A growing number of retail investors are using chatbots to pick stocks – a move that could prove costly for the unwary. CC-licensed photo by Forextime.com 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.

---

It’s Friday, so there’s another post due at the Social Warming Substack at about 0845 UK time.

---

A selection of 10 links for you. Invested. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.

---

Instagram’s teen safety features are flawed, researchers say • Reuters

Jeff Horwitz:

»

Numerous safety features that Meta has said it has implemented to protect young users on Instagram over the years do not work well or, in some cases, don’t exist, according to a report from child-safety advocacy groups that was corroborated by researchers at Northeastern University.

…Of 47 safety features tested, the groups judged only eight to be completely effective. The rest were either flawed, “no longer available or were substantially ineffective,” the report stated.

Features meant to prevent young users from surfacing self-harm-related content by blocking search terms were easily circumvented, the researchers reported. Anti-bullying message filters also failed to activate, even when prompted with the same harassing phrases Meta had used in a press release promoting them. And a feature meant to redirect teens from bingeing on self-harm-related content never triggered, the researchers found.

Researchers did find that some of the teen account safety features worked as advertised, such as a “quiet mode” meant to temporarily disable notifications at night, and a feature requiring parents to approve changes to a child’s account settings.
Titled “Teen Accounts, Broken Promises,” the report compiled and analyzed Instagram’s publicly announced updates of youth safety and well-being features going back more than a decade. Two of the groups behind the report – Molly Rose Foundation in the United Kingdom and Parents for Safe Online Spaces in the U.S. – were founded by parents who allege their children died as a result of bullying and self-harm content on the social-media company’s platforms.

The findings call into question Meta’s efforts “to protect teens from the worst parts of the platform,” said Laura Edelson, a professor at Northeastern University who oversaw a review of the findings. “Using realistic testing scenarios, we can see that many of Instagram’s safety tools simply are not working.”

Meta – which on Thursday said it was expanding teen accounts to Facebook users internationally – called the findings erroneous and misleading. “This report repeatedly misrepresents our efforts to empower parents and protect teens, misstating how our safety tools work and how millions of parents and teens are using them today,” said Meta spokesman Andy Stone. He disputed some of the report’s appraisals, calling them “dangerously misleading,” and said the company’s approach to teen account features and parental controls has changed over time.

«

It’s a constant refrain: Meta introduces safety measures for minors, they turn out not to be as safe as promised. Meta denies it. Rinse and repeat.
unique link to this extract

---

The algorithm will see you now – Works in Progress Magazine

Deena Mousa:

»

CheXNet can detect pneumonia with greater accuracy than a panel of board-certified radiologists. It is an AI model released in 2017, trained on more than 100,000 chest X-rays. It is fast, free, and can run on a single consumer-grade GPU. A hospital can use it to classify a new scan in under a second.

Since then, companies like Annalise.ai, Lunit, Aidoc, and Qure.ai have released models that can detect hundreds of diseases across multiple types of scans with greater accuracy and speed than human radiologists in benchmark tests. Some products can reorder radiologist worklists to prioritize critical cases, suggest next steps for care teams, or generate structured draft reports that fit into hospital record systems. A few, like IDx-DR, are even cleared to operate without a physician reading the image at all. In total, there are over 700 FDA-cleared radiology models, which account for more than three-quarters of all medical AI devices.

Radiology is a field optimized for human replacement, where digital inputs, pattern recognition tasks, and clear benchmarks predominate. In 2016, Geoffrey Hinton – computer scientist and Turing Award winner – declared that “people should stop training radiologists now”. If the most extreme predictions about the effect of AI on employment and wages were true, then radiology should be the canary in the coal mine. 

But demand for human labor is higher than ever. In 2025, American diagnostic radiology residency programs offered a record 1,208 positions across all radiology specialties, a 4% increase from 2024, and the field’s vacancy rates are at all-time highs. In 2025, radiology was the second-highest-paid medical specialty in the country, with an average income of $520,000, over 48% higher than the average salary in 2015.

«

So the obvious question is: why? Why are the humans thriving? The article does explain.
unique link to this extract

---

Microsoft blocks Israel’s use of its technology in mass surveillance of Palestinians • The Guardian

Harry Davies and Yuval Abraham:

»

Microsoft has terminated the Israeli military’s access to technology it used to operate a powerful surveillance system that collected millions of Palestinian civilian phone calls made each day in Gaza and the West Bank, the Guardian can reveal.

Microsoft told Israeli officials late last week that Unit 8200, the military’s elite spy agency, had violated the company’s terms of service by storing the vast trove of surveillance data in its Azure cloud platform, sources familiar with the situation said.

The decision to cut off Unit 8200’s ability to use some of its technology results directly from an investigation published by the Guardian last month. It revealed how Azure was being used to store and process the trove of Palestinian communications in a mass surveillance programme.

In a joint investigation with the Israeli-Palestinian publication +972 Magazine and the Hebrew-language outlet Local Call, the Guardian revealed how Microsoft and Unit 8200 had worked together on a plan to move large volumes of sensitive intelligence material into Azure.

The project began after a meeting in 2021 between Microsoft’s chief executive, Satya Nadella, and the unit’s then commander, Yossi Sariel. In response to the investigation, Microsoft ordered an urgent external inquiry to review its relationship with Unit 8200. Its initial findings have now led the company to cancel the unit’s access to some of its cloud storage and AI services.

Equipped with Azure’s near-limitless storage capacity and computing power, Unit 8200 had built an indiscriminate new system allowing its intelligence officers to collect, play back and analyse the content of cellular calls of an entire population.

The project was so expansive that, according to sources from Unit 8200 – which is equivalent in its remit to the US National Security Agency – a mantra emerged internally that captured its scale and ambition: “A million calls an hour.”

«

This doesn’t quite end the surveillance, though it does inconvenience the Israeli military.
unique link to this extract

---

Amazon to pay $2.5bn FTC settlement over allegations it misled Prime users • WSJ

Erin Mulvaney and Dave Michaels:

»

Amazon has reached a $2.5bn settlement with the Federal Trade Commission, just days into a civil trial examining whether the company duped customers into signing up for its signature Prime service and created a confusing process to cancel.

The e-commerce giant will pay a $1bn civil penalty, the largest in FTC history, and create a $1.5bn fund to pay back to consumers, according to court documents. It will also be required on its Prime interface to include a simple way to cancel.

Amazon agreed to settle the case without admitting or denying the FTC’s allegations that it misled customers in violation of federal consumer protection laws. The company said Thursday that the resolution allows it to move forward and focus on its business.

…“The evidence showed that Amazon used sophisticated subscription traps designed to manipulate consumers into enrolling in Prime, and then made it exceedingly hard for consumers to end their subscription,” FTC Chairman Andrew Ferguson said. “Today, we are putting billions of dollars back into Americans’ pockets, and making sure Amazon never does this again.”

The FTC has said nearly 40 million consumers were affected by Amazon’s alleged conduct.

Amazon’s Prime membership, the largest paid subscription program in the world with at least 200 million users, has helped the company become an integral part of consumers’ shopping habits. Launched in 2005, it offers free and fast shipping, access to Amazon’s streaming service and other perks. It currently costs $139 annually, or $14.99 monthly.

«

The civil trial had just started, but the FTC cut it short. There’s a strong suspicion that this was a nice number that Amazon could afford. The only benefit is that in future it will be as easy to cancel the subscription as it is to sign up. Though only in the US, of course.
unique link to this extract

---

Argentina’s AI hub dream is fading as experts move abroad • Rest of World

David Feliba:

»

Argentina is known for its beef and soy exports. Now, in the age of artificial intelligence, it is also shipping talent overseas.

Last year, Agustín Martínez, a researcher focused on AI safety, finished his doctorate program at the University of Buenos Aires. He looked for a postgraduate research position to specialize further, but couldn’t find one in the country. Martínez then applied to universities in the U.S. and Europe, and landed a position earlier this year at one of the world’s leading AI safety centres in Oxford, England.

“When I started looking into where this research was happening, I realized the only opportunities were abroad,” Martínez told Rest of World. “There isn’t a developed ecosystem around AI safety back home — we are just starting to build it.”

Martínez’s decision to leave Argentina reflects the dilemma many young scientists in the South American nation face: stay and fight an uphill battle to find jobs, support, and funding, or seek opportunities abroad, where research communities, cutting-edge infrastructure, and higher-paying salaries are more accessible. Even as President Javier Milei promises to turn Argentina into an AI powerhouse, many of the country’s best-trained engineers are quietly packing their bags.

“The prospects for scientific development in Argentina today are really bleak,” Sebastián Uchitel, who leads one of the country’s main data science centres at the University of Buenos Aires, told Rest of World. While salaries abroad can be up to ten times higher and research projects more compelling, Uchitel said the biggest problem is structural. “The entire chain for retaining talent here is broken.”

«

Like Brazil, Argentina’s future is always brighter than its present, which is also gloomier than its past.
unique link to this extract

---

Google is coming for Microsoft’s lunch with new Android PCs • Windows Central

Zac Bowden:

»

Google is coming for Microsoft’s lunch, and Qualcomm might just help them along the way. At this years Qualcomm Summit, Google’s Rick Osterloh confirmed that the company is building a unified Android platform that will run on not only smartphones, but PC form factors too: “We are building together a common technical foundation for our products on PCs and desktop computing systems,” Osterloh said.

This isn’t the first time Google has teased that such a project is in the works. In fact, it has mentioned several times now that it has plans to bring Android to PCs in an attempt to seriously compete against Windows and macOS with a real desktop-class operating system. Google is planning to bring the full Android AI stack, along with Android apps and developer community to these new Android PCs.

Qualcomm CEO Cristiano Amon says the project “delivers on the vision of convergence of mobile and PC,” which is something Microsoft attempted to do with Windows a decade ago with Windows 10, Continuum, and eventually Windows Core OS. Unfortunately, Microsoft pulled the plug on its converged operating system before it had the chance to take off. Now, it looks like Google wants to bring to life what Microsoft failed to.

«

Woo-hoo! Here comes Google! Take that, slovenly Microsoft! What do you know about operating systems?

»

Both Apple and Google have slowly been converging their mobile platforms with the PC form factor. iPadOS is Apple’s attempt at scaling up iOS to become more desktop-like, and now Google is doing the same with Android. With the foundations it has already laid with Chrome OS, the relationships it has built with PC makers, and the thousands of Android apps that are already available, Google is in a pretty good spot to make a good go at this.

Of course, Chrome OS never really took off in the way Google was hoping it would, and so there’s no guarantee Android-powered PCs will be any different.

«

Oh. So.. a desktop-class operating system? Wouldn’t that need desktop-class apps? Microsoft abandoned convergence because it didn’t have a comprehensible mobile story. Apple, by contrast, has managed the shift from desktop to mobile through enormous effort over two decades. The Android stack isn’t a desktop stack.
unique link to this extract

---

‘ChatGPT, what stocks should I buy?’ AI fuels boom in robo-advisory market • Reuters

Joice Alves:

»

As ChatGPT nears its third birthday, at least one in 10 retail investors is using a chatbot to pick stocks, fuelling a boom in the robo-advisory market, but even fans say it is a high-risk strategy that cannot replace traditional advisors just yet.

Thanks to artificial intelligence, anyone can select stocks, monitor them and obtain investment analysis that was once only available to big banks or institutional investors.

The robo-advisory market – which includes all companies providing automated, algorithm-driven financial advice such as fintech, banks and wealth managers – is forecast to grow to $470.91bn in revenues in 2029 from $61.75bn last year, marking a roughly 600% increase, according to data analysis firm Research and Markets.

Jeremy Leung, who spent almost two decades analysing companies for UBS, has been using ChatGPT to chase stocks for his multi-asset portfolio since he left the Swiss bank late last year.

“I no longer have the luxury of a Bloomberg (terminal), or those kinds of market-data services which are very, very expensive,” Leung said. “Even the simple ChatGPT tool can do a lot and replicate a lot of the workflows that I used to do,” he said, cautioning that such a tool might however miss some crucial analyses as it can’t access data behind a paywall.

Leung isn’t alone. The industry is growing fast and exponentially. About half of retail investors say they would use AI tools such as ChatGPT, whose launch in November 2022 ignited the AI boom on the markets, or Google’s Gemini to pick or alter investments in their portfolio, and 13% of them already use these tools, according to a survey from broker eToro, which polled 11,000 retail investors across the world.

In the UK, 40% of the respondents to a survey by comparative company Finder said they have used chatbots and AI for personal finance advice.

ChatGPT itself warns it should not be relied on for professional financial advice and says its owner OpenAI has not released data on the number of people who use its chatbot to choose investments. “AI models can be brilliant,” said Dan Moczulski, UK managing director at eToro, which boasts 30 million users worldwide. “The risk comes when people treat generic models like ChatGPT or Gemini as crystal balls.”

«

Yet another space where chatbots are taking over people’s intelligence.
unique link to this extract

---

The strange and hilarious history of the word “OK” • History Defined

Carl Seaver:

»

Journalist Charles Gordon Greene was responsible for the first confirmed use of the word OK in the March 23, 1839 issue of the Boston Morning Post.It was found in a humorous article about their rival paper, the Providence Journal.

There are a few theories about the origins of OK, and some of them make perfect sense. 

One such theory is that OK stems from the Scottish phrase of agreement, “och aye”. After all, OK also conveys agreement or acquiescence, and both och aye and OK sound similar. Surely that’s the most sensical answer, right? Well, it might be sensical, but it’s wrong. 

The next theory is that OK comes from the Choctaw word “okeh”, which translates to “it is so”. Again, both words sound phonetically similar, and they both convey similar ideas. This explanation also makes sense! But it’s still wrong. 

When OK was used in the Boston Morning Post article, it was an abbreviation of “Oll Korrect”, or “Ole Kurreck”, a humorous misspelling of “All Correct”. This misspelling and subsequent abbreviation was quite popular, and readers of the time found it hilarious.

But why was spelling something wrong so funny? 

«

Not entirely sure about “hilarious”. Turns out it was an electoral ploy which went viral.
unique link to this extract

---

Apple responds to iPhone 17 Pro scratch and durability concerns • 9to5Mac

Chance Miller:

»

It wouldn’t be a new iPhone launch without some drama. While the iPhone Air has impressed in all bend and scratch tests, some users have raised concerns over the durability of the iPhone 17 Pro’s aluminum body.

I spoke to Apple about these concerns, and here’s what it said.

The “scratchgate” narrative first materialized in a Bloomberg story published on iPhone 17 launch day. The article highlighted several examples of wear and tear on iPhone 17 Pro demo units in Apple Stores and other retail partners. The marks were mostly contained to the back of the iPhone 17 Pro, especially around the MagSafe cutout.

Apple tells me it has determined these imperfections are caused by worn MagSafe stands used in some stores. It also clarifies that the marks aren’t scratches, but rather material transfer from the stand to the phone that is removable with cleaning. The company says it is working to address these problems at the stores, presumably by replacing the worn MagSafe stands. Other iPhones on display are also affected by this, including iPhone 16 models.

JerryRigEverything highlighted another durability concern in a video over the weekend. In his testing, he found that the raised edges around the camera plateau on the back of the iPhone 17 Pro are particularly susceptible to scratches. He explained that this is largely because Apple didn’t add a chamfer, fillet, or radius around the camera plateau.

Apple tells me that iPhone 17 Pro’s camera plateau edges have similar characteristics to the edges of the anodized aluminum cases on other Apple products, including other iPhone models and MacBooks. While those edges are durable and undergo Apple’s rigorous testing, the company says users may see normal wear and tear, including small abrasions, over time.

«

We go from scratching (iPod nano in 2005) to bending (iPhone 6 in 2014) and back to scratching. It’s quite the cycle.
unique link to this extract

---

iPadOS 26 review: A computer? • Six Colours

Jason Snell:

»

Another huge feature that Mac users take for granted, but wasn’t really a part of iPadOS before, is the ability to open files in specific apps and to set default apps for file types. To assign all your files to a specific app, just select one, choose Get Info, and choose a default from that panel. This, combined with the arrival of the Preview app on the iPad, has really changed how I work. It’s so much easier to double-click on a PDF and then do the same with a Markdown file and get to work, making the “classic” oblique app-centric iPad approach — launch each app, navigate through a bespoke file picker, repeat — feel archaic.

There’s also a major improvement when it comes to long file copies, especially ones happening across a network: they generate a progress window that can be made into a Live Activity, allowing you to leave Files while you keep tabs on the progress of the operation.

This is part of a larger upgrade to iPadOS that allows apps that perform lengthy, finite tasks to do so in the background. Previously, apps with lengthy exports—Final Cut Pro, Logic Pro, and Ferrite Recording Studio are good examples—had to be kept in front until they were done with their jobs. That sort of single-mindedness would never fly on the Mac, where you can always switch to another app while you’re doing an export (or file copy). That ends in iPadOS 26, and the first time I exported a podcast edit from a version of Ferrite that supported this feature was almost magical. What do you mean, I can do more work while the export is happening? It’s so delightful for something so commonplace and frictionless on the Mac to reach that same level of frictionlessness on the iPad.

«

It seems that iPadOS has essentially turned into macOS Lite, but with useful additions. Given the chips that power them, it was ridiculous to hold them back with the old version. Though I still haven’t upgraded, personally – I like the Slide Over function of the previous version, which this completely abandons.
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