AI Is Too Risky to Insure
Sundar Pichai hand-waves his way to obsolescence, Ozempic becomes a life sentence, and we’re all using ChatGPT for emotional support instead of work
Dear Friend,
As the US celebrates its grand tradition of Thanksgiving, the world surely doesn’t slow down – just last week we saw another round on the seemingly eternal merry-go-round of who has the best AI model. First shots were fired by Google with the launch of Gemini 3 Pro, then OpenAI returned the salvo with their latest ChatGPT model, only to be trumped by Anthropic’s release of Opus 4.5. The lesson? Whatever frontier AI floats your boat – stick with it, they are all about as good as each other (with nuances).
Meanwhile, the team at radical enjoys some delicious food, good company, and stories told by actual humans, not AI-generated slop (though human stories can also become a little sloppy after the third glass of wine).
P.S. Podcast alert! I just dropped the latest episode of our podcast – a superbly insightful conversation with Jeff Seibert, Founder and CEO of Digits, the AI-native accounting platform. Jeff is wicked smart, and it shows! I hope you enjoy listening to this as much as I did when we recorded it.
P.P.S. With this Friday Briefing, we’re launching a new section – inspired by Kai Ryssdal and Molly Wood’s fantastic Make Me Smart podcast, we’ll be asking their clever signature question at the end of our interviews. Scroll down to see both the question and the answer.
And now, this…
Headlines from the Future
The AI Insurance Conundrum. Insurance companies are balking at insuring anything AI – from risks involving companies using AI to generate content, make decisions, or run processes, to the sheer idea of using AI in the first place.
Major insurers including Great American, Chubb, and W. R. Berkley are asking U.S. regulators for permission to exclude widespread AI-related liabilities from corporate policies.
Make no mistake – this is a huge issue not just for companies building AI models and AI-powered apps, but for any company using AI in their processes. Simply put, if you are using AI and something goes wrong (say, you are an accountant and your use of AI in accounting resulted in an error in your client’s tax return), you – not the software vendor, not the AI model your software vendor is using – are liable to your client. This could prove to be a major hurdle to overcome when it comes to the widespread adoption of AI-powered workflows and systems.
↗ AI is too risky to insure, say people whose job is insuring risk
━━━━━
Google CEO Puts Himself Out of a Job. In one of the more remarkable “let’s just pretend AI is going to solve every problem” hand-waving statements, Google’s CEO Sundar Pichai made the bold assertion that AI will soon be able to do his job:
“I think what a CEO does is maybe one of the easier things maybe for an AI to do one day,” he said. Although he didn’t talk specifically about CEO functions that an AI could do better, Pichai noted the tech will eliminate some jobs but also “evolve and transition” others—ramifications that mean “people will need to adapt.”
The important part here is, “Although he didn’t talk specifically about CEO functions that an AI could do better […]” If only every problem in the world could be solved by making a vague statement and moving on. But hey, Sundar will soon have plenty of time to work on other things, as AI will steer his company.
↗ Google’s Sundar Pichai says the job of CEO is one of the ‘easier things’ AI could soon replace
━━━━━
GLP-1 – The Forever Drug. We have talked about GLP-1 weight-loss drugs here before – they seemingly came out of nowhere (at least in the public eye), have skyrocketed into a massive category, and promise to solve much more than just our weight issues. But they come with a massive downside (which, I am sure, big pharma won’t mind): you can’t get off them without losing all the benefits (and gains you made).
An analysis published this week in JAMA Internal Medicine found that most participants in a clinical trial who were assigned to stop taking tirzepatide (Zepbound from Eli Lilly) not only regained significant amounts of the weight they had lost on the drug, but they also saw their cardiovascular and metabolic improvements slip away. Their blood pressure went back up, as did their cholesterol, hemoglobin A1c (used to assess glucose control levels), and fasting insulin.
Another good reminder that there is no such thing as a free lunch.
P.S. Meanwhile there is evidence popping up, showing that GLP-1 class drugs can lead to chronic cough…
↗ There may not be a safe off-ramp for some taking GLP-1 drugs, study suggests
What We Are Reading
😤 Tech Should Help Us Be Creative. AI Rips Our Creativity Away Tech companies are mass-producing musical spam and calling it “democratizing creativity,” because nothing says democracy like replacing musicians with robots! @Jane
🤗 We Analyzed 47,000 ChatGPT Conversations. Here’s What People Really Use It For. People primarily use ChatGPT for emotional support, personal advice, and help navigating real-life problems more than for tasks related to work productivity. So more of a companion, less of a search engine. @Mafe
🌍 Why Solarpunk Is Already Happening in Africa Stepping away from our own borderline cyberpunk day-to-day, we can see a very different (and arguably more inspiring) example of a convergent, tech-enabled future unfolding in Africa. @Jeffrey
🌀 Quantum Thinking Can Help You Solve Complex Strategy Challenges As we move beyond linear models, quantum thinking reminds us that the messy overlap of choices, consequences, and system dynamics is where the new strategic insight lives. @Kacee
🤔 Beyond the Machine – Creative Agency in the AI Landscape A wonderful, long read on an artist’s perspective on AI, from tool to instrument. Discusses AI being above, next to, or under us. Highly recommended food for thought! @Pascal
Down the Rabbit Hole
🎤 Pixar: The Early Days – A never-before-seen 1996 interview
⚛️ Kodak Ran a Secret Nuclear Device in Its Basement for Decades. It Was a Scientific Marvel.
🧐 Microsoft’s head of AI doesn’t understand why people don’t like AI, and I don’t understand why he doesn’t understand because it’s pretty obvious
🤖 Figure AI sued by whistleblower who warned that startup’s robots could ‘fracture a human skull’
😧 Ouch: Study finds nearly two-thirds of AI-generated citations are fabricated or contain errors
🤿 Gorgeous storytelling: How Deep is Challenger Deep?
📜 Use a poem, hack a LLM: Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Models
💣 How The Atomic Tests Looked From Los Angeles
🏦 The ’S&P 493’ reveals a very different U.S. economy
🛰️ Voyager 1 Is About to Reach One Light-day from Earth
Make Me Smart
What’s something you thought you knew that you later found out you were wrong about?
I thought you could cool your drink faster by adding more ice or using those massive cubes you see at cocktail bars without diluting it as much, but my wife, who’s a chemical engineer, proved me completely wrong – ice only cools stuff when it melts, so no matter what shape or size your ice is, you need the same amount to melt to get your drink from one temperature to another.
— Jeff Seibert, Founder and CEO of Digits
Pascal is working through the responses from our inaugural radical Pulse… Stay tuned – we will publish the findings next week.