A new study found that endoscopists who used AI were less good at diagnosis afterwards without it – a “deskilling” effect. CC-licensed photo by Rosen & Meseguer Atlas of Medical Foreign Bodies 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. Machine-guided. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.

---

AI is a mass-delusion event • The Atlantic

Charlie Warzel:

»

It is a Monday afternoon in August, and I am on the internet watching a former cable-news anchor interview a dead teenager on Substack. This dead teenager—Joaquin Oliver, killed in the mass shooting at Marjory Stoneman Douglas High School, in Parkland, Florida—has been reanimated by generative AI, his voice and dialogue modelled on snippets of his writing and home-video footage. The animations are stiff, the model’s speaking cadence is too fast, and in two instances, when it is trying to convey excitement, its pitch rises rapidly, producing a digital shriek. How many people, I wonder, had to agree that this was a good idea to get us to this moment? I feel like I’m losing my mind watching it.

Jim Acosta, the former CNN personality who’s conducting the interview, appears fully bought-in to the premise, adding to the surreality: he’s playing it straight, even though the interactions are so bizarre. Acosta asks simple questions about Oliver’s interests and how the teenager died. The chatbot, which was built with the full cooperation of Oliver’s parents to advocate for gun control, responds like a press release: “We need to create safe spaces for conversations and connections, making sure everyone feels seen.” It offers bromides such as “More kindness and understanding can truly make a difference.”

…The interview triggered a feeling that has become exceedingly familiar over the past three years. It is the sinking feeling of a societal race toward a future that feels bloodless, hastily conceived, and shruggingly accepted. Are we really doing this? Who thought this was a good idea? In this sense, the Acosta interview is just a product of what feels like a collective delusion. This strange brew of shock, confusion, and ambivalence, I’ve realized, is the defining emotion of the generative-AI era. Three years into the hype, it seems that one of AI’s enduring cultural impacts is to make people feel like they’re losing it.

«

unique link to this extract

---

Why Apple Vision Pro can’t win without immersive video • Mac Observer

Rajat Saini:

»

When the Vision Pro launched, Apple highlighted immersive video as the device’s strongest selling point. The format offers a level of immersion that no traditional screen can match. Watching sports, concerts, or documentaries in this environment can feel transformative. Yet Apple has released only 27 pieces of content in the format since launch, leaving users with little to watch.

For example, Apple continues to promote an immersive highlight reel of the 2024 NBA All-Star Game, despite the 2025 event having taken place six months ago without an immersive version. The same applies to concerts. While users can access shows from Metallica, Bono, and a short music video by The Weeknd, the library lacks depth and variety to keep younger tech audiences engaged.

Apple’s series offerings remain sparse. Wild Life has four episodes, Elevated just one, Boundless two, and Prehistoric Planet only two. Even Adventure, with its extreme sports footage, offers only five installments. For a product pitched as a new way to experience storytelling, the content feels more like a demo reel than a growing library.

According to Bloomberg’s Mark Gurman, Apple intentionally slow-walked immersive content to preserve its reserve, as each production is expensive and resource-heavy. But this strategy has created a catch-22: immersive video is the feature that sells the Vision Pro, yet the lack of it makes the device harder to justify.

Apple has updated the Vision Pro’s software and plans a faster chip in the next model, but these changes won’t address the fundamental issue. A cheaper and lighter version is reportedly set for 2027, but two years is a long wait in a fast-moving industry. If immersive content remains scarce until then, the product risks fading into irrelevance.

«

“Risks fading”? I’ll say it: the Vision Pro is completely forgotten by everyone who doesn’t own one, and probably by quite a few who do. The lack of content; the price; the terrible ergonomics and battery life; the weird design decisions (fake eyes on the outside? Why??). It’s been a bad idea from start to finish, and the lack of a sensible strategy around immersive video – the real selling point – has been the garnish on the awful salad.
unique link to this extract

---

Trump administration in talks to take 10% stake in Intel, Bloomberg News reports • Reuters

Zaheer Kachwala and Jaspreet Singh:

»

The Trump administration is in talks to take a 10% stake in Intel by converting some or all of the struggling company’s Chips Act grants into equity, Bloomberg News reported, citing a White House official and other people familiar with the matter.

Shares of Intel closed about 3.7% lower on Monday, after rallying last week on hopes of US federal support.

A 10% stake in the American chipmaker would be worth about $10bn. Intel has been slated to receive a combined $10.9bn in Chips Act grants for commercial and military production, and the figure is roughly enough to pay for the government’s holding, according to the Bloomberg report on Monday.

Intel declined to comment on the report, while the White House did not respond to a request for comment. Reuters could not immediately verify the report.

Media reports said last week that the US government may buy a stake in Intel, after a meeting between CEO Lip-Bu Tan and President Donald Trump that was sparked by Trump’s demand for the new Intel chief’s resignation over his ties to Chinese firms.

Federal backing could give Intel more breathing room to revive its loss-making foundry business, analysts have said, but it still suffers from a weak product roadmap and challenges in attracting customers to its new factories.

“The fact that the U.S. government is stepping in to save a blue-chip American company likely means that Intel’s competitive position was much worse than what anybody feared,” said David Wagner, head of equity and portfolio manager at Intel shareholder Aptus Capital Advisors.

«

What fun to watch America trying state-owned capitalism. Not sure it’s going to pull Intel out of its nosedive, but certainly adds a bit of spice to the process.
unique link to this extract

---

Indeed recruiter text scam: I responded to one of the “job” messages. It got weird quickly • Slate

Alexander Sammon:

»

On WhatsApp, I met Cathy, my “coach,” from a company she referred to as Interleave. She had gotten my number, she said, from “Elena who works in Indeed Recruitment Department” and was eager to work with me. Her number had a 424 area code, or Los Angeles. (Indeed is its own company—it essentially offers a job board—and I have no reason to believe that it was actually involved at all.)

Cathy was not altogether patient. When I didn’t respond within two hours, she sent me a voice note that sounded sort of humanoid: “Hello, are you still there?” The next day, she called me and I missed it. When I did respond, via WhatsApp message, she was curt: “Hello, you finally replied to my message. I thought you were taken away by aliens.”

I was going to be doing “music promotion,” I was told. It would take just one or two hours a day. “We use an A.I.–powered system developed by Interleave to help increase the play count of music singles and albums,” Cathy told me, on platforms like YouTube and TikTok. In effect, we were going to be boosting play counts: “Artificial intelligence cannot do this, only real people can participate,” she said. “All we need to do is create a personal account on the Interleave platform, use our real information, and create real playback records.”

Like so many middle school girlfriends, Interleave was based in Canada. The compensation was similarly sketchy. I’d get $100 for two days of work. For 30 days, I’d get $8,200, though it would all have to be routed through a crypto wallet. The job and the compensation had nothing to do with the original text I’d received, but no matter.

Besides, this wasn’t just crude self-enrichment, Cathy said. Interleave was going to “donate a portion of its profits to the World Food Programme charity to help those who really need help gain a brighter life.”

«

You already have a feeling for where it’s going to go – there will be money “paid” to him which somehow can’t be withdrawn without paying more money to the “employer”, but then for some reason the “salary” still can’t be withdrawn, for ever and ever – and all that varies between these scams is what the Macguffin is. (In this case, some sort of music promotion.)
unique link to this extract

---

Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study • The Lancet Gastroenterology & Hepatology

Krzysztof Budzyń, MD et al:

»

We conducted a retrospective, observational study at four endoscopy centres in Poland taking part in the ACCEPT (Artificial Intelligence in Colonoscopy for Cancer Prevention) trial. These centres introduced AI tools for polyp detection at the end of 2021, after which colonoscopies had been randomly assigned to be conducted with or without AI assistance according to the date of examination.

We evaluated the quality of colonoscopy by comparing two different phases: three months before and three months after AI implementation. We included all diagnostic colonoscopies, excluding those involving intensive anticoagulant use, pregnancy, or a history of colorectal resection or inflammatory bowel disease. The primary outcome was change in adenoma detection rate (ADR) of standard, non-AI assisted colonoscopy before and after AI exposure.

…Interpretation: continuous exposure to AI might reduce the ADR of standard non-AI assisted colonoscopy, suggesting a negative effect on endoscopist behaviour.

«

I’ve only got access to this summary, so it’s not clear whether the AI-assisted rate of detection was better or worse than without; only that after people used AI and then reverted to not, their detection was worse. (If anyone can enlighten, that would be welcome.)

The Science Media Centre has a number of comments from scientists: one points out that the number of colonoscopies performed nearly doubled after the AI tool was introduced. Another points out that this paper suggests there would be a problem if a cyber attack (or similar) took out the AI assistant.
unique link to this extract

---

Big Tech expands AI for Good in Africa amid skepticism • Rest of World

Damilare Dosunmu:

»

in July, Google opened an AI Community Center in Ghana to support local innovation, and announced a $37m investment in social impact projects under the catch-all label “AI for Good” across Africa. 

Microsoft, Meta, Amazon, and Apple are funding similar projects, aligning themselves with a growing trend in which governments, global organizations, and private tech companies promote AI for public good — a push that aims to normalize the technology and soften the anxieties surrounding it. 

Regional AI experts and industry leaders, however, urge caution, saying Africa could become a testing ground for AI models and a vast source of data collection in the U.S.-China tech rivalry, making the continent dependent on foreign-owned systems.

“AI for Good is still very much embedded and rooted in the same saviorism from the West towards the global south. Africa needs to start building reliable infrastructures that can power all these systems,” Asma Derja, founder of Ethical AI Alliance, a Spain-based advocacy group for safe AI, told Rest of World. “Otherwise, Big Tech will continue to make money off the region and then take a [corporate social responsibility] budget to finance a few projects that are addressing climate change or, you know, a particular topic in Tanzania or in Mongolia and call it AI for Good.”

«

“AI for Good” is quite a slogan. Definitely uplifting, something for people to follow – a bit like, say, “Don’t be evil”.
unique link to this extract

---

Chinese homebuyers snub government incentives as they bet on further price falls • South China Morning Post

Daniel Ren and Yuke Xie:

»

Homebuyers in mainland China’s major cities are responding tepidly to new policy relaxations, anticipating further price drops amid an uncertain outlook for the sector.
Analysts and potential buyers warn that the bearish sentiments might weaken the effectiveness of the stimulus measures aimed at reviving the stagnant property market

“The market sentiment is terribly weak now, and the consensus among potential buyers is that prices will slide in the coming months if homeowners are eager to get deals done,” said Qiu Lixiao, a property agent with Pacific Rehouse in Shanghai. “Any measure to encourage purchases of new and pre-owned flats may turn out to be a damp squib.”

He added that his agency had hardly received any inquiries since Beijing’s surprise move last week to lift home-purchase restrictions in the city’s outlying areas, which had also heightened expectations that authorities in Shanghai would take similar action.

…Over the past decade, the mainland’s biggest cities like Shanghai and Beijing have taken many steps to curb home purchases to rein in what was once a red-hot property market.

Lindsay Zhang, a Beijing-based finance professional, said the policy change was not enough to inspire confidence to step into the market.

“I don’t think the relaxed rules are very meaningful because there is not much room for value appreciation for most homes,” she said. “Only properties in the very core locations can retain their value, or at least are more immune to a citywide [price] decline.”

«

Price deflation in a housing market? Almost unknown when there isn’t a recession happening. But China would never admit that it’s in recession.
unique link to this extract

---

MIT report: 95% of generative AI pilots at companies are failing • Fortune

Sheryl Estrada:

»

The GenAI Divide: State of AI in Business 2025, a new report published by MIT’s NANDA initiative, reveals that while generative AI holds promise for enterprises, most initiatives to drive rapid revenue growth are falling flat.

Despite the rush to integrate powerful new models, about 5% of AI pilot programs achieve rapid revenue acceleration; the vast majority stall, delivering little to no measurable impact on P&L. The research—based on 150 interviews with leaders, a survey of 350 employees, and an analysis of 300 public AI deployments—paints a clear divide between success stories and stalled projects.

To unpack these findings, I spoke with Aditya Challapally, the lead author of the report, and a research contributor to project NANDA at MIT.

“Some large companies’ pilots and younger startups are really excelling with generative AI,” Challapally said. Startups led by 19- or 20-year-olds, for example, “have seen revenues jump from zero to $20 million in a year,” he said. “It’s because they pick one pain point, execute well, and partner smartly with companies who use their tools,” he added.

But for 95% of companies in the dataset, generative AI implementation is falling short. The core issue? Not the quality of the AI models, but the “learning gap” for both tools and organizations. While executives often blame regulation or model performance, MIT’s research points to flawed enterprise integration. Generic tools like ChatGPT excel for individuals because of their flexibility, but they stall in enterprise use since they don’t learn from or adapt to workflows, Challapally explained.

The data also reveals a misalignment in resource allocation. More than half of generative AI budgets are devoted to sales and marketing tools, yet MIT found the biggest ROI in back-office automation—eliminating business process outsourcing, cutting external agency costs, and streamlining operations.

«

I was musing, as I saw various people on social media touting how wonderful ChatGPT is for them, why I haven’t noticed any evidence of their productivity being vastly higher. I suspect we’ll need to wait a few years to see it being used even slightly effectively.
unique link to this extract

---

HR giant Workday discloses data breach after Salesforce attack • Bleeping Computer

Sergiu Gatlan:

»

Human resources giant Workday has disclosed a data breach after attackers gained access to a third-party customer relationship management (CRM) platform in a recent social engineering attack.

Headquartered in Pleasanton, California, Workday has over 19,300 employees in offices across North America, EMEA, and APJ. Workday’s customer list comprises over 11,000 organizations across a diverse range of industries, including more than 60% of the Fortune 500 companies.

As the company revealed in a Friday blog, the attackers gained access to some of the information stored on the compromised CRM systems, adding that no customer tenants were impacted.

“We want to let you know about a recent social engineering campaign targeting many large organizations, including Workday,” the HR giant said. “We recently identified that Workday had been targeted and threat actors were able to access some information from our third-party CRM platform. There is no indication of access to customer tenants or the data within them.”

However, some business contact information was exposed in the incident, including customer data that could be used in subsequent attacks.

“The type of information the actor obtained was primarily commonly available business contact information, like names, email addresses, and phone numbers, potentially to further their social engineering scams,” it added.

«

Just the normal stuff that hackers love to get hold of, no big deal, don’t worry about the call apparently from one of those people you know.
unique link to this extract

---

Government-built “Humphrey” AI tool reviews responses to consultation for first time, in bid to save millions • GOV.UK

»

A new AI tool has summarised what the public have told the government in response to a consultation for the first time – providing nearly identical results to officials.

The tool, called ‘Consult’, was first used on a live consultation by the Scottish Government when it was seeking views on how to regulate non-surgical cosmetic procedures – like lip fillers and laser hair removal – as use of the treatments has risen.

The tool now set to be used across departments in a bid to cut down the millions of pounds spent on the current process, which often includes outsourcing analysis to expensive contractors – helping to build a productive and agile state to deliver the Plan for Change.

Reviewing comments from over 2,000 consultation responses using generative AI, Consult identified key themes that feedback fell into across each of six qualitative questions. These themes were checked and refined by experts in the Scottish Government, the AI tool then sorted individual responses into themes and gave officials more time to delve into the detail and evaluate the policy implications of feedback received.

As this was the first time Consult was used on a live consultation, experts at the Scottish Government manually reviewed every response too. Identifying what an individual response is saying, and putting it in a ‘theme’ is subjective, humans don’t always agree. When we compare Consult to the human reviewer, we see they agree the majority of the time – with differences in view having a negligible impact on how themes were ranked overall.

‘Consult’ is part of ‘Humphrey’, a bundle of AI tools designed to speed up the work of civil servants and cut back time spent on admin, and money spent on contractors.

«

Calling it “Humphrey” is a joke about the old Yes Minister British TV comedy series, where the senior civil servant urbanely guiding the minister away from pitfalls was called Sir Humphrey. Given that people are going to be offering AI-written responses, might as well fight fire with fire.
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