Miles Brundage: 'Recursive self-improvement' talk is as misleading as 'FSD (Supervised)'

Miles_Brundage · x · 2026-10-10

Former OpenAI AGI Readiness lead Miles Brundage argues that phrasing like "recursive self-improvement (and we'll figure out later what role humans play)" carries the same energy as "Full Self-Driving (Supervised)" — vague enough plus the right keywords to satisfy some critics, while being fundamentally misleading about the goal actually being pursued.

Related event: Ex-OpenAI safety lead says no frontier AI firm has a safe self-improvement plan(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →