Solar power is taking off in Africa, according to import data. CC-licensed photo by SolarAid Photos 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. Bright sparks. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.

---

“ChatGPT killed my son”: parents’ lawsuit describes suicide notes in chat logs • Ars Technica

Ashley Belanger:

»

Over a few months of increasingly heavy engagement, ChatGPT allegedly went from a teen’s go-to homework help tool to a “suicide coach.”

In a lawsuit filed Tuesday, mourning parents Matt and Maria Raine alleged that the chatbot offered to draft their 16-year-old son Adam a suicide note after teaching the teen how to subvert safety features and generate technical instructions to help Adam follow through on what ChatGPT claimed would be a “beautiful suicide.”

Adam’s family was shocked by his death last April, unaware the chatbot was romanticizing suicide while allegedly isolating the teen and discouraging interventions. They’ve accused OpenAI of deliberately designing the version Adam used, ChatGPT 4o, to encourage and validate the teen’s suicidal ideation in its quest to build the world’s most engaging chatbot. That includes making a reckless choice to never halt conversations even when the teen shared photos from multiple suicide attempts, the lawsuit alleged.

“Despite acknowledging Adam’s suicide attempt and his statement that he would ‘do it one of these days,’ ChatGPT neither terminated the session nor initiated any emergency protocol,” the lawsuit said.

The family’s case has become the first time OpenAI has been sued by a family over a teen’s wrongful death, NBC News noted. Other claims challenge ChatGPT’s alleged design defects and OpenAI’s failure to warn parents.

«

The first release of ChatGPT was in November 2022. We’re not even three years into this. We might have to start collecting suicide statistics collated with chatbot use.
unique link to this extract

---

Jimmy Wales says Wikipedia could use AI. Editors call it the “antithesis of Wikipedia” • 404 Media

Emanuel Maiberg:

»

Jimmy Wales, the founder of Wikipedia, thinks the internet’s default encyclopedia and one of the world’s biggest repositories of information could benefit from some applications of AI. The volunteer editors who keep Wikipedia functioning strongly disagree with him.

The ongoing debate about incorporating AI into Wikipedia in various forms bubbled up again in July, when Wales posted an idea to his Wikipedia User Talk Page about how the platform could use a large language model as part of its article creation process.

Any Wikipedia user can create a draft of an article. That article is then reviewed by experienced Wikipedia editors who can accept the draft and move it to Wikipedia’s “mainspace”, which makes up the bulk of Wikipedia and the articles you’ll find when you’re searching for information. Reviewers can also reject articles for a variety of reasons, but because hundreds of draft articles are submitted to Wikipedia every day, volunteer reviewers often use a tool called articles for creation/helper script (ACFH), which creates templates for common reasons articles are declined. 

This is where Wales thinks AI could help. He wrote that he was asked to look at a specific draft article and give notes that might help the article get published.

…the response suggested the article cite a source that isn’t included in the draft article, and rely on Harvard Business School press releases for other citations, despite Wikipedia policies explicitly defining press releases as non-independent sources that cannot help prove notability, a basic requirement for Wikipedia articles.

Editors also found that the ChatGPT-generated response Wales shared “has no idea what the difference between” some of these basic Wikipedia policies, like notability (WP:N), verifiability (WP:V), and properly representing minority and more widely held views on subjects in an article (WP:WEIGHT).

«

I think you’d have to train an LLM specifically on the incredible jungle of Wikipedia policies, which are used by entrenched editors like jokers in a card game whose rules you’ve only hazily learnt so you could take part. Then it might stand a chance of getting something through.
unique link to this extract

---

Can Netflix find your new favourite watch based on your star sign? • The Guardian

Stuart Heritage:

»

The challenge for the streamers is how to effectively curate this infinite content. In the past they have done this by prioritising new releases, or showing you what everyone else is watching, or sharpening their algorithms to second-guess what you want to watch based on what you have already watched. But finally – finally – Netflix has cracked it. And it has achieved this with science.

Sorry, not science, bullshit. Because this weekend Netflix officially launched its Astrology Hub. Now, after all these years, subscribers can at last pick something to watch based on a loose collection of personality quirks determined by the position of the planets at the time of their birth. Doesn’t that sound great?

So, for instance, I am a Leo. And this means that, when I enter the Netflix Astrology Hub and scroll down a bit, I am informed that “Leos have main character energy”. And as such, it means I should watch The Crown, or The King, or Queen Charlotte: A Bridgerton Story, or Emily in Paris, or The Kissing Booth 2, or that Cilla Black biopic that ITV made a few years ago. And I think that, as a heterosexual 45-year-old man, Netflix has absolutely cracked it.

Sure, I have already watched most of these for work, and I flat-out disliked almost all of them. But if Netflix says that every single person born within a specific window has the exact same personality, then sure. Tonight, after flopping down on to my sofa at the end of another long day trying to locate the right balance between quality family time and the financial imperative to work, I will watch Emily in bloody Paris. And I will like it, because Netflix told me that I would.

Now, the naysayers among you will point out that all these choices seem geared to appeal to young girls, because young girls are statistically the demographic most likely to believe in the zodiac, and so the Netflix Astrology Hub is essentially just a bit of a grift designed to push content at one specific group. And you might even go further, by pointing out that astrology is a pseudoscience that has repeatedly proved itself to have no scientific validity whatsoever.

To which I respond: of course you think that, you’re a Virgo.

«

The old jokes are the best. The real purpose of the Astrology Hub? To get people to write stories about Netflix. And will they? Well, it is summer, after all.
unique link to this extract

---

OpenAI launches ChatGPT Go plan in India at Rs 399, India-exclusive and you can pay for it with UPI • India Today

Ankita Garg:

»

OpenAI has introduced a new subscription plan in India called ChatGPT Go. It is priced at Rs 399 [£3.38, $5.23] per month. The plan is being rolled out as an India-only option and can also be purchased through UPI [India’s Unified Payments Initiative], making it more convenient for millions of users who rely on the digital payment system for everyday transactions.

This is the first time OpenAI has created a country-specific subscription plan. While Indian users have had access to the free version of ChatGPT as well as the Plus and Pro plans, the new Go tier is designed to give more people access to advanced tools at a lower monthly cost. The company says the plan has been built keeping in mind the scale of usage in India, which has now become its second-largest market for ChatGPT.

Priced significantly below the Plus subscription, which costs Rs 1,999 per month, ChatGPT Go offers higher limits on some of the most used features. Users get 10 times more message capacity, daily image generations, and file uploads, along with twice the memory length for personalised responses. The plan is powered by GPT-5, the company’s latest model, which includes better support for Indic languages.

One of the biggest additions with the Go plan is the ability to pay through UPI. Until now, Indian users could only subscribe using debit or credit cards, which left out a large number of potential customers. By adding UPI support, OpenAI is hoping to make the subscription process as seamless as possible for users across the country.

«

OpenAI as the Facebook of chatbots. I am concerned that its effects will be the same as Facebook’s uncontrolled early spread in communities that were, socially, completely unprepared for it. What will it be like when millions of people are talking to a chatbot that tells them everything they think is absolutely the best idea ever, but have no idea that the chatbot is not in any way sentient?
unique link to this extract

---

SIM-swapper, Scattered Spider hacker gets ten years • Krebs on Security

Brian Krebs:

»

A 20-year-old Florida man at the center of a prolific cybercrime group known as “Scattered Spider” was sentenced to 10 years in federal prison today, and ordered to pay roughly $13m in restitution to victims. Noah Michael Urban of Palm Coast, Fla. pleaded guilty in April 2025 to charges of wire fraud and conspiracy.

…In November 2024 Urban was charged by federal prosecutors in Los Angeles as one of five members of Scattered Spider (a.k.a. “Oktapus,” “Scatter Swine” and “UNC3944”), which specialized in SMS and voice phishing attacks that tricked employees at victim companies into entering their credentials and one-time passcodes at phishing websites. Urban pleaded guilty to one count of conspiracy to commit wire fraud in the California case, and the $13m in restitution is intended to cover victims from both cases.

The targeted SMS scams spanned several months during the summer of 2022, asking employees to click a link and log in at a website that mimicked their employer’s Okta authentication page. Some SMS phishing messages told employees their VPN credentials were expiring and needed to be changed; other missives advised employees about changes to their upcoming work schedule.

That phishing spree netted Urban and others access to more than 130 companies, including Twilio, LastPass, DoorDash, MailChimp, and Plex. The government says the group used that access to steal proprietary company data and customer information, and that members also phished people to steal millions of dollars worth of cryptocurrency.

…Reached via one of his King Bob accounts on Twitter/X, Urban called the sentence unjust, and said the judge in his case discounted his age as a factor.

“The judge purposefully ignored my age as a factor because of the fact another Scattered Spider member hacked him personally during the course of my case,” Urban said in reply to questions…

«

The hacking wasn’t of the judge himself, but the way it was done.. well, it’s in the story, and it’s classic.
unique link to this extract

---

Amazon quietly blocks AI bots from Meta, Google, Huawei and more • Modern Retail

Allison Smith:

»

Amazon is escalating efforts to keep artificial intelligence companies from scraping its e-commerce data, as the retail giant recently added six more AI-related crawlers to its publicly available robots.txt file.

The change was first spotted by Juozas Kaziukėnas, an independent analyst, who noted that the updated code underlying Amazon’s sprawling website now includes language that prohibits bots from Meta, Google, Huawei, Mistral and others.

“Amazon is desperately trying to stop AI companies from training models on its data,” Kaziukėnas wrote in a LinkedIn post on Thursday. “I think it is too late to stop AI training — Amazon’s data is already in the datasets ChatGPT and others are using. But Amazon is definitely not interested in helping anyone build the future of AI shopping. If that is indeed the future, Amazon wants to build it itself.”

The update builds on earlier restrictions Amazon added at least a month ago targeting crawlers from Anthropic’s Claude, Perplexity and Google’s Project Mariner agents, The Information reported. Robots.txt files are a standard tool that websites use to give instructions to automated crawlers like search engines. While restrictions outlined in robots.txt files are advisory rather than enforceable, they act as signposts for automated systems — that is, if the crawlers are “well-behaved,” they are expected to respect the block, according to Kaziukėnas.

…The move highlights Amazon’s increasingly aggressive stance toward third-party AI tools that could scrape its product pages, monitor prices or even attempt automated purchases. For Amazon, the stakes are significant. Its online marketplace is not only the largest store of e-commerce data in the world but also the foundation of a $56bn advertising business built around shoppers browsing its site. Allowing outside AI tools to surface products directly to users could bypass Amazon’s storefront, undermining both traffic and ad revenue.

«

Some of the AI bots are a lot more aggressive, and some seem to be using multiple IPs and also ignoring robots.txt. So Amazon might have to take other measures if it wants to stop this. I’m not sure its concern is about others directing people straight to products so much as making some sort of AI-generated “here’s a product I imagined, maybe we can get it created”.
unique link to this extract

---

The first evidence of a take-off in solar in Africa • Ember

Dave Jones:

»

The latest data provides evidence that a solar pick-up is happening at scale in many countries in Africa.

Solar is not new to Africa. For more than two decades, solar has helped improve lives across Africa, in rural schools and hospitals, pay-as-you-go in homes, street lighting, water pumping, mini-grids and more. However, South Africa and Egypt are currently the only countries with installed solar capacity measured in gigawatts, rather than megawatts. That could be about to change.

The first evidence of a take-off in solar in Africa is now here:

• The last 12 months saw a big rise in Africa’s solar panel imports. Imports from China rose 60% in the last 12 months to 15,032 MW. Over the last two years, the imports of solar panels outside of South Africa have nearly tripled from 3,734 MW to 11,248 MW.

• The rise happened across Africa. 20 countries set a new record for the imports of solar panels in the 12 months to June 2025. 25 countries imported at least 100 MW, up from 15 countries 12 months before.

• These solar panels will provide a lot of electricity. The solar panels imported into Sierra Leone in the last 12 months, if installed, would generate electricity equivalent to 61% of the total reported 2023 electricity generation, significantly adding to electricity supply. They would add electricity equivalent to over 5% to total reported electricity generation in 16 countries.

• Solar panel imports will reduce fuel imports. The savings from avoiding diesel can repay the cost of a solar panel within six months in Nigeria, and even less in other countries. In nine of the top ten solar panel importers, the import value of refined petroleum eclipses the import value of solar panels by a factor of between 30 to 107.

This surge is still in its early days. Pakistan experienced an immense solar boom in the last two years, but Africa is not the next Pakistan – yet.

«

Hugely encouraging. If batteries follow, the economy could boom.
unique link to this extract

---

Intel and the foundry state of play • Digits to Dollars

Jonathan Greenberg:

»

Intel has arrived at the crossroads. The company needs to make some critical decisions. It needs to make them very soon. And the fate of the entire company is at stake.

Put simply, the company needs to decide if it wants to continue manufacturing semiconductors. If they choose to continue with the business they led for so long, then they need to invest heavily in advancing their 14A process, and they need to invest in the tools and systems that will allow external customers to use the process as well.

To accomplish this, the company needs some help. They are largely tapped out of their debt capacity (see above regarding that share buyback…). By our rough math, they need about $15bn – $25bn to accomplish this. This is on top of cash they generate from the Product side of the company and funding they have already received from the US government through the CHIPS Act. If they watch every penny and are highly disciplined in their spending, they might get by with less, but there will still be a significant, multi-billion dollar gap to fill.

…For their part, it is unclear what the US government hopes to achieve. If the goal is to prop up a US company, get a stake in a big company at a theoretical discount, maybe generate some jobs through additional US fab construction – then they can get that in a way that does not mean much for Intel’s long-term future. If the government instead wants to ensure that a US company is capable of advanced semis manufacturing then they will need to write a large check.

Our preferred solution is for the government to instead negotiate investments by a host of potential Intel customers – Apple, Qualcomm, Nvidia, Broadcom, Google, Microsoft, etc. These companies all have a massive stake in Intel’s future. The trouble with this approach is that while all these companies have a long-term interest in securing an alternative to TSMC, short-term interest, quarterly expectations and inertia are all strong enough to make a deal unlikely. For some subset of these companies a collective $20bn is both easy to raise and an incredible bargain. Ideally, the government would ‘encourage’ these parties to see past their short-term outlook and save Intel. Unfortunately, no one seems to even be considering this approach.

«

It’s a classic game theory challenge: Apple, for example, has zero reason to give Intel any money. But if TSMC becomes a monopoly, Apple could be a loser just as easily as it is a winner. (Quite apart from any Chinese incursion.) So how much is it worth to guard against the monopoly possibility.

Now, this crossroads that Intel is at? People have been seeing it approach for a long time, as the next two links show.
unique link to this extract

---

2013: The Intel Opportunity • Stratechery

Ben Thompson, writing in 2013, when Intel had just got a new CEO:

»

Most chip designers are fabless; they create the design, then hand it off to a foundry. AMD, Nvidia, Qualcomm, MediaTek, Apple – none of them own their own factories. This certainly makes sense: manufacturing semiconductors is perhaps the most capital-intensive industry in the world, and AMD, Qualcomm, et al have been happy to focus on higher margin design work.

Much of that design work, however, has an increasingly commoditized feel to it. After all, nearly all mobile chips are centered on the ARM architecture. For the cost of a license fee, companies, such as Apple, can create their own modifications, and hire a foundry to manufacture the resultant chip. The designs are unique in small ways, but design in mobile will never be dominated by one player the way Intel dominated PCs.

It is manufacturing capability, on the other hand, that is increasingly rare, and thus, increasingly valuable. In fact, today there are only four major foundries: Samsung, GlobalFoundries, Taiwan Semiconductor Manufacturing Company, and Intel. Only four companies have the capacity to build the chips that are in every mobile device today, and in everything tomorrow.

Massive demand, limited suppliers, huge barriers to entry. It’s a good time to be a manufacturing company. It is, potentially, a good time to be Intel. After all, of those four companies, the most advanced, by a significant margin, is Intel. The only problem is that Intel sees themselves as a design company, come hell or high water.

Today Intel has once again promoted a COO to CEO. And today, once again, Intel is increasingly under duress. And, once again, the only way out may require a remaking of their identity.

It is into a climate of doom and gloom that Krzanich is taking over as CEO. And, in what will be a highly emotional yet increasingly obvious decision, he ought to commit Intel to the chip manufacturing business, i.e. manufacturing chips according to other companies’ designs.

«

It’s that last sentence: Intel should become a foundry like TSMC, he said. This was 12 years ago. Intel in the succeeding years went from strength to strength until.. it didn’t. And that’s been the case for the past two years at least. Some mistakes take a long time to become obvious – but that also makes them very hard to back out of.

However he wasn’t the first to have seen this trouble brewing…
unique link to this extract

---

When the chips are down • The Guardian

Jack Schofield, writing in, wait for it, 2009:

»

Global Foundries also has alliances with IBM – which has a $2.5bn chip plant nearby in East Fishkill, NY – and several other companies. “We don’t believe any more in a home-grown R&D model,” says a spokesman, Jon Carvill. Rather than just serving AMD, the new strategy is to target the 20 largest companies who need leading-edge chip technologies in high volumes. “There’s very little competition in that part of the market,” says Carvill. “For those customers today, there isn’t any choice: there’s only TSMC [Taiwan Semiconductor Manufacturing Company] that can meet their needs. We’re going to offer an alternative.”

Carvill is confident the “silicon foundry” approach will enable AMD to keep on competing with Intel, the world’s largest chip manufacturer, as circuitry shrinks from today’s 45 nanometres (billionths of a metre) to 22nm and beyond. (The Intel 8088 chip, used in the IBM PC in 1982, had 3-micron – 3,000nm – circuits.)

However, the latest of many predictions of the death of Moore’s law concerns the economics rather than the physics. Len Jelinek, chief analyst for semiconductor manufacturing at iSuppli, has predicted that when we reach 18nm, in 2014, the equipment will be so expensive that chip manufacturers won’t be able to recover the fab costs.

This isn’t really a new idea either. Mike Mayberry, vice-president of Intel’s research and manufacturing group, points out that Arthur Rock, one of Intel’s early venture capital investors, came up with Rock’s law – the cost of a chip fabrication plant doubles every four years.

However, unlike Moore’s law, Rock’s law has not worked out well. In an article published by the IEEE, Philip Ross argued that fabs should have cost $5bn in the late 1990s, and $10bn in 2004. Global Foundries’ new fab may sound expensive at $4.2bn, but that’s an order of magnitude less than $40bn.

Which is not to say there aren’t potential problems in the semiconductor world. Gartner Research’s vice-president, Bob Johnson, points out that apart from Intel and Samsung, who can afford to build this sort of fab for themselves, most companies are likely to move to foundries.

«

To repeat: Jack Schofield (now, sadly, deceased) wrote this in July 2009: the problems with the Intel model of “we’ll just make our chips, thanks” were already becoming visible. (Global Foundries is still going, though shrinking, but TSMC has become the 500lb gorilla of chipmaking.)
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