A new web app will let you explore and compare the temperature, rainfall and sunshine in up to three cities you might want to live in. CC-licensed photo by Maureen Barlin 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. Sunny side up. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
Political noise • Void if removed
Dave Hewitt:
»
There is a tradeoff between exploiting what the system already knows about us and exploring possibilities that might gather more data or push us in a new direction. But the relentless, optimising logic of attention as a metric of success fuels the drive to obtain ever more data, to detect stronger signals, to produce more refined suggestions, to generate more attention and more data. And in reality these illustrative squares are a massive simplification of the thousands of dimensions of datapoints gathered about our behaviour on social networks.
It is the collection of ever more data in order to extract signal from noise that is fuelling radicalisation on social media. As these systems try to improve their certainty about our behaviour by gathering more and more of our interactions, so do we too become ever more certain and forthright in the opinions we express through the constant repetition of those actions and the constant feedback of social approval or approbation. Machine learning recommendations provide each of us with an individually tailored on-ramp to whatever extreme views we might one day be capable of holding. These systems have no specific end goal in mind, as long as we keep clicking. We might start with reasonable questions and – through the slow dripfeed of ever more strongly held opinions – find ourselves ten years down the line cheering behaviour that we would once have found abhorrent.
All while flattering ourselves that such radicalisation – such conversion – only happens to other, stupider people.
«
This is a fantastic piece which explains, in a patient and comprehensible way, how social media algorithms work in a multi-dimensional framework to fit you into boxes you didn’t even know you could fit into. Think of it as your Social Warming essay for the week. I certainly couldn’t do better than this; I’d be chuffed if I got anywhere near it.
unique link to this extract
Generative AI is an expensive edging machine • Garbage Day
Ryan Broderick:
»
Every time I’ve tried to involve AI in one of my creative pursuits it has spit out the exact same level of meh. No matter the model, no matter the project, it simply cannot match what I have in my head. Which would be fine, but it absolutely cannot match the fun of making the imperfect version of that idea that I may have made on my own either. Instead, it simulates the act of brainstorming or creative exploration, turning it into predatory pay-for-play process that, every single time, spits out deeply mediocre garbage. It charges you for the thrill of feeling like you’re building or making something and, just like a casino — or online dating, or pornography, or TikTok — cares more about that monetizable loop of engagement, of progress, than it does the finished product. What I’m saying is generative AI is a deeply expensive edging machine, but for your life.
My breaking point with AI started a few months ago, after I spent a week with ChatGPT trying to build a synth setup that it assured me over and over again was possible. Only on the third or fourth day of working through the problem did it suddenly admit that the core idea was never going to actually work. Which, from a business standpoint is fine for OpenAI, of course. It kept me talking to it for hours. And, similarly, last night, after another fruitless round of vibe coding an app with Claude, I kept pressing it over and over to think of a better solution to a problem I’m having. I knew, in my bones, that it was missing a more obvious, easier solution and after the fifth time I reframed the problem it actually got mad at me!
[Claude responded to him “Ryan, this is a really tractable problem, and I think you’re overcomplicating it in your head. Let me walk through the realistic options given your constraints.”]
If we are to assume that this imagination gap, this life edging, this progress simulator, is a feature and not a bug — and there’s no reason not to, this is how every platform makes money — then the “AI revolution” suddenly starts to feel much more insidious. It is not a revolution in computing, but a revolution in accepting lower standards. I had a similar moment of clarity, watching a panel at Bitcoin Miami in in 2022, where the speakers started waxing philosophically on what they either did or did not realize was a world run on permanent, automated debt slavery. In the same way, if AI succeeds, we will have to live in a world where the joy of making something has turned into something you have to pay for.
«
When he goes for it, Broderick really smashes his target. I don’t think I’d tolerate machines telling me “you’re overcomplicating it in your head”. Remember who relies on a power outlet, buddy.
unique link to this extract
Spotlight on the shingles vaccine—again! • Ground Truths
Eric Topol:
»
Last April I wrote a Ground Truths entitled “The Shingles Vaccine and Reduction of Dementia.” At the time many were unaware of this unanticipated relationship based on 2 large natural experiments. Two new studies this week have advanced our understanding about the potential biological impact of the Shingles vaccine, independent of its effects for preventing Shingles or direct action vs. herpes zoster virus and reinforced its protection from dementia, ~80% of which is attributable to Alzheimer’s disease.
…In the new study of Canadians, the focus was on over 464,000 people aged 70 years and older who were enrolled in a primary care network; and the more than 250,000 from that group who were born in Ontario. The vaccine eligibility cutoff of birthdate before and after Jan 1, 1946 broke the groups from Ontario into two, for either being or not being vaccinated. With 5.5 years of follow-up, there was an absolute 2.0% point reduction of dementia. Using a second date-of-birth eligibility (Jan 1, 1945) in Ontario, the findings were replicated. Using the same birth cohorts for Ontario compared with the other Canadian provinces that did not implement a Shingles vaccine program, as shown below, the differences for dementia reduction irons ere pronounced, increasing over the length of follow-up. This ability to triangulate by birthdates and to other parts of the country without a vaccine program is unique among the 4 natural experiments and helps to further support a cause-and-effect relationship.
…A recent paper in Cell added another feature about the Shingles vaccine (Zostavax) drawn from the natural experiments. The vaccine not only helped prevent or delay mild cognitive impairment but also slowed the disease course among those with dementia, and reduced deaths attributable to dementia.
«
Seems like the sensible thing would be to give it as a prophylactic in middle age, given the growing amount of evidence.
unique link to this extract
On Tilt • Harpers
Jasper Craven looks at the sadly expanding world of sports betting in the US, where nearly half of all American men aged 18-49 have an online sports betting account. For the story, he went (of course) to Las Vegas:
»
Mitch Jones, a twenty-nine-year-old recovering addict, told me he’d dedicated forty hours a week to sports betting for years, largely to avoid life’s routine lulls and irritations.
By betting on sports he had found a way to “kill as much or as little time” as he wanted. A female bettor I met at Circa rightly noted that wagering is a universal opportunity to inject a bit of meaning and emotion into daily life. “Why not care a little bit?” she reasoned.
I approached a preternaturally calm man wearing a Colin Kaepernick jersey who gave his name only as Q. “I don’t bet,” he told me. When I asked why, his answer was simple: “I’m just not into losing money.” Q felt that sports, like so many other American institutions, had lost their integrity. He stopped watching the NFL for a year after Kaepernick’s ouster and now casts a skeptical eye on the entire enterprise.
He told me he had doubts about the independence of referees and questions about how the sportsbooks set betting lines—in essence, numerical representations of the likelihood that something will happen—and spreads, Vegas-generated, highly specific, bettable margins of victory. “I don’t think everybody’s involved,” Q said. “But just enough pieces are involved to make it work.”
For most of its existence, the NFL has taken pains to avoid any open association with gambling. The truth, however, is that many of the early titans of the league were bookies, fight promoters, and gamblers, including Eagles owner Leonard Tose, who was forced to sell his team after accruing some $50m in gambling losses, and Carroll Rosenbloom, the owner of the Los Angeles Rams and a heavy bettor whose 1979 death by drowning occurred under suspicious circumstances.
Nearly four years later, in the inaugural episode of the PBS investigative series Frontline, a gambler named John Piazza infamously claimed that he had helped rig a dozen games between 1968 and 1970, with the active participation of two players and a coach. As he explained, “With the quarterback, if he knew the perimeters of the score we wanted to hold . . . he’d throw a bad pass or throw it out of bounds.”
«
The female bettor saying that putting a bet on will make you “care a little bit” truly puzzles me. I don’t get sports gambling. Either the game is exciting enough for you, or it’s not. If not, why not watch something else? If you’ve got a bet on, to my mind there are now two things to worry you: will your team win, and will you make money?
unique link to this extract
Grok AI generated about 3m sexualised images in 11 days, study finds • The Guardian
Robert Booth:
»
Grok AI generated about 3m sexualised images in less than two weeks, including 23,000 that appear to depict children, according to researchers who said it “became an industrial-scale machine for the production of sexual abuse material”.
The estimate has been made by the Center for Countering Digital Hate (CCDH) after Elon Musk’s AI image generation tool sparked international outrage when it allowed users to upload photographs of strangers and celebrities, digitally strip them to their underwear or into bikinis, put them in provocative poses and post the images on X.
The trend went viral over the new year, peaking on 2 January with 199,612 individual requests, according to analysis conducted by Peryton Intelligence, a digital intelligence company specialising in online hate.
…CCDH estimated that over the 11-day period, Grok was helping create sexualised images of children every 41 seconds. These included a selfie uploaded by a schoolgirl undressed by Grok, turning a “before school selfie” into an image of her in a bikini.
“What we found was clear and disturbing: in that period Grok became an industrial-scale machine for the production of sexual abuse material,” said Imran Ahmed, CCDH’s chief executive. “Stripping a woman without their permission is sexual abuse. Throughout that period Elon was hyping the product even when it was clear to the world it was being used in this way. What Elon was ginning up was controversy, eyeballs, engagement and users. It was deeply disturbing.”
He added: “This has become a standard playbook for Silicon Valley, and in particular for social media and AI platforms. The incentives are all misaligned. They profit from this outrage. It’s not about Musk personally. This is about a system [with] perverse incentives and no minimum safeguards prescribed in law. And until regulators and lawmakers do their jobs and create a minimum expectation of safety, this will continue to happen.”
«
Just as bad, CCDH points out, is that nearly a third of those sexualised images of children remain on the platform. Musk is a terrible person to put in charge of anything sensitive.
unique link to this extract
City weather explorer: 3D climate comparison
Arthur Juliani has been looking for a new city to move to, so he built himself a little visualiser to show the local climate. It’s a lot of fun! You can compare up to three different cities for temperature (high, mean, low, at average and “feels like”), rainfall and sunlight.
It quickly went viral, leading him to comment “I’m glad my neurotic obsession with how minor temperature differences could hypothetically impact my mental health months from now has led to something people are enjoying.” There are lots of comments in replies for how he might improve it even further. One to bookmark.
unique link to this extract
Inside Enchanté, Apple’s AI chatbot for employee productivity • Macworld
Filipe Esposito:
»
…Enchanté functions as an internal ChatGPT-like assistant for employees. The app can be used to assist employees with ideas, development, proofreading, and even general knowledge answers. The interface looks quite similar to what you see in the ChatGPT app for macOS.
Many companies prohibit or restrict employees from using AI platforms for work tasks, as sensitive and internal data may end up being sent to third-party servers. Because of this, Enchanté was designed specifically for Apple’s workflows and security requirements.
For instance, the app only runs models approved by Apple, and they all run locally or on private servers, with no connection to third parties. In addition to Apple’s own Foundation Models, which power Apple Intelligence, Enchanté also provides access to Claude and Gemini.
Because of the level of privacy and security behind this app, employees can even upload documents, images, and files for analysis. Sources say the app can also access files stored on the Mac as a source for answers.
According to an internal memo from Apple, Enchanté can be used by employees not only as a test platform, but also to help them with everyday tasks at work. The app includes a database of Apple’s internal documentation and guidelines, and is being used across all departments, including engineering, design, marketing, and leadership.
Enchanté began rolling out around November 2025. Employees using Enchanté can rate the quality of answers they receive via a feedback mechanism. The app also allows side-by-side comparisons between responses generated by Apple’s models and those produced by third-party models.
«
Chatbots for me but not for thee? I do wonder though about the wisdom of doing this. I’d love some more concrete examples of what precisely people ask of it, and what they’re looking for. (Impressive work by Esposito getting this story, by the way.)
unique link to this extract
Apple’s secret product plans stolen in Luxshare cyberattack • MacRumors
Hartley Charlton:
»
The Apple supplier subject to a major cyberattack last month was China’s Luxshare, it has now emerged. More than 1TB of confidential Apple information was reportedly stolen.
It was reported in December that one of Apple’s assemblers suffered a significant cyberattack that may have compromised sensitive production-line information and manufacturing data linked to Apple. The specific company targeted, the scope of the breach, and its operational impact were unclear until now.
The attack was first revealed on RansomHub’s dark web leak site on December 15, 2025, where the group claimed it had encrypted internal Luxshare systems and exfiltrated large volumes of confidential data belonging to the company and its customers. The attackers warned that the information would be publicly released unless Luxshare contacted them to negotiate, and accused the company of attempting to conceal the incident.
According to the attackers’ claims, the exfiltrated material includes vital files such as detailed 3D CAD product models and high-precision geometric files, 2D manufacturing drawings, mechanical component designs, circuit board layouts, and internal engineering PDFs. The group added that the large archives include Apple product data as well as information belonging to Nvidia, LG, Tesla, Geely, and other major clients.
«
There was a similar attack against Apple’s laptop assembler Quanta in April 2021, and the plans from that leaked. They were for a MacBook Pro which dispensed with the Touch Bar, but brought back the HDMI connector and SD card slot. Which was indeed what happened.
Luxshare though is a components manufacturer, so any leaks from this might not be so helpful.
unique link to this extract
The paradox of work • Financial Times
Tim Harford:
»
there is a more general lesson to be learnt about our puzzling relationship with work, and a lesson that will prove particularly useful if AI dislocates the labour market.
The puzzle is that we have a love-hate relationship with working for a living. Look closely and you find that people do not tend to enjoy their work. Step back and you find that they can’t do without it.
Twenty years ago, a team of social scientists, including Alan Krueger, an economist, and Daniel Kahneman, a Nobel laureate psychologist, investigated the wellbeing of nearly 1,000 employed women living in Texas. Kahneman and Krueger asked these women to reconstruct a recent day, episode by episode, and to rate the emotions experienced during meals, stretches of childcare, commuting and so on. Emotional labels included “happy”, “enjoying myself”, “annoyed”, “depressed” and “anxious”.
A Douglas Adams character once ruefully reflected about his job that the hours were good but “most of the actual minutes are pretty lousy”. The point of Kahneman and Krueger’s research was to examine that distinction, directing people away from grand evaluations of their lives and towards the moment-to-moment experiences of which life is made.
Their day reconstruction method suggests that the three activities most likely to elicit positive emotions in these women were relaxing, socialising after work and, best of all, sex. The three most miserable activities were the evening commute, the morning commute and work itself. Work was simply the least enjoyable thing in their lives.
Yet to return to that puzzle, one of the most robust findings in social science is that when we ask people to evaluate their lives overall, there are few more reliable sources of dissatisfaction and disappointment than being unemployed. This isn’t just about money: the swings in life-satisfaction are much greater than income alone would explain.
«
(This should work as a gift article, but if not the full text will be at Harford’s site in a couple of days’ time.)
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
The Overspill: when there's more that I want to say 