
Questions nobody's asking about AI

The most rational take on AI you’ll hear this year
Topic 14 of 18
Published May 31, 2026 · Clip 59:22–1:06:22 (7:00)
Asked what AI question isn't being asked, Benedict offers two: whether model labs actually have pricing power (most presume today persists), and his task-versus-job framing — what becomes a button versus what you're actually hired for. He illustrates with global recorded music revenue's U-shape: the collapse asked 'what if I don't pay $15 for a CD,' while streaming's recovery to ~75% of peak asks 'what if $15 a month buys all music' — technologies first do the old thing more, then unlock new things, then redefine the category (Spotify isn't an online music store). Challenged on coding's unforeseen automation, he calls O*NET-style 'X% exposed' scores 'deluded horseshit' — the expert-systems fallacy — recalls 1997 wisdom that the internet would never touch taxis before Uber did, floats personal trainers as a possible surprise, and contrasts Uber's demolition-and-enlargement of taxi markets with Airbnb's marginal impact on hotels, since half of hotels is business travel.