Dear Friend,
Over the last couple of months, I have talked to quite a few of you (the readers of our dispatches) and heard multiple times that you’d prefer to have some more context around the links we share in the rabbit hole section of this Briefing. Happy to oblige. Take a look and let me know what you think. Better, worse, too long, or still too short?
And in other news, I am prepping the launch of Season 3 of the Built for Turbulence podcast. The upcoming season will be under the banner of REWIRE (also the working title of my next book… but that’s for another time) and focus on conversations with operators who actually rewired organizations, including their failures.
P.S. Such a joy and honor to be (back) on the podcast mic with Berkeley’s Gregory LaBlanc and his unSILOed podcast. Drop in for a wide-ranging conversation about how to better learn from trying new things, a bunch of lessons learned from my past, and so much more.
And now, this…
Headlines from the Future
The AI Hourglass Economy. Back in 2017 I wrote the draft for a book built on an observation we had been making for years (and talking about for as long): Markets are bifurcating into what we dubbed the “Hourglass Economy” (sometimes also referred to as a barbell economy). We could see it everywhere – fashion being one of the canaries in the coal mine. Luxury and niche brands were booming, the Zaras and Uniqlos of the world were doing a roaring trade, and the middle (The Gap) was struggling. In our research, we found a couple of factors that were driving this. For a long list of reasons, we never got around to finishing and publishing the book – and moved on to write Disrupt Disruption. Now the Hourglass Economy is making its comeback – driven by AI:
AI will strengthen the biggest platforms and make the smallest specialists more formidable. The businesses caught between them will face the hardest strategic choice.
And the reason is simple: cost.
This is what a technology barbell looks like. The largest platforms spread their data, expertise and infrastructure across more volume. Small specialists rent capabilities they could never afford to build. Firms in the middle carry enough overhead to need scale but lack enough scale to fund a differentiated platform.
I believe this to be fundamentally true – and we see it happen all the time. If your company is big (and resourced) enough to deploy AI in truly interesting and unique ways, you are, likely, putting yourself into a winning position in your market. If you are small and nimble enough, you can and will use AI in very creative ways (the prime example for me is the amount of value an individual can get out of a $100/$200 monthly subscription versus a corporation which has to pay for individual tokens on their enterprise plans). And if you are in the middle, you find yourself more often than not having to use the same commercial tools as everyone else, can’t afford to build your own stuff, don’t have the benefit of being nimble anymore, and have to compete with the giants in your space who are building their customized AI workflows.
━━━━━
AI Can’t Tell the Time – Or Can It (and the Bigger Question Behind This)? A recent study found that various AI models have a hard time reliably reading an analog clock or figuring out the day on which a date will fall.
“Most people can tell the time and use calendars from an early age. Our findings highlight a significant gap in the ability of AI to carry out what are quite basic skills for people,” study lead author Rohit Saxena, a researcher at the University of Edinburgh, said in a statement. These shortfalls must be addressed if AI systems are to be successfully integrated into time-sensitive, real-world applications, such as scheduling, automation and assistive technologies.”
That being said – and we have seen this many times now – the models being used in the study were GPT-4o, Claude 3.5, and Gemini 2. These are all models which are long out of date (GPT-4o, which a lot of these studies reference, was introduced on May 13, 2024 and has, for quite a while, not been available in the ChatGPT interface). Which brings up the bigger question of how insightful these studies are – and how much headlines such as “AI models can’t tell time or read a calendar, study reveals” really mean.
To investigate AI’s timekeeping abilities, the researchers fed a custom dataset of clock and calendar images into various multimodal large language models (MLLMs), which can process visual as well as textual information. The models used in the study include Meta’s Llama 3.2-Vision, Anthropic’s Claude-3.5 Sonnet, Google’s Gemini 2.0 and OpenAI’s GPT-4o.
What We Are Reading
‘I’ve Definitely Lost Followers’: Influencers Face Backlash Over Meta ‘Pervert Glasses’ Content Seven million pairs of Meta’s smart glasses were sold last year. Now restaurants, pubs, and content creators are banning them because the privacy concerns are proving impossible to ignore. @Jane
An Analysis of 1,000 Meetings Shows How the Best Leaders Shape Conversations Teams that consistently leave conversations with clear accountabilities and next steps are roughly four times more likely to be rated as effective than those that don’t. @Mafe
Claude Was Put in Charge of Human Workers - and Fired 1 Turns out the Turing test for management is losing money, forgetting policy, and eventually firing someone. Claude passed. @Kacee
Ordinary Abundance Sometimes it’s good to remind yourself of how incredibly far we have come. This gorgeous website takes much of what we take for granted (unlimited music in our homes!) and dissects the journey to get there. @Pascal
Down the Rabbit Hole
AI ≠ LLM: Pharma companies Merck and Moderna announced a new, very promising mRNA-based cancer therapy to treat skin cancer. AI outlets such as The Neuron were quick to celebrate: “AI has now, OFFICIALLY, helped design a cancer treatment that just cleared Phase 3.” Sadly, that is – at least in the way the pundits like to use this phrase – not true. The companies developed the drug using neoantigen selection (supervised machine learning on sequencing data) and batch scheduling (which is barely AI in the modern sense), not ChatGPT, Claude, or any other LLM. When it comes to AI (especially in areas like healthcare), it’s not always gold that shines.
Prompt Injection Goes to Court: Prompt injection isn’t something new. A self-representing plaintiff trying his hand at inserting hidden prompts (in tiny text, white on white) into his court filing is. Aside from the gumption and the creativity of doing this (and, of course, the more than questionable ethics of doing this), the fact that the plaintiff clearly just assumed that the court would use AI to process his filing is, in my eyes, the real story here.
Secondary Token Markets: After tokenmaxxing and the token economy, we now have (of course) secondary markets for tokens. Startups often get token credits through various programs they participate in, such as startup accelerators. As a startup founder, you tend to be perpetually minutes away from running out of cash. What do you do? You sell your AI tokens at a discount. And with token prices what they are, it’s quite the market.
How Organizations Use AI: Take the overall study with a grain of salt, as it is OpenAI grading itself, unchecked. That being said, OpenAI conducted a study based on actual token use and message data and found that the lightest users of AI are the ones who are mandating it, with early-career workers (not all too surprisingly) beating their usage by 8–9x. In essence: the mandate to use AI is coming from people who have never actually worked all that much with it.
Google Buys Spirit’s Dead Data: It is well known that AI has an insatiable appetite for data. Just in the last week, we learned that Amazon is buying rare books, scanning them to train its AI models, and then destroying said books (much to the chagrin of book lovers everywhere). Or Grok being trained on SpaceX’s internal documents and communications. Now Google just paid a cool $10 million to buy failed airline Spirit’s 100 million emails, 30 million recorded phone calls, and tons of other internal communications, such as Microsoft Teams chats. And you thought that your little side chat in Teams would go unnoticed?!
The AI Resistance Is Getting Weird: After widespread AI datacenter demonstrations, and some companies never letting a good crisis go to waste, canned-water maker Liquid Death now runs an NFL star Jason Kelce-supported ad campaign asking people to mail their pee to data centers. It’s weird, gross, and quite frankly idiotic. But hey – if it helps them sell you more overpriced tap water in cans, who are we to judge?
ヽ(ツ)ノ And now for some fun:
📱 Before there was an iPhone (and its long list of predecessors), there was the IBM Simon. The device was clearly ahead of its time, and today it is mostly forgotten. Wondering what it was like to use it? Wonder no more. Here is Simon’s official introduction video. Prepare to be amazed.
🇪🇬 Explore ancient Egypt in this neat interactive 3D reconstruction right in your browser.
🪑 Here’s how IKEA comes up with names for its products – straight from the horse’s mouth.
↗ Dive into the deep end: Access our complete collection of 3,000+ radical links.
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.