Dear Friend,
The breathless (and largely mindless) reporting on all things tech, especially AI, is truly getting old. From a press which defaults to printing Elon Musk’s timeline fantasies as news, to reporters quoting Dwarkesh Patel’s rather unhinged summary of the Hugging Face incident as fact – the lack of critical thinking and basic fact-checking is astounding. Please do call us out on this if you ever see a whiff of it sneaking into your radical Briefing.
And now, this…
Headlines from the Future
Cybersecurity in the Age of AI: Anil Madhavapeddy is a professor of Planetary Computing at the University of Cambridge and one of the maintainers of the OCaml programming language. In a recent blog post he sounded the alarm about a worrying trend in cybersecurity, now that your adversaries have access to increasingly powerful AI models and tooling: Within minutes of a bug disclosure, attackers are deploying AI-generated exploits:
This normally takes a few days and a release within a week or two is reasonable. Within about ten minutes (!) this website was fielding probes for percent-encoded traversal sequences, indicating that automated watchers are keeping an eye on public repositories.
AI expert Simon Willison points out:
Modern coding agents have become so effective at finding flaws that the slightest hint at a new bug can be enough information for them to find it, something Anil has been able to demonstrate using his own agents, switching to DeepSeek V4 Pro when Claude Fable refused the task. […] Anil points out that this rate of discovery appears incompatible with existing open source embargo practices for new issues. If an issue can become an exploit this fast, we need to figure out new processes for keeping our communities safe.
Prepare yourself for a rapidly increasing number of AI-generated exploits, and if you are in IT, the necessity to apply patches within minutes of disclosure. Personally, I think the vast majority of companies are not ready for this.
━━━━━
The Data Center Reality Check: People love to hate on data centers. They are big, ugly, don’t create many jobs, and gobble up enormous amounts of energy and water. But what does this look like? The Verge went to Loudoun County, Virginia, to find out – here, the future arrived twenty years ago.
AI Societies: MIT’s Markus Buehler and team dropped up to 200 LLM agents into a shared simulated world without any assigned roles and just let them explore, gather resources, and build. What emerged was a technological society: agents spontaneously split into roles, with about a quarter of the population clustering around constructing and maintaining shared artifacts that others could then inherit, fork, and improve. It’s the closest we’ve seen (so far) to a controlled experiment in what an “AI civilization” might actually look like.
AI Sameness: By now we all know that AI slop has a unique smell – all those lovely emdashes, the “it’s x, not y” tone, the short, breathless sentences. AI researcher Louis Abraham has analyzed 461,000+ GitHub pull request descriptions and found a clear pattern in the specific words Anthropic’s Claude AI uses. I bet you money you have seen these before – they are plainly load-bearing.
AI for Everyone: South Korea plans to make AI chatbots available to all citizens – turning AI into a public utility. Even more remarkable is the fact that South Korea will use mostly homegrown AI models for this – a strong move toward AI sovereignty (and possibly a way to wean South Koreans from the dominant US models).
Drone Deliveries Are a Dud: I’ve been saying this for years – I just can’t see drone deliveries taking off in urban areas. The Amazon-and-Walmart fantasy of me barbecuing in my backyard, running out of ketchup, and having a drone drop off a fresh bottle is something I never really understood – imagine living in the nightmarish dystopia where drones fly over your house all the time. Now Amazon’s drones are at it again: They dropped parcels straight into the swimming pools of their customers.
Homework Post-Homework: In a world where AI can credibly do nearly all of a student’s undergraduate homework, the question becomes: what is homework good for and how do we measure our students’ learning? MIT, being at the forefront of this, is considering changing course and moving away from take-home work to oral exams and portfolios – something that previews the future of education. And as correct a move this might be, I doubt it will scale well in an already overloaded education system.
The Biggest AI Haters in America: A recent study by Glassdoor found that the number one AI hater is your friendly neighborhood insurance claims adjuster. Ninety-eight percent of them were unhappy with AI: “AI-obsessed leaders forcing error-prone AI on them and their clients,” according to the study. Leadership in insurance companies is doubling down on their use of AI; employees are dissatisfied and leave.
What We Are Reading
The Singularity Is Not What It Seems We built a machine we can’t control, and then we hired another machine to figure out why. @Jane
Why AI Detectors Can’t Solve The Problem They Were Built For AI-tells like em dashes are becoming unreliable proxies for detecting AI writing. The real problem isn’t detection at all; it’s that readers use perceived AI use as a shortcut to judge credibility, regardless of whether the content is actually good. @Mafe
Your Perception Of Time Is, In Fact, Warped It’s not just you. Now here’s why it’s happening. @Jeffrey
Algorithms Trap Us in the Familiar. Can They Also Spark Breakthroughs? The AI paradox in a nutshell: we’ve given everyone access to more knowledge, and somehow risk sending everyone toward the same answer. Efficiency is easy; designing for productive weirdness is where breakthroughs live. @Kacee
The Forgotten History of Small Nuclear Reactors Small nuclear reactors (SMRs) have been proposed as a potential solution to our energy needs – sadly, due to their mental proximity to nuclear disasters such as Chernobyl and Three Mile Island, they never had a chance to prove their ground. Here is their long history. @Pascal
↳ And now for some fun…
🤖 When I grew up, we had Lego and Fischertechnik kits. Now we have Microduck, the adorable, open-source biped robot. Makes me want to be a kid again and train this thing to fetch my coffee.
🈺 This makes me want to change both the name of my company, as well as my title: A selection of Los Alamos rolodex business cards.
🚶➡️ Ever thought about hiking the Appalachian Trail? Here is a gorgeous trail journal made in the style of the cult-classic computer game from yesteryear, The Oregon Trail.
📽️ If you ever miss that feeling of wandering through your local Blockbuster – now you can do the same with your personal video collection in a 3D store right in your web browser.
🗺️ Talking about movies – ever wondered where a particular movie was filmed, or which movies were filmed in your city? Movie Scene Map puts movie locations on a map.
🚽 The Internet is a weird place. Point in case: The Restroom Archive is an ongoing repository of publicly accessible restrooms.
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.