The killing of more than ten people and wounding of hundreds in Lebanon by exploding pagers has raised many questions. CC-licensed photo by Hades2k 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.
There’s another post coming this week at the Social Warming Substack on Friday at 0845 UK time. Free signup.
A selection of 10 links for you. Ride on time. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. Observations and links welcome.
New Instagram changes for teens include private accounts and other restrictions • WSJ
Julie Jargon:
»
Instagram is placing teens in a protective bubble.
Starting this week, it will begin automatically making youth accounts private, with the most restrictive settings. And younger teens won’t be able to get around it by changing settings or creating adult accounts with fake birth dates.
Account restrictions for teens include direct messaging only with people they follow or are already connected to, a reduction in adult-oriented content, automatic muting during nighttime hours and more.
Building on changes to teen accounts it announced earlier this year—and following years of criticism about child safety—the Meta Platforms-owned social network said it would shift 100 million teenagers in the US and around the world into the guardrailed accounts. The move applies to all accounts with an under-18 birth date, though teens 16 and older will be able to change their settings without parental approval.
Any new teen accounts will be similarly restricted starting Tuesday. Parents will no longer have to manually enter those settings using Instagram’s parental supervision tool.
Teens are unlikely to be happy with the changes. Instagram is expecting to lose “some meaningful amount of teen growth and teen engagement,” Instagram head Adam Mosseri said in an interview. “I have to believe earning some trust from parents and giving parents peace of mind will help business in the long run, but it will certainly hurt in the short term.”
Instagram plans to go even further, starting next year: Using artificial intelligence, it said, it will identify children who are lying about their age—then automatically place them into the restricted teen accounts.
«
There’s a writeup at The Guardian too, quoting Ian Russell, whose daughter Molly killed herself in 2017 after viewing suicide-related content shown to her via Instagram’s algorithm. So it’s only taken them seven years to get to this.
unique link to this extract
Hezbollah pagers explode in apparent attack across Lebanon • WSJ
Adam Chamseddine, Summer Said and Stephen Kalin:
»
Pagers carried by thousands of Hezbollah operatives exploded at about the same time Tuesday afternoon in what appeared to be an unprecedented attack that authorities said injured more than 2,700 and killed eight [update: at least 11 – Overspill Ed] across Lebanon.
Many of the affected pagers were from a new shipment that the group received in recent days, people familiar with the matter said. A Hezbollah official said many fighters had such devices, speculating that malware might have caused the devices to explode. The official said some people felt the pagers heat up and disposed of them before they burst.
Hezbollah said a number of pagers carried by its members exploded simultaneously at 3:30 p.m. local time. It couldn’t immediately be determined what caused the blasts, which were spread out across the country in several areas where Hezbollah has a heavy presence.
Hezbollah and the Lebanese government blamed Israel for the attack. Both said civilians were killed, and Hezbollah threatened to retaliate. The Israeli military declined to comment.
Iranian state television said the country’s ambassador to Lebanon, Mojtaba Amani, was injured by his pager but was conscious and not in danger. Iran is the main supporter of Hezbollah, a US-designated terrorist group that has grown into one of the world’s best-armed nonstate militias.
«
No details at the time of compilation, but here’s the obvious method: infiltrate the supply chain and pack C-4/Semtex into the pagers. The details of how the battery charge is diverted into the explosive is for the bombmakers. That’s the “new shipment”. Hezbollah had stopped using mobile phones because they considered them unsafe for various reasons – including the fact that Israel, apparently, killed a Hamas operative using a boobytrapped one in 1996.
Hezbollah has been indiscriminate with its attacks: in July, one of its rockets hit a playground and soccer field in northern Israel, killing nine children and injuring 20 more. Making pagers explode is indiscriminate too: among those killed on Tuesday was the 10-year-old daughter of a Hezbollah operative who was standing by her father.
The US said it had nothing to do with it. Unconnected: the American University of Beirut Medical Center replaced the pagers of all doctors and staff at the end of August following an upgrade in April. (Thanks Jimbo.)
unique link to this extract
Streaming struggles with churn as even buzzy shows can’t keep subscribers • Deadline
Katie Campione:
»
Even as streaming continues to grow, subscribers are still cycling through the premium services to watch their favorite shows, rather than committing to any or all of them.
According to Samba TV‘s latest State of Viewership Report, U.S. households watched a record level of streaming content in the first half of the year, up 40% from the same time frame in 2023. However, the same report also states that about 44% of subscribers are only willing to watch one or two platforms per six months.
Subscription Video On Demand (SVOD) platforms are still consistently adding subscribers each quarter, but cancellations each quarter have also grown, and the services only added about 4.8m net subscribers in Q1 2024 — down more than 3m from the previous quarter.
Samba’s data shows that viewers are generally invested in a specific piece of content, choosing to leave the platform after they’ve finished watching. The only exception to this is Netflix, which is unsurprising given that the streamer has proven highly effective at using its algorithm to draw users to new content and keep them engaged on the platform. It’s a tool that many of the newer platforms have yet to perfect.
Netflix had the lowest level of churn and also represented 60% of the Top 50 shows in the first half of the year with series like Fool Me Once, Griselda, American Nightmare, Bridgerton and The Gentlemen in the Top 10.
Elsewhere, subscription cyclers come back for buzzy new shows, Samba says, including Peacock’s Ted, AppleTV’s Masters of the Air, and FX’s Shōgun, which aired in Hulu and Disney+. New seasons of House of the Dragon (the second most-popular premiere of the half, behind Fool Me Once), True Detective, and Reacher were also a big draw.
But still, once those shows are over, subscribers often leave the service until they’re enticed back again for another anticipated release.
«
Fabulous irony: I’ve been glued to the BBC’s Sherwood series, which is completely bingeable on iPlayer. Meanwhile, Apple TV+ parcels out one episode per week of Slow Horses.
unique link to this extract
Ban warnings fly as users dare to probe the “thoughts” of OpenAI’s latest model • Ars Technica
Benj Edwards:
»
Unlike previous AI models from OpenAI, such as GPT-4o, the company trained o1 specifically to work through a step-by-step problem-solving process before generating an answer. When users ask an “o1” model a question in ChatGPT, users have the option of seeing this chain-of-thought process written out in the ChatGPT interface. However, by design, OpenAI hides the raw chain of thought from users, instead presenting a filtered interpretation created by a second AI model.
Nothing is more enticing to enthusiasts than information obscured, so the race has been on among hackers and red-teamers to try to uncover o1’s raw chain of thought using jailbreaking or prompt injection techniques that attempt to trick the model into spilling its secrets. There have been early reports of some successes, but nothing has yet been strongly confirmed.
Along the way, OpenAI is watching through the ChatGPT interface, and the company is reportedly coming down hard on any attempts to probe o1’s reasoning, even among the merely curious.
One X user reported (confirmed by others, including Scale AI prompt engineer Riley Goodside) that they received a warning email if they used the term “reasoning trace” in conversation with o1. Others say the warning is triggered simply by asking ChatGPT about the model’s “reasoning” at all.
«
Simon Willison (a smart developer, for those who don’t know) has a good writeup of what he reckons is going on with the “reasoning”. Ben Thompson (in his paywalled Stratechery) got it to solve a 7×7 crossword, and John Gruber got it to solve a logic puzzle that I found too exhausting to fight through.
It’s getting closer to something that “reasons”.
unique link to this extract
TikTok fights ban in court hearing, facing skeptical judges • The Washington Post
Drew Harwell and Eva Dou:
»
The fate of the wildly popular TikTok app hangs in the balance after the company tried to persuade a Washington appeals court Monday to halt a fast-approaching ban on the platform’s use across America.
A deep legal discussion over the potential ban’s constitutionality Monday morning offered no clear answers, leaving the company’s fate uncertain even as it nears a Jan. 19 deadline to divest from Chinese ownership, with a ban coming into place if it hasn’t done so by that date.
The political backdrop is striking: Vice President Kamala Harris and former president Donald Trump are both seizing on the viral short-video platform to vie for young voters, even though both the Trump and Biden administrations backed banning it.
In Monday morning’s hearing in federal D.C. Court of Appeals, the panel of three judges — Chief Judge Sri Srinivasan and Judges Neomi Rao and Douglas H. Ginsburg — grilled TikTok attorney Andrew Pincus on why he thought the company’s right to free speech outweighed national security concerns over its ownership, citing wartime precedents of the United States curbing the broadcast of foreign propaganda into America.
The judges pressed Pincus on a hypothetical scenario involving a war between the United States and a foreign country, asking whether Congress would be within its right to bar “the enemy’s ownership of a major media source” — in line with Congress’s designation of TikTok and other China-based apps as controlled by a “foreign adversary.”
They also cited the government’s concern over how ByteDance developers in China could curate TikTok’s recommendation algorithm for American users.
Pincus argued that ByteDance, TikTok’s parent company in China, is a private company, not a state-owned enterprise, and that the United States isn’t at war with China. The judges didn’t seem wholly convinced.
«
An information war is still a war, after all.
unique link to this extract
Mozilla exits the fediverse and will shutter its Mastodon server in December • TechCrunch
Sarah Perez:
»
Mozilla.social was a small instance, having only 270 active users as of the time of Tuesday’s announcement. By comparison, the most popular Mastodon instance, Mastodon.social, has over 247,500 monthly active users.
Mozilla had telegraphed its plans to scale back on its fediverse investments earlier this year after the CEO stepped down. At the time, Mozilla board member Laura Chambers took over the job as the interim CEO of Mozilla Corporation through the end of 2024. Shortly after the change in leadership, Mozilla said it would refocus its product strategy around Firefox and AI and significantly scale back or even shutter other efforts. Among those products affected by the pullback were its VPN, Relay, and Online Footprint Scrubber, in addition to its Mastodon instance, the company said at the time. Meanwhile, its virtual world Hubs was shut down.
The redirection of Mozilla’s efforts came after its flagship product, the Firefox web browser, spent years losing market share. That left room for other competitors, like the startup Arc, to take hold in the alternative browser market.
Months prior to this change in strategy, Mozilla had been touting the fediverse’s potential, but under Chambers, the company said that a more “modest approach” to the fediverse would have allowed it to participate with “greater agility.”
«
Mozilla is really hunkering down. The Mastodon server can’t have been costing anything, but it seems like anything surplus is too much. Presently Mozilla has nine products (but that includes Relay and the VPN). Odd to abandon the VPN: companies like Nord are making good money from them (even though their adverts wildly exaggerate the need for them).
unique link to this extract
CMA objects to Google’s ad tech practices in bid to help UK advertisers and publishers • GOV.UK
»
An investigation by the Competition and Markets Authority (CMA) has provisionally found that Google is using anti-competitive practices in open-display ad tech, which it believes could be harming thousands of UK publishers and advertisers.
As set out in a statement of objections issued to Google today (Friday 6 September), the CMA has provisionally found that, when placing digital ads on websites, the vast majority of publishers and advertisers use Google’s ad tech services in order to bid for and sell advertising space.
The CMA is concerned that Google is actively using its dominance in this sector to preference its own services. Google disadvantages competitors and prevents them competing on a level playing field to provide publishers and advertisers with a better, more competitive service that supports growth in their business.
In its 2019 market study into digital advertising, the CMA found that advertisers were spending around £1.8bn annually on open-display ads, marketing goods and services via apps and websites to UK consumers.
…The CMA has provisionally found that, since at least 2015, Google has abused its dominant positions through the operation of both its buying tools and publisher ad server in order to strengthen AdX’s market position and to protect AdX from competition from other exchanges. Moreover, due to the highly integrated nature of Google’s ad tech business, the CMA has provisionally found that Google’s conduct has also prevented rival publisher ad servers from being able to compete effectively with DFP, harming competition in this market.
«
This may look abstruse and acronym/abbreviation-laden, but it’s the same topic as the DOJ trial in the US: Google abuses its position at both ends of the advertising market (selling space to advertisers, finding buyers for publishers’ space, aka “inventory”) to charge excess fees.
unique link to this extract
Is My Blue Your Blue?
»
Colour perception is tricky to measure–vision scientists use specialized calibrated equipment to measure colour perception. Graphic designers use physical colour cards, such as those made by Pantone, so that they can communicate colours unambiguously. Here we use your monitor or phone to test how you categorize colours, which is far from perfect, since your calibration may differ from mine.
The validity of the inference is limited by the calibration of your monitor, ambient lighting, and filters such as night mode. Despite these limitations, the results should have good test-retest reliability on the same device, in the same ambient light, which you can verify by taking the test multiple times. If you want to compare your results with friends, use the same device in the same ambient light.
Getting outlier results doesn’t mean there’s anything wrong with your vision. It might mean you have an idiosyncratic way of naming colours, or that your monitor and lighting is unusual.
«
Fun game! When you see a colour on the screen, press the “This is blue” or “This is green” button at the bottom. Then keep on going. Apparently my green pushes over into most people’s blue. Look, you just need to move your idea of “blue”.
unique link to this extract
South Korea: the deepfake crisis engulfing hundreds of schools • BBC News
Jean Mackenzie and Leehyun Choi:
»
South Korean journalist Ko Narin published what would turn into the biggest scoop of her career. It had recently emerged that police were investigating deepfake porn rings at two of the country’s major universities, and Ms Ko was convinced there must be more.
She started searching social media and uncovered dozens of chat groups on the messaging app Telegram where users were sharing photos of women they knew and using AI software to convert them into fake pornographic images within seconds.
“Every minute people were uploading photos of girls they knew and asking them to be turned into deepfakes,” Ms Ko told us.
Ms Ko discovered these groups were not just targeting university students. There were rooms dedicated to specific high schools and even middle schools. If a lot of content was created using images of a particular student, she might even be given her own room. Broadly labelled “humiliation rooms” or “friend of friend rooms”, they often come with strict entry terms.
Ms Ko’s report in the Hankyoreh newspaper has shocked South Korea. On Monday, police announced they were considering opening an investigation into Telegram, following the lead of authorities in France, who recently charged Telegram’s Russian founder for crimes relating to the app. The government has vowed to bring in stricter punishments for those involved, and the president has called for young men to be better educated.
Telegram said it “actively combats harmful content on its platform, including illegal pornography,” in a statement provided to the BBC.
The BBC has viewed the descriptions of a number of these chatrooms. One calls for members to post more than four photos of someone along with their name, age and the area they live in.
“I was shocked at how systematic and organised the process was,” said Ms Ko. “The most horrific thing I discovered was a group for underage pupils at one school that had more than 2,000 members.”In the days after Ms Ko’s article was published, women’s rights activists started to scour Telegram too, and follow leads. By the end of that week, more than 500 schools and universities had been identified as targets. The actual number impacted is still to be established, but many are believed to be aged under 16, which is South Korea’s age of consent. A large proportion of the suspected perpetrators are teenagers themselves.
«
Agenda Hero: we make calendars magically simple
»
We make calendars magically simple. An AI companion for calendars you already use and any new ones you create
«
Neat: I gave it an image of a spreadsheet of a set of tennis matches (without saying they were tennis matches anywhere; they were just described as “Mixed”) and it correctly created the calendar events and added a tennis ball in front of each event.
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 