Unsupervised Learning with Jacob Effron
Unsupervised Learning with Jacob Effron

Ep 91: Top AI Analyst Unpacks Today's AI Hype Cycle

July 16, 2026

AI Summary

5 min read

“The problem here is he doesn’t actually understand what radiologists do.” That’s Benedict Evans on Geoffrey Hinton’s famous claim, a decade ago, that we should stop training radiologists. Evans’s point isn’t just about one prediction being wrong. It’s about a recurring pattern in AI discourse: mistaking what a model can do in a narrow test for what a job actually involves. This episode of Unsupervised Learning is a patient, rigorous unpacking of that pattern—and many others—by one of tech’s most clear-eyed analysts.

The Limits of Historical Analogy

Evans pushes back hard on the impulse to declare that AI is “like electricity” or “like the internet.” He doesn’t dismiss the comparisons, but he insists they be used carefully. “You can wave your hands and say, no, this is like electricity. Okay, fine, it’s like electricity. Well, what happened with electricity?” The point is not to prove a prediction by analogy, but to mine the past for useful patterns. He runs through several: semiconductors had escalating costs (Moore’s Law doubling the cost of a cutting-edge fab every four years); mobile networks had marginal costs (adding traffic meant building more base stations); mobile data traffic grew 1,000–2,000x in 15 years, created a trillion-dollar industry with $200B annual CapEx, and “they don’t make any money… all the value went upstack.” The question for AI is which layer

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

Listen to Audio Summary Open in App

What you'll learn

  • 1 (00:00) **Episode Introduction** - Host Jacob Effron introduces guest Benedict Evans, a top tech analyst, and previews the conversation on the AI hype cycle, value accrual, and job impact.
  • 2 (01:46) **The "How Big Is This?" Question** - Evans argues that comparing AI's scale to past technologies is less useful than studying *how* those technologies actually unfolded.
  • 3 (06:08) **The Radical Uncertainty of Model Progress** - Evans highlights a critical difference from past platform shifts: we have no scientific understanding of why these models work or what comes next.
  • 4 (08:54) **The "Descartes Fallacy" and Historical Hype** - Evans warns against philosophical hand-waving and reminds listeners of the wild-eyed utopianism that surrounded the early internet.
  • 5 (11:34) **The "Daily Active User" Problem** - Evans argues that AI's current usage is shallow, with only 10-15% of people as daily active users, and most using it once or twice a day.
  • 6 (14:12) **The Hardest Part: Isolating the Task** - The real bottleneck isn't model capability, but the user's ability to isolate and describe a task they could give to the AI.
  • 7 (17:38) **The "Lump of Labor" Fallacy and Job Impact** - Evans argues that predicting job loss by measuring current tasks is a delusional exercise, akin to the "expert in the system fallacy."

+ Full timestamped outline available in the app

Show Notes

Benedict Evans, one of tech's most widely-read analysts, joins Jacob Effron. The conversation centers on Benedict's core thesis that comparing AI's scale to past platform shifts (the internet, mobile, PCs) is analytically useless, and that the more productive move is studying how those previous technologies actually evolved economically to reason about where AI's value will accrue. He argues the one genuine difference this time is that we don't know AI's physical or scientific limits, unlike past shifts where the boundaries were at least knowable, and that this uncertainty is what fuels both AGI hype and doomerism without resolving anything. Benedict unpacks why capabilities remain jagged, meaning usage is jagged too, why coding became the first real enterprise use case thanks to scalable verification, and why most consumer and enterprise use cases still have to be invented by entrepreneurs rather than emerging spontaneously once models improve. He also lays out why foundation model labs may end up structurally like TSMC rather than Windows, valuable but bounded rather than owning the entire stack, walks through why automation has historically meant more work rather than less (using a hundred years of rising accountant headcount as evidence), and explains why industries like Uber and Airbnb, or Caterpillar and the internet, show just how unevenly this kind of technology actually lands. Throughout, he offers candid, historically grounded takes on OpenAI's product sprawl versus Anthropic's narrow coding bet, Apple's stumbled AI moment, and why most companies, unlike Silicon Valley, have far bigger priorities than AI on their minds.

 

(0:00) Intro
(1:31) Is AI Bigger Than the Internet?
(10:10) Barriers of Getting From Demos to Daily Use
(20:15) Why Job Predictions Fail
(25:52) Where's the Moat?
(33:55) Will Models Eat the App Layer?
(39:25) When Average Isn't Enough and Models Don't Work
(45:58) Reflections on OpenAI
(55:04) Consumer Usage Is Still Shallow
(58:51) What's Required for More Enterprise Adoption
(1:03:47) Opinion on Sora
(1:06:27) Quickfire

 

With your host:  

@jacobeffron  

- Managing Director at Redpoint

Unsupervised Learning with Jacob Effron

More from this podcast

Unsupervised Learning with Jacob Effron →