Dear Friend,
This week Bill Gates published a longer think piece on his position on AI. As was to be expected, it was (and continues to be) hotly debated – especially as Gates raises some strong concerns about the downsides of the laissez-faire approach to regulating AI we see around the world. I encourage you to read it – Gates surely is a smart person, and his views are strong, but well articulated and reasoned. Give it a read, and if you like, hit reply and let me know what you think.
And now, this…
Headlines from the Future
Screen Time and Kids? It’s Complicated. The common narrative these days is that screen time is (universally) bad for kids. So bad, that we (increasingly) ban phones from schools in countries around the world.
Low levels of physical activity among adolescents remain a significant public health concern. Sedentary behavior has been associated with poorer academic achievement, while physical activity is known to support brain function, particularly in adults.
But, as so often, the real world is more complicated than that. A new study came to very different conclusion:
“The findings suggest that screen time can support children’s and adolescents’ cognitive processing. Presumably, the essential point here is what kind of things they do in their screen time. Teachers and parents should encourage children to use devices and screens in such ways that promote active thinking, problem-solving, creativity and learning,” states Doctoral Researcher Petri Jalanko from the University of Jyväskylä.
And, of course, screen time is not screen time – the type of activity does matter.
“We should not regard screen time solely as harmful but seek balance between physical activity and screen time that promotes active thinking,” Jalanko summarizes.
And just to keep things in perspective – here is the counter-argument (another new study which was just published).
━━━━━
Reuters’ Deep Dive into How Meta’s AI-native Plan Imploded: Reuters’ reporting into the details of the company’s plans to transform itself into an AI-native organization, cutting some teams by as much as 60%, is a good and detailed read on why, despite all the hopes, wishes, and continued hype, we are still in the very first inning of the AI “revolution”: Meta’s engineers wrote far more code and shipped no more product.
The Agents Never Sleep, So Neither Do They: AI agents work around the clock – and regularly require human steering, oversight, or a human sign-off. All of this creates a whole new type of burnout. Startup founders report interrupted sleep schedules, the constant pressure of being on call for their agents (who don’t have any of those pesky human needs), and “agent FOMO.” In the words of Peter Pezaris (four prior exits, who promised his wife he’d retire this year and instead founded Proxon – a company that makes software for managing AI agent workforces): “Every minute that I’m not working, I’m missing out on doing a week’s worth of work.”
Trackers Aren’t Tracking: Ever since the first Garmin watches, and certainly since Apple introduced the Apple Watch, we (collectively) have become obsessed with tracking our health stats. Oura, the company behind the health-tracking ring, now faces a class-action lawsuit over its claims that the ring can accurately track your sleep. Note: Many or most consumer electronics are not medical-grade – regardless of what they claim.
TTT – TikTok Travel: European travel group TUI conducted a survey and found that four in ten travelers booked their trip after seeing some viral content on social media – and an equal number of people feel pressured to post about their trip and make it look exciting (you know the type – miserable faces throughout, but happy smiles when the selfie is taken). And, of course, there is a whole counterculture where people share carefully curated trip suggestions on private Google Docs.
Better Grades, Lower Exam Scores: The impact of AI on education is a hot topic (to say the least). A new study from China, looking at 27,000 kids, found that although AI increased the average homework scores by a significant amount (while also reducing the time spent on it: scores up 18% in 45 fewer minutes), it had the inverse effect on exam scores (down by 20%). But, of course, it’s more complicated than that; the difference is entirely explained by whether kids kept doing the work slowly or just handed in the AI-generated answers.
Open AI (not OpenAI) Models Are Catching Up – Fast: Open-weight AI models, such as the recently released GLM 4.0 and Kimi k1.5, can be hosted on your own servers and are often available through model providers such as OpenRouter for significantly lower costs than their closed counterparts from companies such as OpenAI, Anthropic, and Google. The usual argument against using them is that they lack capabilities when compared to the latest frontier models. But the gap is closing fast – a recent analysis and comparison found that “with each generation, open-source models take half as long to catch up to the first closed-source model of the era.”
AI Eats The Turks: Twenty-one years ago, Amazon launched Mechanical Turk – a site that allowed you to hire people to do menial tasks for pennies. I remember running user experience research tasks on there by having people click on our prototype interfaces. It’s over soon, as Amazon said it will shut the service down next month – AI has taken over many of the work people previously farmed out to Mechanical Turkers.
Robot Surgery Is More Precise – And No Better: Robot-assisted surgery (in this case knee surgery) has long been touted as more precise and without the messiness of unsteady human hands doing fine motor-skilled work inside your body. The RACER-Knee trial, published in The Lancet, found that robot-assisted knee surgery is, indeed, more precise – alas, when it comes to the one thing you should care about, it’s not better. Pain, mobility and recovery came out the same at twelve months – while the robotic operations ran about ten minutes longer and cost around £950 more each.
Learn Claude: Anthropic launched its Claude Academy, a neat collection of learning resources for all things Claude (and AI at large). It’s comprehensive (350+ courses), free, and well done.
What We Are Reading
The False Alignment Trap Most leadership teams think they’re aligned because nobody disagreed, but that silence is exactly what kills the opportunity for change. @Mafe
The Year Every Buyer Tries to Build Their Own Software AI didn’t kill SaaS. It just made every vendor compete with their customer’s weekend project, and every customer underestimate the long-term management of their vibe-coded solutions. @Kacee
How Credit Card Rewards Became a $9.2 Billion Wealth Transfer I do love to geek out on some obscure economic insights. This one is a fascinating look into credit card reward programs, the cost of collecting money from your customers, what that does to your pricing and how those price increases are disproportionately paid for by people who have to pay cash or with debit cards. @Pascal
↳ And now for some fun…
🤘 I guess it doesn’t come as a total surprise that metalheads (folks who like to listen to heavy metal music) are special: Metalheads can endure pain longer thanks to the festival experience.
🤖 Vintage Artificial Intelligence: Before It Got Awkward. A lovely collection of very early AI software, hosted by the wonderful folks at the Internet Archive.
📡 Another good reminder that we truly live in the future: New Worlds.
What We Do When We’re Not Writing
Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.
↗ Dive into the deep end: Access our complete collection of 3,000+ radical links.