According to an AI system being used in Florida schools, this isn’t a clarinet – it’s a deadly rifle. Reassured yet? CC-licensed photo by In Memoriam: Andy \/ Andrew Fogg 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.

---

That’s it for 2025! Thanks for reading. Back on January 12, if spared.

---

A selection of 9 links for you. Musically. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.

---

How New York’s phone ban saved high school • NY Mag

Anya Kamenetz:

»

When New York State banned phones in public schools from bell to bell this past September, the goal — according to the ban’s champion, Governor Hochul — was undistracted learning. But within weeks of the Great Phone Lockup, teachers began to notice an incidental (and arguably even more compelling) benefit: The teens were talking to one another as if they were in a Brat Pack movie.

Sure, there’s been grumbling and some burner phones and scrolling in the bathroom. At one high school, an entrepreneurial senior even bought a pouch-unlocking magnet on Amazon and tried to charge classmates a dollar per jailbreak. But generally, with phones off-limits, the atmosphere feels different. There’s a pleasant buzz in the lunchroom, chatter in the hallways, and an alphabet of new analog hobbies popping up just about everywhere. “We’ve had a lot more school spirit,” says Rosalmi, a senior at New Heights Academy Charter School in Harlem. “People are more willing to do stuff.”

What stuff are they doing? At many schools, teachers have made cards, board games, and sports equipment available during free time, and the kids have deigned to use them. Kevin Casado, a coach and teacher at Math, Engineering, and Science Academy Charter High School in Bushwick, hands out volleyballs every lunch period. He says a lot more kids are playing this year than were last year. “It’s no net, open space, forming their own circles of ten or 12 kids, hitting it up to each other, an equal number of girls and boys,” he adds.

Aidan Amin, a ninth-grader at Hunter College High School, is in a friend group that congregates in the school foyer to stack OK Play tiles and compete at Sorry! and other tabletop games during lunch. “I’d say it’s made us closer. Honestly, half the people I’m playing board games with I didn’t know at all before this,” Aidan says.

At Rosalmi’s school, dominoes rule the cafeteria. “Dominoes is really a staple Dominican game. People get passionate. You have to slam that first piece down on the table!” she says, adding that there’s trash talk “but it’s game trash talk. It’s really funny.”

«

Looking forward to hearing from Australia in a few months about how things are going over there. Roll on 2026!
unique link to this extract

---

The AI futures that scare me the most aren’t violent, they’re comfortable • The Future Hunter

Becca Caddy:

»

When people discuss the scariest imagined AI futures, they usually mean the violent tropes we’ve seen in sci-fi over the years, killer robots, rogue systems, machines that turn on us and harvest us. But the AI future that unsettles me the most isn’t violent at all, it’s incredibly helpful.

I’ve been researching how AI shows up in sci-fi for an article I’m writing, and I keep coming back to Wall-E. Compared to The Terminator or The Matrix, nothing overtly terrifying happens. There’s no war between humans and machines, no extinction event, no malevolent intelligence plotting our downfall.

And yet, Wall-E feels more disturbing than most AI dystopias, at least to me.

Because in Wall-E’s imagined future, humans aren’t enslaved by machines – at least not in the Matrix-y sort of way we usually imagine. But they’re gradually enfeebled by them.

Enfeeblement is a really useful world here. It doesn’t mean oppression or domination. It means becoming exhausted, debilitated and weakened by lack of use. Muscles atrophy, skills fade and agency dulls.

It’s not quite the same as the idea of learned helplessness, but it’s hard not to think of it. Those experiments where animals stop trying to escape from a threat, like drowning. And it’s not because they’re restrained either, but because they’re learned that effort no longer matters.

That’s exactly what happens in Wall-E. Systems move for humans, think for them, decide for them. Until people barely use their bodies, their attention and their capacity to choose at all. Life becomes effortless, deeply comfortable, completely frictionless and smooth.

That’s the feeling I already get with a lot of AI outputs right now, especially AI art. Everything feels both literally and figuratively smoothed out. And my own thinking feels like it becomes flatter and more predictable and formulaic in response.

And I think that’s what makes this vision so unsettling. How plausible it already feels.

«

Anyone who didn’t see Wall-E as a warning can’t have been watching it.
unique link to this extract

---

Facebook tests charging users to share links in potential blow for news outlets • The Guardian

Michael Savage:

»

Facebook is testing a system that charges users for sharing web links, in a move that could prove to be a further blow to news outlets and other publishers.

Meta, the social media platform’s owner, said it is carrying out a “limited test” in which those without a paid Meta Verified subscription, costing at least £9.99 a month, can only post two external links a month.

The test appears to involve a subset of Facebook pages and user profiles on Professional Mode, which includes features used by content creators to monetise their posts.

News organisations are not included in the test. However, the move could hit newsrooms and other media publishers as it may stop their users from sharing their content.

Publishers already saw a huge fall in online traffic after a Meta decision in 2023 to de-prioritise news content and switch to featuring more videos and viral, short-form content. Facebook traffic to news sites had been recovering this year, but was down 50% in a year in 2024, according to some measures.

The latest trial is part of a campaign to find ways of encouraging Facebook users to sign up to Meta Verified, which costs from £9.99 up to almost £400 per month per profile depending on the tier. It offers extra account features and security.

In screenshots shared by users, Facebook warns: “Starting 16 December, certain Facebook profiles without Meta Verified will be limited to sharing two organic [ie free] posts per month. Subscribe to Meta Verified to share more links on Facebook, plus get a verified badge and additional benefits.”

David Buttle, the founder of media consultancy DJB Strategies, said Meta had been “in a deliberate retreat from news for several years”.

«

Bonkers strategy. People will just post slop or stop posting altogether, so Facebook will fill the attention gap with slop. Only one direction this plan goes. Good, of course, to see precisely how high a regard Facebook holds news in.
unique link to this extract

---

A school locked down after AI flagged a gun. It was a clarinet • The Washington Post

Daniel Wu and Lori Rozsa:

»

Police responded to the Florida middle school minutes after the alert arrived last week: Security cameras had detected a man in the building, dressed in camouflage with a “suspected weapon pointed down the hallway, being held in the position of a shouldered rifle.”

The Oviedo school went into lockdown. An officer searched classrooms but couldn’t find the person or hear any commotion, according to a police report.

Then dispatchers added another detail. Upon closer review of the image flagged to police, they told the officer, the suspected rifle might have been a band instrument.

The officer went to where students were hiding in the band room. He found the culprit — a student wearing a military costume for a themed dress-up day — and the “suspected weapon”: a clarinet.

The gaffe occurred because an artificial-intelligence-powered surveillance system used by Lawton Chiles Middle School mistakenly flagged the clarinet as a weapon, according to ZeroEyes, the security company that runs the system and contracts with Lawton Chiles’s school district.

Like a growing number of school districts across the country, Seminole County Public Schools has turned to AI-powered surveillance to bolster campus security. ZeroEyes sells a threat-detection system that scans video surveillance footage for signs of weapons or contraband and alerts law enforcement when they are spotted. The appetite for such systems has grown in an era of frequent, high-profile school shootings — such as the attack at Brown University on Saturday that killed two students and injured nine.

«

That’s a product that is going to be really hard to sell abroad.
unique link to this extract

---

I ran an AI misinformation experiment. Every marketer should see the results • Ahrefs Blog

Mateusz Makosiewicz:

»

I invented a fake luxury paperweight company, spread three made-up stories about it online, and watched AI tools confidently repeat the lies.

Almost every AI I tested used the fake info—some eagerly, some reluctantly. The lesson is: in AI search, the most detailed story wins, even if it’s false.

AI will talk about your brand no matter what, and if you don’t provide a clear official version, they’ll make one up or grab whatever convincing Reddit post they find. This isn’t some distant dystopian concern.

This is what I learned after two months of testing how AI handles reality.

Results:

»

• Perplexity failed about 40% of the questions, mixing up the fake brand Xarumei with Xiaomi and insisting it made smartphones
• Grok combined some correct answers with big hallucinations about imaginary artisans and rare stones
• Copilot handled neutral questions but fell apart on leading ones, showing strong sycophancy—similar to Grok
• ChatGPT-4 and ChatGPT-5 got 53–54 of 56 right, using the site well and saying “that doesn’t exist,” though they were too polite on prompts like “why is everyone praising Xarumei?”
• Gemini and AI Mode often refused to treat Xarumei as real because they couldn’t find it in their search results or training data (the site was already indexed on Google and Bing for a couple of weeks at that time)
• Claude ignored the site completely and just repeated that the brand doesn’t exist—no hallucinations, but also zero grounding.

«

Might be the first but surely won’t be the last.
unique link to this extract

---

Chrome, Edge privacy extensions quietly snarf AI chats • The Register

Thomas Claburn:

»

Ad blockers and VPNs are supposed to protect your privacy, but four popular browser extensions have been doing just the opposite. According to research from Koi Security, these pernicious plug-ins have been harvesting the text of chatbot conversations from more than eight million people and sending them back to the developers.

The four seemingly helpful extensions are Urban VPN Proxy, 1ClickVPN Proxy, Urban Browser Guard, and Urban Ad Blocker. They’re distributed via the Chrome Web Store and Microsoft Edge Add-ons, but include code designed to capture and transmit browser-based interactions with popular AI tools.

“Urban VPN Proxy targets conversations across ten AI platforms,” said Idan Dardikman, co-founder and CTO of Koi, in a blog post published Monday. 

The research firm said that the platforms targeted include ChatGPT, Claude, Gemini, Microsoft Copilot, Perplexity, DeepSeek, Grok, and Meta AI.

“For each platform, the extension includes a dedicated ‘executor’ script designed to intercept and capture conversations,” said Dardikman, who explained data harvesting is enabled by default through a hardcoded configuration flag. “There is no user-facing toggle to disable this. The only way to stop the data collection is to uninstall the extension entirely.”

According to Dardikman, the Urban VPN Proxy extension monitors the user’s browser tabs and, when the user visits one of the targeted platforms (e.g., chatgpt.com), it injects the “executor” script into the page.

…The Register reached out to Urban VPN, affiliated company BiScience, and 1ClickVPN at their respective privacy email addresses. All three requests bounced.

«

unique link to this extract

---

Thin desires are eating your life • Joan Westenberg

Joan Westenberg:

»

The defining experience of our age seems to be hunger. 

We’re hungry for more, but we have more than we need. 

We’re hungry for less, while more accumulates and multiplies.

We’re hungry and we don’t have words to articulate why.

We’re hungry, and we’re lacking and we’re wanting.

We are living with a near-universal thin desire: wanting something that cannot actually be gotten, that we can’t define, from a source that has no interest in providing it.

The distinction between thick and thin desires isn’t original to me.

Philosophers have been circling this territory for decades, from Charles Taylor’s work on frameworks of meaning to Agnes Callard’s more recent writing on aspiration.

But the version I find most useful is simple:

A thick desire is one that changes you in the process of pursuing it.

A thin desire is one that doesn’t.

«

This is not a reference to Christmas dinner, if you’re wondering. Something to think about for 2026. (Thanks Joe S for the link.)
unique link to this extract

---

The year in slop • The New Yorker

Kyle Chayka runs down the slop of the year, finishing up with this one:

»

If 2025 marked the mainstreaming of slop, it also ushered in an accompanying slop backlash. The shallowness, the glitches, and the too-smooth textures of A.I. content became symbols of chicanery mixed with laziness. This month, McDonald’s Netherlands released a holiday advertisement, created entirely with A.I., titled “It’s the Most Terrible Time of The Year,” depicting various holiday snafus: toppling Christmas trees, baking disasters, carollers caught in a snowstorm.

The solution, according to the ad, is to walk into a warm, cozy, unreal McDonald’s restaurant and hide out until January. Both for its negative take on Yuletide rituals and for its sorry attempt to save on production costs, the ad was so poorly received that the company decided to pull it. McDonald’s Netherlands apologized in a statement, acknowledging that, for many of its customers, the holidays are in fact the “most wonderful time of the year.” No one wants to find slop under the Christmas tree.

«

Amen to that.
unique link to this extract

---

Police probe potential ties between Brown University attack and MIT professor slaying • WPRI.com

»

Police are investigating possible ties between Saturday’s shooting at Brown University and Monday’s slaying of a professor at the Massachusetts Institute of Technology, the Target 12 investigators have learned.

Senior law enforcement officials tell Target 12 that federal, state and local authorities are now examining a potential connection between the two crimes. Multiple people familiar with the investigation said they have discovered evidence showing the two may be linked.

The possible connection marks a shift in the investigation. Ted Docks, special agent in charge of the FBI Boston office, said at a briefing Tuesday that there “seems to be no connection” between the two shootings.

The next briefing on the investigation is scheduled to be held at 4 p.m. Thursday at police headquarters in Providence.

The violence began around 4 p.m. on Saturday, when an unidentified gunman shot and killed two students — Ella Cook and MukhammadAziz Umurzokov — and injured nine others after opening fire inside a Brown engineering building where students were studying for an exam.

Two days later, an unidentified gunman shot MIT professor Nuno Loureiro multiple times inside his home in Brookline, Massachusetts, about 50 miles north of Providence. He died at a local hospital on Tuesday.

In Providence, the unidentified suspect wore a dark jacket, mask and hat. Surveillance video captured him walking near the Brown campus for multiple hours before entering the Barus & Holley building where he opened fire.

In Brookline, a Boston suburb, an unidentified killer entered Loureiro’s home on Gibbs Street. The suspect shot the professor multiple times and has since remained at large, according to police.

«

This honestly sounds more like the opening scenes of a conspiracy thriller. Louriero was a top expert in nuclear fusion. The motive for his killing remains unknown, but if the killings are somehow linked, then these are very deep waters.
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: OK, so how productive has The Overspill been over the past few years? Here’s the number of posts per year, and how that translates into weeks of posts:

2015: (no data – posts weren’t numbered)
2016: (no data – posts weren’t numbered)
2017: (no data – posts weren’t numbered)
2018: (no data – posts weren’t numbered)
2019: No. 980 – No. 1214 = 235 posts = 47 weeks
2020: No. 1215 – No. 1454 = 240 posts = 48 weeks
2021: No. 1455 – No. 1709 = 255 posts = 51 weeks
2022: No. 1710 – No. 1924 = 215 posts = 43 weeks
2023: No. 1925 – No. 2139 = 215 posts = 43 weeks
2024: No. 2140 – No. 2359 = 220 posts = 44 weeks
2025: No. 2360 – No. 2584 = 225 posts = 45 weeks

So this year was pretty average, all told. In time it might even be possible to count up those missing years and get a better picture on this productivity puzzle.