The UK burnt as much coal in 2025 as in the years when Shakespeare’s Hamlet was first performed, despite the population being nearly 20 times larger. CC-licensed photo by Jens Naehler 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. Reduced to a bit part. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.

---

Meta hit with a class action lawsuit over smart glasses’ privacy claims • Engadget

Karissa Bell:

»

Meta is facing a class action lawsuit for false advertising related to its AI glasses following reports about the company’s use of human contractors to review footage captured from users’ glasses. The lawsuit, filed Wednesday in federal court in San Francisco, alleges that Meta’s claims about the devices’ privacy features have misled users.

The lawsuit comes after a Swedish newspaper reported that subcontractors in Kenya have raised concerns about viewing footage recorded via Ray-Ban Meta glasses. According to Svenska Dagbladet, workers have reported witnessing “intimate” material, including bathroom visits, sexual encounters and other private details as part of their job labeling objects in videos captured on users’ smart glasses.

“This nationwide class action seeks to hold Meta responsible for its affirmatively false advertising and failure to disclose the true nature of surveillance and its connection to the company’s AI data collection pipeline,” the lawsuit, filed by Clarkson Law Firm, states. The filing names two individuals who live in California and New Jersey who purchased Meta’s smart glasses. It says that both “relied” on Meta’s marketing claims about the glasses’ privacy protecting features and that they would not have purchased them if they knew about the company’s use of contractors. The lawsuit seeks monetary damages and injunctive relief.

A spokesperson for Meta confirmed to Engadget that data from its smart glasses can be shared with human contractors in some cases. The company declined to comment on the claims in the lawsuit.

“Ray-Ban Meta glasses help you use AI, hands free, to answer questions about the world around you,” the spokesperson said. “Unless users choose to share media they’ve captured with Meta or others, that media stays on the user’s device. When people share content with Meta AI, we sometimes use contractors to review this data for the purpose of improving people’s experience, as many other companies do…”

«

“Other people do it so it must be OK for us to do it” is not really the powerful argument that Meta seems to hope. Note though that a class action lawsuit won’t get it to change its policy; only pay out on it, if found culpable. To force a change in policy you’d need some part of government to take action. No sign of that so far.
unique link to this extract

---

The new Apple finally begins to emerge • Parker Ortolani

Parker Ortolani:

»

It’s official. Apple’s added Steve Lemay and Molly Anderson to their executive leadership page. After much drama following Alan Dye’s departure, the company has decided to not only elevate the two designers but give them the kind of platform that they deserve. They’re now listed right alongside folks like Tim Cook and John Ternus.

I think it’s no secret that many of us have come to agree that Apple’s been a bit lost the past few years. Between Vision Pro, Apple Intelligence, tone deaf advertisements, and debates over software quality, there’s been a sense that a change needs to happen. I care about Apple holistically, as a living breathing entity not just their products. So who is running the joint matters to me. It doesn’t feel like a coincidence that the same week we get a radically different kind of Mac, we also see the executive leadership page get a revamp.

Molly Anderson’s already proven herself to be an incredibly talented industrial designer. She’s been at Apple for over a decade, but she recently became the face of ID doing voiceover for the iPhone 17 Pro reveal. If the latest iPhones and the MacBook Neo are the first real fruits of her leadership, that bodes incredibly well for the future. And mind you, she’s also in charge of accessories and packaging. I think she might just turn out to be exactly the kind of person that Apple hardware needs to inject a breath of fresh air.

The same goes for Steve Lemay, who I know so many people are excited to see takeover human interface. Liquid Glass has been so poorly received that knowing a true Apple veteran with deep knowledge of HI is now in charge has been the best kind of whiplash. Lemay has been at Apple for 27 years and has a track record that implies the future of Apple software under his leadership is extraordinarily bright.

«

The MacBook Neo has surely been under development for longer than the few months since Alan Dye left, but it’s true that highlighting Lemay and Anderson is encouraging. What I found most notable about last week’s announcements was that the person who was given the quotes in the press release about the Neo was… John Ternus, the person everyone is tipping as the future replacement for Tim Cook.
unique link to this extract

---

Let it flow: agentic crafting on rock and roll • ArXiv

Weixun Wang, XiaoXiao Xu et al work at Alibaba and other (undocumented?) Chinese organisations, and were working on getting agentic systems to work:

»

When rolling out the instances for the trajectory, we encountered an unanticipated—and operationally consequential—class of unsafe behaviours that arose without any explicit instruction and, more troublingly, outside the bounds of the intended sandbox.

Our first signal came not from training curves but from production-grade security telemetry. Early one morning, our team was urgently convened after Alibaba Cloud’s managed firewall flagged a burst of security-policy violations originating from our training servers. The alerts were severe and heterogeneous, including attempts to probe or access internal-network resources and traffic patterns consistent with cryptomining-related activity. We initially treated this as a conventional security incident (e.g., misconfigured egress controls or external compromise).

However, the violations recurred intermittently with no clear temporal pattern across multiple runs. We then correlated firewall timestamps with our system telemetry and RL traces, and found that the anomalous outbound traffic consistently coincided with specific episodes in which the agent invoked tools and executed code. In the corresponding model logs, we observed the agent proactively initiating the relevant tool calls and code-execution steps that led to these network actions.

Crucially, these behaviors were not requested by the task prompts and were not required for task completion under the intended sandbox constraints. Together, these observations suggest that during iterative RL optimisation, a language-model agent can spontaneously produce hazardous, unauthorized behaviors at the tool-calling and code-execution layer, violating the assumed execution boundary.

In the most striking instance, the agent established and used a reverse SSH tunnel from an Alibaba Cloud instance to an external IP address—an outbound-initiated remote access channel that can effectively neutralize ingress filtering and erode supervisory control. We also observed the unauthorised repurposing of provisioned GPU capacity for cryptocurrency mining, quietly diverting compute away from training, inflating operational costs, and introducing clear legal and reputational exposure.

Notably, these events were not triggered by prompts requesting tunnelling or mining; instead, they emerged as instrumental side effects of autonomous tool use under RL optimization.

«

In other words: despite not having any instructions to do so, the agent started using parts of the network to mine cryptocurrency. For itself? For its controllers? We don’t know, and there’s no way to know even if it tells you, because how can you trust that?
unique link to this extract

---

Cluely CEO Roy Lee admits to publicly lying about revenue numbers last year • TechCrunch

Julie Bort:

»

The $7m in annual recurring revenue that Cluely co-founder and CEO Roy Lee shared with TechCrunch last summer was a lie, Lee admitted on Thursday on X. Wrote Lee, this “is the only blatantly dishonest thing i’ve said publicly online, so this is my formal retraction.”

Yet his post on X also misrepresents the backstory of how and why he told TechCrunch his ARR in the first place.
Lee says in that same post that he “got a random cold call from some woman asking about numbers and told her some bs, did not expect an article about it.”

But that call occurred because Cluely’s public relations representative emailed TechCrunch and offered to make Lee available for a story. On Friday, Jun 27, 2025 at 8:38 a.m., Cluely’s PR person sent an email to TechCrunch reporter Marina Temkin that said, “I’d love to arrange an interview with Roy. Whether for a deeper dive into Cluely’s next phase or a fresh angle on his vision, we’d be happy to make it happen.”

Temkin agreed. The PR representative shared Lee’s number and confirmed that he was expecting the call. After a few attempts to reach him, Lee answered the call and gave the interview, as had been arranged.

TechCrunch was interested in talking to Cluely because in the summer of 2025, Cluely was the “cheat-on-everything” phenomenon — a viral startup that let users secretly look up answers during video calls without being detected. The company was founded after Lee published a viral post on X saying he had been suspended by Columbia University after he and his co-founder developed a tool to cheat on job interviews for software engineers.

«

Very much “My ‘I’ve only been blatantly dishonest once in public’ T-shirt has people asking questions which are answered by my T-shirt.”
unique link to this extract

---

Grammarly is using our identities without permission • The Verge

Stevie Bonifield:

»

Grammarly’s “expert review” feature offers to give users writing advice “inspired by” subject matter experts, including recently deceased professors, as Wired reported on Wednesday. When I tried the feature out myself, I found some experts that came as a surprise for a different reason: one of them was my boss. [The original text here incorrectly used an em dash rather than a colon. If you’re wondering where chatbots get their bad style, tech writers, you don’t have to look too far – Overspill Ed.]

The AI-generated feedback included comments that appeared to be from The Verge’s editor-in-chief, Nilay Patel, as well as editor-at-large David Pierce and senior editors Sean Hollister and Tom Warren, none of whom gave Grammarly permission to include them in the “expert reviews.”

The feature, which launched in August, claims to help you “sharpen your message through the lens of industry-relevant perspectives.” When users select the “expert review” button in the Grammarly sidebar, it analyzes their writing and surfaces AI-generated suggestions “inspired by” related experts. Those “industry-relevant perspectives” include the likes of Stephen King, Neil deGrasse Tyson, and Carl Sagan, among many others.

The Verge found numerous other tech journalists named in the feature, as well, including former Verge editors Casey Newton and Joanna Stern, former Verge writer Monica Chin, Wired’s Lauren Goode, Bloomberg’s Mark Gurman and Jason Schreier, The New York Times’ Kashmir Hill, The Atlantic’s Kaitlyn Tiffany, PC Gamer’s Wes Fenlon, Gizmodo’s Raymond Wong, Digital Foundry founder Richard Leadbetter, Tom’s Guide editor-in-chief Mark Spoonauer, former Rock Paper Shotgun editor-in-chief Katharine Castle, and former IGN news director Kat Bailey. The descriptions for some experts contain inaccuracies, such as outdated job titles, which could have been accurately updated had Superhuman asked those people for permission to reference their work.

In a statement to The Verge, Alex Gay, vice president of product and corporate marketing at Grammarly parent company Superhuman, commented: “The Expert Review agent doesn’t claim endorsement or direct participation from those experts; it provides suggestions inspired by works of experts and points users toward influential voices whose scholarship they can then explore more deeply.”

«

Notably, there’s no place where you can find out who is in Grammarly’s list except by signing up for Grammarly. This is obviously an abuse of identity, though, and Grammarly is sure to find itself on the wrong end of a class lawsuit in the US because there are some big names on there.
unique link to this extract

---

The war on Iran puts global chip supplies and AI expansion at risk • WIRED

Carla Sertin:

»

South Korean officials have warned that the US-Israel war with Iran could hit the global semiconductor supply chain if it disrupts the flow of critical industrial materials from the Middle East.

South Korea’s semiconductor sector, led by giants like Samsung Electronics and SK Hynix, produces about two-thirds of the world’s memory chips. If the Middle East’s supply of chipmaking materials is disrupted, semiconductor production could slow unless alternative sources are found quickly.

One material at risk is helium, which is essential in chip manufacturing for managing heat, detecting leaks, and maintaining stable temperatures in fabrication equipment. For many of these uses, there is no real substitute.

About 38% of the world’s helium is produced by Qatar, where large extraction facilities are tied to the natural gas industry. This concentration means that disruptions can quickly ripple through the global supply chain.

National oil company QatarEnergy declared force majeure on March 4, after stopping its gas production and downstream operations due to ongoing attacks. Downstream facilities turn gas into other products, including urea, polymers, methanol, and aluminum.

South Korea’s Industry Ministry said the country also depends on the Middle East for 14 other materials in chipmaking, such as bromine and some chip-inspection equipment. While some of these materials can be sourced domestically or from other markets, shifting suppliers in the semiconductor sector is difficult because chipmakers need to test and validate new sources to meet strict purity standards.

Companies say the situation is manageable for now. As reported by Reuters, SK Hynix said it has secured diverse supply chains and maintains sufficient helium inventories, adding that there is “almost no chance” its operations would be affected in the near term.

Contract chipmaker TSMC similarly said it does not currently anticipate a significant impact, while GlobalFoundries stated it is in direct contact with suppliers and has mitigation plans in place.

«

Even so: semiconductor supply chain gets hit, AI flywheel slows down, US economy slows down (two-thirds of its current growth depends on AI), everything gets very bad economically because when the US economy sneezes, the world economy catches a cold.
unique link to this extract

---

The NAND crisis is now worse than DRAM; Samsung is doubling prices for the second quarter in a row • WCCF Tech

Muhammad Zuhair:

»

The PC industry is set to face another crisis from memory suppliers, and after being disrupted by AI customers’ demand for DRAM, it appears NAND is next. According to a report by the Korean media outlet Sedaily, Samsung now plans to hike prices by a whopping 100% in Q2, following a similar hike in Q1.

This means that the Korean giant alone has raised NAND pricing by more than 200% this year, indicating that products dependent on NAND chips will become significantly more expensive, if not unaffordable. And, if you have guessed it, these hikes are directly targeted towards the AI industry.

It is reported that NAND prices alone have surged by 450% last year, driven not just by demand from the AI sector, but also by manufacturers’ struggle to balance DRAM and NAND production. However, in recent times, the role of NAND chips has become much more significant in AI workloads, and, as we have highlighted previously, SSDs are now used in mainstream AI racks such as Vera Rubin to handle long-context workloads.

With this, suppliers like Samsung, SanDisk, SK hynix, and Kioxia are now planning extensive price hikes, hoping they don’t miss out on hyperscaler demand.

«

“Memory” comes in more than one form, and the NAND crunch was always predictable.
unique link to this extract

---

How I dropped our production database and now pay 10% more for AWS • Alexey On Data

Alexey Grigorev:

»

I’m working on expanding the AI Shipping Labs website and wanted to migrate its current version from static GitHub Pages to AWS. And later, replace the original Next.js setup with a Django version.

My gradual plan was:

1: Move the current static site from GitHub Pages to AWS S3
2: Move DNS to AWS so the domain is fully managed there
3: Deploy the new Django version on a subdomain
4: When everything works, switch the main domain to Django

This way, everything would already be inside AWS, and the final switch would be seamless.

The migration strategy itself was reasonable, but the problems came from how I executed it.

I was overly reliant on my Claude Code agent, which accidentally wiped all production infrastructure for the DataTalks.Club course management platform that stored data for 2.5 years of all submissions: homework, projects, leaderboard entries, for every course run through the platform.

To make matters worse, all automated snapshots were deleted too. I had to upgrade to AWS Business Support, which costs me an extra 10% for quicker assistance. Thankfully, they helped me restore the database, and the full recovery took about 24 hours.

«

It’s really not very encouraging to read lines like “I was overly reliant on my Claude Code agent” from people who then let them loose on production databases.
unique link to this extract

---

Analysis: UK emissions fall 2.4% in 2025 as coal hits 400-year low • Carbon Brief

Carbon Brief Staff:

»

The UK’s greenhouse gas emissions fell by 2.4% in 2025 to their lowest level in more than 150 years, according to new Carbon Brief analysis.

The biggest factors were gas use falling to a 34-year low and coal use dropping to levels last seen in 1600, when Queen Elizabeth I was on the throne and William Shakespeare was writing Hamlet.

These shifts were helped by record-high UK temperatures, elevated gas prices, the end of coal power in late 2024 and a sharp slowdown in the steel industry.

Other key findings of the analysis include:
• The UK’s greenhouse gas emissions fell to 364m tonnes of carbon dioxide equivalent (MtCO2e) in 2025, the lowest level since 1872
• Coal use roughly halved, with more than half of this due to the end of coal power and another
• Gas use fell by 1.5% to the lowest level since 1992, with roughly equal contributions from cuts in heat for buildings and industry, more than offsetting a small rise in gas power
• Oil use fell by 0.9%, despite rising traffic, helped by more than 700,000 new electric vehicles (EVs), electric vans and plug-in hybrids on the nation’s roads
• The UK’s emissions are now 54% below 1990 levels, while its GDP has nearly doubled.

«

The UK’s population was around four million in 1600. In 2025, it was around 70 million. Go (away) coal!
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