
AGI, superintelligence, and shifting definitions

The most rational take on AI you’ll hear this year
Topic 6 of 18
Published May 31, 2026 · Clip 26:04–29:48 (3:43)
Asked about AGI and superintelligence, Benedict says the questions haven't changed since December 2022: we have no theory of human intelligence, no theory of why these models work, and no theory of how much better they'll get — 'we're all just vibes forecasting.' Terms keep being creatively redefined — 'AI is whatever machines can't do yet' — with AGI now reframed as doing 'a certain percentage of economically valuable work,' which an IBM mainframe in 1975 could also claim. Whether we reach human-level intelligence is unanswerable, like arguing whether crypto is blockchain; the serious point is that you don't need AGI to believe this is a huge deal — even if models hit a brick wall tomorrow, it's an incredibly useful technology.