Concept
AGI
Version 2 — The guest states his theory of AGI directly — stack specialized intelligences trained in RL environments until no gaps remain, with verifiability as the only requirement — and marks taste as the residue that resists it. (define-concept@5, claude-opus-5)
AGI is machine intelligence that matches human competence across the full range of tasks rather than excelling at one. The term marks breadth, not peak skill, which is why proposals differ on whether generality falls out of scale or accumulates as specialized capabilities stacked until no gaps remain. The stacking route needs each task to be verifiable, so it grades itself.
Used by
A five-hour conversation about what AI does to the craft of programming. What DHH claims, and what it rests on.