Seeing faces in clouds, as humans do, resembles what LLMs do with text – mistaking unstructured content for content. CC-licensed photo by Jaxs Powell 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 9 links for you. Use them wisely. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.

---

WhatsApp backs Apple in its legal row with the UK over user data • BBC News

Zoe Kleinman:

»

WhatsApp has told the BBC it is supporting fellow tech giant Apple in its legal fight against the UK Home Office over the privacy of its users’ data.

The messaging app’s boss, Will Cathcart, said the case “could set a dangerous precedent” by “emboldening other nations” to seek to break encryption, which is how tech firms keep customers’ information private.

Apple went to the courts after receiving a notice from the Home Office demanding the right to access the data of its global customers if required in the interests of national security.

It and other critics of the government’s position say the request compromises the privacy of millions of users.
The Home Office told the BBC it would not comment on ongoing legal proceedings.

“But more broadly, the UK has a longstanding position of protecting our citizens from the very worst crimes, such as child sex abuse and terrorism, at the same time as protecting people’s privacy,” it said in a statement.

WhatsApp has applied to submit evidence to the court which is hearing Apple’s bid to have the Home Office request overturned.

Mr Cathcart said: “WhatsApp would challenge any law or government request that seeks to weaken the encryption of our services and will continue to stand up for people’s right to a private conversation online.”

This intervention from the Meta-owned platform represents a major escalation in what was an already extremely high-profile and awkward dispute between the UK and the US.

«

Interesting: by making this public, WhatsApp is bolstering its position as a secure messaging platform (it has an advertising campaign running now along exactly these lines – “we can’t see your messages”) but is also letting people know that the case hasn’t been decided, and that it is proceeding. Quite a lot from a fairly simple statement. And it previously threatened to withdraw from the UK rather than weaken its encryption or offer a backdoor.
unique link to this extract

---

Apple knows AI isn’t what people really want, but it can’t say that • Gizmodo

James Pero:

»

It’s clear Apple is under pressure to deliver AI features via Apple Intelligence, which is no surprise given the way rhetoric (and money) around AI has exploded, but what’s most interesting about that isn’t the pressure necessarily; it’s where the pressure is coming from.

As I wrote before the start of WWDC, in a lot of ways, Apple’s Siri stumbles, and AI issues writ large are more optical than consequential. That’s to say that people don’t necessarily care about AI features yet, and as a result, Apple likely doesn’t either—or not that it doesn’t care, it’s that it doesn’t care about rushing them out. Of course, it can’t really say that for the optical reasons I already mentioned. Despite the consumer-side collective shrug, investors are paying attention, and that may be exactly why, in an AI-light, year of WWDC, Apple’s stock immediately dipped following its keynote. But the fact remains: AI, while still on the roadmap, isn’t make-or-break for selling hardware yet.

Call me a skeptic, but I don’t think AI Siri will be the determining factor of whether or not most people rush out to buy this year’s iPhone. It’ll be cameras; it’ll be a thinner form factor; it’ll be the fact that they need a new phone and simply cannot bear the thought of switching to Android and getting green bubbled by the Messages app. If Apple is going to care about AI in the here and now, then it’ll be the result of consumers, not market forces.

That being said, Apple may need to care about AI a lot more in the near future. Progress in generative AI and LLMs has been rapid, and while skeptical of AI features, I’m not ruling out a watershed moment for AI on your phones or laptops quite yet. I’m also not ruling out that said watershed moment could actually come from Apple, despite the fact that it’s “behind” in rolling out Apple Intelligence features.

«

This is a very solid point. Apple has made onboard LLM processing available to app developers in the upcoming iOS release, and they will start integrating that processing into their apps. That, finally, might be the way that apps start becoming smart – all provisos about LLMs applying, of course.

And Pero is right: nobody is buying an “AI phone”. Google isn’t getting people buying the Pixel because of it; Samsung isn’t switching buyers over because of Gemini on its Galaxy phones. There is time before AI is something in our hands.
unique link to this extract

---

Apple’s “Liquid Glass” and what it means for accessibility • dreezus

Idrees Isse:

»

Tim Cook once said “When we work on making our devices accessible by the blind, I don’t consider the bloody ROI.”

Yesterday Apple unveiled their new “Liquid Glass” design language at WWDC 2025. The whole user interface becomes layers of translucent elements floating over your content, with glass-like behavior that refracts light and reacts to movement. It’s interesting stuff. But I’m wondering what this means for readability for the millions of people who struggle with traditional interfaces already.

Here we have a company that positions accessibility as a human right introducing a design system that seems at odds with basic accessibility principles. Like they built a glass house and forgot that not everyone can walk in it without bumping into walls.

I can’t even tell how I feel about how it looks yet. But I know I feel weird about how it works. Once you start thinking about stuff like contrast ratios, and the whole thing feels off.

Why should I care? I run a digital design agency called Mumino, where we create custom websites and brand identities. So I spend time thinking about WCAG, the Web Content Accessibility Guidelines. These guidelines require at least a 4.5:1 contrast ratio for normal text. That’s not some random number, it’s based on research about what people need to read comfortably. When you have translucent elements letting background colors bleed through, you’re creating variable contrast ratios that might work well over one background, but fail over a bright photo of the sunset.

Complying with accessibility guidelines is deep & complex. There are layers to this stuff that even experienced designers miss. I’m fortunate to have healthy eyesight, no disabilities, and I’m aware there’s always more to learn about creating fully inclusive experiences. But that’s exactly why it’s interesting to see Apple, a company with all the resources and expertise in the world, making design choices that raise a bunch of questions.

«

I’m extremely suspicious of this new design. Alan Dye, Apple’s design chief, doesn’t seem to have heard the saying that “no plan survives contact with the enemy”, where in this case the “plan” is his software designs and the “enemy” is a billion or so iPhone users. There is time, but not a lot, for it to be improved, but the main changes needed (tinting on overlays) may need years to win Dye’s approval. Design is how it works – or in this case, doesn’t.
unique link to this extract

---

How a hazelnut spread became a sticking point in Franco-Algerian relations • The New Yorker

Lauren Collins:

»

El Mordjene is produced by Cebon, a baking-products company based in Tipaza, near Algiers. It comes in two main varieties: cocoa hazelnut, colloquially known as “brown,” and hazelnut cream, colloquially known as “white.” (There is also rocher, a crunchy version of each color.) The spread débuted in 2021 on the Algerian market, where it retails for roughly five euros a jar. El Mordjene is not widely sold in France, but it has long been procurable in small quantities at slightly elevated prices, mostly in independent shops offering products from North Africa. The white kind is the most popular, for its silky texture and intensely sweet flavor, which is said to resemble the filling of a Kinder Bueno candy bar.

Last summer, as traffic between France and Algeria underwent its annual spike, influencers, many of them with Algerian connections, started touting the spread. “This thing, it’s sick!” one TikToker raved; another swore that it was “better than Nutella!”

…In early September, customs authorities at the port of Marseille and at Charles de Gaulle Airport, outside Paris, refused to allow two separate shipments of El Mordjene to enter French territory. Both lots—about a dozen pallets total, of eight hundred and forty jars each—belonged to independent importers, who had never before run into trouble. The rationale for the goods’ rejection wasn’t clear. On one form, customs officials wrote that the spread appeared “to infringe the trademark, designs, and model of the Ferrero Group,” the makers of Nutella, who sell tens of millions of jars in France each year.

For several days, the pallets remained in limbo. Then a decision came down from the French Ministry of Agriculture, which insures food safety, formalizing the ban. Apparently, trademark infringement wasn’t the issue; the problem was powdered milk and whey—ingredients in El Mordjene, alongside sugar, hazelnuts, vegetable fat, and emulsifiers. E.U. law limits the importation of products that contain even a small amount of dairy; Algeria does not appear on a list of approved countries. The offending jars of El Mordjene would therefore have to be repatriated or sent elsewhere. A ministry spokesman vowed that it would open an investigation “to determine the mechanisms of circumvention that until now have allowed this product to be placed on the market.”

«

Next? Gunpoint thefts, racial tensions, political tensions…
unique link to this extract

---

Curious humpback whales approach humans and blow bubble “smoke” rings • Whale SETI

»

A team of scientists from the SETI Institute and the University of California at Davis documented, for the first time, humpback whales producing large bubble rings, like a human smoker blowing smoke rings, during friendly interactions with humans. This previously little studied behavior may represent play or communication. Humpback whales are already known for using bubbles to corral prey and creating bubble trails and bursts when competing to escort a female whale. These new observations show humpback whales producing bubble rings during friendly encounters with humans. This finding contributes to the WhaleSETI team’s broader goal of studying non-human intelligence to aid in the search for extraterrestrial intelligence.

“Because of current limitations on technology, an important assumption of the search for extraterrestrial intelligence is that extraterrestrial intelligence and life will be interested in making contact and so target human receivers,” said Dr. Laurance Doyle, SETI Institute scientist and co-author on the paper. “This important assumption is certainly supported by the independent evolution of curious behavior in humpback whales.”

“Humpback whales live in complex societies, are acoustically diverse, use bubble tools and assist other species being harassed by predators,” said co-lead author Dr. Fred Sharpe, UC Davis Affiliate. “Now, akin to a candidate signal, we show they are blowing bubble rings in our direction in an apparent attempt to playfully interact, observe our response, and/or engage in some form of communication.”

«

It’s basically Arrival, but underwater, isn’t it?
unique link to this extract

---

Saab achieves AI milestone with Gripen E • Saab Newsroom

»

Saab, in collaboration with Helsing, today announced the successful completion of the first three flights integrating Helsing’s Artificial Intelligence (AI) agent ‘Centaur’ into a Gripen E fighter jet. As part of Saab’s ‘Project Beyond’, the flights, where the first flight, was conducted on May 28, mark a significant advance in bringing AI capabilities to military aircraft. It is also yet another proof point of Gripen E´s unparalleled ability to rapidly update software without disregarding safety requirements.

During the flights, the Gripen E gave control to Centaur which successfully autonomously executed complex manoeuvres in a Beyond Visual Range (BVR) combat environment and cued the pilot to fire.

“This is an important achievement for Saab, demonstrating our qualitative edge in sophisticated technologies by making AI deliver in the air. The swift integration and successful flight testing of Helsing’s AI in a Gripen E exemplifies the accelerated capability gain you can get from our fighter. We are excited to continue developing and refining how this and other AI agents can be used, while once again showing how our fighters will outperform faster than the opponent can evolve,” said Peter Nilsson, head of Advanced Programmes, from Saab’s Aeronautics Business Area.

Thanks to the unique design of Gripen E, the fighter can fly with the AI software onboard and fully integrated without being restricted to solely military test ranges or have to rely on an experimental X-plane to do flight trials with the software.

“Within Project Beyond and other programmes, we utilise the power of software to rapidly explore and blur the lines between “now” and the future; in software there are no generations, only speed,” says Peter Nilsson.

«

Saab, in case you didn’t know, stopped making cars a while back, but still does fighter jets – now with added AI. How soon before the pilot is superfluous? (Thanks Peter R for the link, who remarked “seems like another Black Mirror moment.” Indeed!)
unique link to this extract

---

AI chatbots are making LA protest disinformation worse • WIRED

David Gilbert:

»

Disinformation about the Los Angeles protests is spreading on social media networks and is being made worse by users turning to AI chatbots like Grok and ChatGPT to perform fact-checking.

As residents of the LA area took to the streets in recent days to protest increasingly frequent Immigration and Customs Enforcement (ICE) raids, conservative posters on social media platforms like X and Facebook flooded their feeds with inaccurate information. In addition to well-worn tactics like repurposing old protest footage or clips from video games and movies, posters have claimed that the protesters are little more than paid agitators being directed by shadowy forces—something for which there is no evidence.

In the midst of fast-moving and divisive news stories like the LA protests, and as companies like X and Meta have stepped back from moderating the content on their platforms, users have been turning to AI chatbots for answers—which in many cases have been completely inaccurate.

On Monday, the San Francisco Chronicle published images of National Guard troops sleeping on floors. They were later shared on X by California governor Gavin Newsom, who responded to a post from President Donald Trump by writing: “You sent your troops here without fuel, food, water or a place to sleep.”

Within minutes of the posts being shared, many users on X and Facebook were claiming that the images were either AI-generated or taken from a completely different situation.

«

The number of people who now lean on Grok or similar for their “fact-checking” is just appalling. There should be a health warning on these systems. But of course, if they did that, then people wouldn’t use them as much. It’s not in the tech companies’ interests to tell people about the limitations of their systems.
unique link to this extract

---

Large language models and pareidolia • Terence Eden’s Blog

Terence Eden:

»

Have you ever looked up at the sky and seen a face staring back at you from the clouds? Of course you have; you’re human. Our delicious meaty brains are hardwired to recognise certain shapes – and faces are a useful shape to recognise. A few false positives are a worthwhile trade-off for such a powerful feature.

Mistakenly seeing faces where there are none is a phenomenon called pareidolia. If you’ve ever used facial recognition on a computer, you’ll know that machines also suffer from it.

…LLMs are hardwired to regurgitate text which statistically matches what they’ve seen before. Their makers believe that a few false-positives are an acceptable error rate for such a useful feature. The LLM form of pareidolia is to recognise text as being syntactically and linguistically correct, even though the contents are rubbish. This is an inherent feature of LLMs. No amount of manually tweaking their parameters or prompts can fix this.

At the moment, Artificial Intelligence – whether Machine Learning or Large Language Models – only works well on a narrowly defined set of tasks and with humans checking the output.

Imagine you’ve just hired an intern. They’ve graduated top of their class from the best university and, apparently, excel at what they do. Because you’re the boss and they’re the intern, you ask them to make you a mug of tea. White, no sugar.
They return with the teabag still in the mug. OK, not everyone knows the intricacies of how to serve tea.

The tea tastes funny. You ask them if they sniffed the milk. “Milk? I used Tipp-Ex to make it white!”

At which point, after throwing up, you throw them out.

Most people encountering [Google] Gemini’s repeated and unacceptable failures will decide, perhaps rightly, that AI isn’t even close to good enough yet.

«

The space where LLMs are good and useful is very limited – coding, for example, because it’s a very limited subset of language with strict rules.
unique link to this extract

---

Bluesky isn’t a bubble. It’s a containment dome • Very Serious

Josh Barro:

»

My friend Megan McArdle warns in a column that the social media platform Bluesky is a harmful bubble for liberals. By decamping together for Bluesky, she writes, liberals have cloistered themselves in a place where their views won’t be challenged. And because the conversational norms on Bluesky are so hostile and obnoxious — do you ever use AI? Former “Reply All” host Alex Goldman wants you to know you should be thrown into a volcano — the platform fails to appeal beyond its niche political audience, is losing users, and is unlikely to become a place where posting is a good way to influence public opinion.

Megan correctly describes these dynamics, but she’s wrong about them being harmful. In fact, these dynamics are why Bluesky is an important harm reduction tool for liberals. Twitter used to be a place where the most neurotic and censorious liberal influencers were highly effective at influencing events within media organizations and the Democratic Party. But was that actually ever good for liberal causes?

…A lot of the blame for the self-inflicted wounds of the 2020 Democratic presidential primary should go to The Groups: it was the ACLU that got Kamala Harris to commit to taxpayer-funded sex changes for criminals and detained migrants. But one of the reasons Democrats didn’t realize it was a big mistake to make promises and statements that made them sound wacky was that they were constantly being yelled at on Twitter by people whose unpopular viewpoints they mistook for broad public opinion. The screamers won the battle but they lost the war: they pressured their own candidates into manufacturing attack ad fodder for Republicans, and as a result, Donald Trump is president again.

…By rupturing the Twitter user base, [Elon Musk] (accidentally?) created a firewall between the most maladjusted liberal posters on the internet and the reporters, Democratic politicians and operatives who used to pay an excessive amount of attention to their harangues. (Media reporter Max Tani wrote about this for Semafor last month: “I spoke with a few congressional staffers who said that they had tried using Bluesky as an alternative to Twitter after Twitter was purchased by Elon Musk, but they gave up after their bosses kept getting yelled at by Democratic users angry at their impotence.”)

I believe the emergence of this firewall is one reason for the renaissance that we were seeing at WelcomeFest last week: Democrats are becoming more cognizant of public opinion and less fearful of breaking with the activist base because they are no longer receiving so much activist messaging in the form of aggrieved Twitter push alerts on their phones.

«

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