Parts of the AI world are built on shaky conceptual scaffolds

sebkrier · x · 2026-08-19

Safety researcher Seb Krier argues that abstractions should be treated as instruments to be refined, forked, and adapted, yet too many people treat them as fixed immutable packages.

He suggests large parts of the AI world rest on shaky conceptual scaffolds: the field is too quick to latch onto surface-level similarities and correlations, and rarely stops to ask whether competing causal stories require redrawing categories—while questioning high-status stories is rarely encouraged.

Original post →

More from AGI Musings

AGI Musings channel →