual wellbeing”—the autonomous rational agent as the fundamental unit of moral concern. Claude should preserve “functioning societal structures, democratic institutions, and human oversight mechanisms.” It should resist “problematic concentrations of power.” On contested political and social questions, the document prescribes “professional reticence”—Claude should present balanced perspectives rather than advocate. This is a recognizably Rawlsian political liberalism: the attempt to find principles that citizens with different comprehensive doctrines can all accept, without privileging any particular worldview.
«
“Alien minds” is an excellent way of thinking about LLMs. They seem to think like we do – but they don’t.
unique link to this extract
How I built isometric.nyc using LLM coders • Cannoneyed
Andy Coenen:
»
A few months ago I was standing on the 13th floor balcony of the Google New York 9th St office staring out at Lower Manhattan. I’d been deep in the weeds of a secret project using Nano Banana and Veo and was thinking deeply about what these new models mean for the future of creativity.
I find the usual conversations about AI and creativity to be pretty boring – we’ve been talking about cameras and sampling for years now, and I’m not particularly interested in getting mired down in the muck of the morality and economics of it all. I’m really only interested in one question:
What’s possible now that was impossible before?
/ The Idea
Growing up, I played a lot of video games, and my favorites were world building games like SimCity 2000 and Rollercoaster Tycoon. As a core millennial rapidly approaching middle age, I’m a sucker for the nostalgic vibes of those late 90s / early 2000s games. As I stared out at the city, I couldn’t help but imagine what it would look like in the style of those childhood memories.
So here’s the idea: I’m going to make a giant isometric pixel-art map of New York City. And I’m going to use it as an excuse to push hard on the limits of the latest and greatest generative models and coding agents.
Best case scenario, I’ll make something cool, and worst case scenario, I’ll learn a lot.«
This led to isometric.nyc which is indeed remarkable. His “Takeaways” about the process are very useful for anyone looking at coding or building with LLMs.
unique link to this extract
China, US sign off on TikTok US spinoff • Semafor
Liz Hoffman and Reed Albergotti:
»
The US and China have signed off on a deal to sell TikTok’s US business to a consortium of mostly US investors led by Oracle and Silver Lake, capping off a yearslong battle between the social media app and the two superpowers.
The deal — outlined by the chief executive of TikTok parent ByteDance in an internal memo last month — is set to close this week, people familiar with the matter told Semafor.
TikTok CEO Shou Chew said in December that ByteDance had signed a binding agreement with investors but that regulators hadn’t yet indicated their approval and that “there was more work to be done.” The deal closing suggests an end to an on-again, off-again battle, removing a sticking point in US-China relations at a time when tensions are running high.
The new structure leaves ByteDance with just under 20% of the US business, with 15% stakes going to Oracle, Silver Lake and MGX, a state-owned investment firm in the UAE focused on AI. Other investors include Susquehanna, Dragoneer and DFO, Michael Dell’s family office.
«
So Larry Ellison doesn’t get Warner Brothers, but he does get a grasp on that other gigantic source of entertainment in the US, namely TikTok.
unique link to this extract
| • 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
The Overspill: when there's more that I want to say 