Critique of Rationalist AI Safety: Bad Abstractions and Wrong Predictions
sebkrier · x · 2026-08-29
The author critiques the Rationalist AI Safety corpus for being full of 'wordcelery' and bad abstractions, arguing it lacks fine-grained mechanics of intelligence. It claims the group was wrong about AI's form in the 2010s and fails to apply Hayekian insights to their own thinking. The author argues that a vague, vibes-based approach will outperform in the long run.
More from AGI Musings
- AGI-5 only requires superhuman capability at all work, a crazy low bar — RileyRalmuto · 2026-08-29
- AI should be in the service of humans — sierracatalina · 2026-08-29
- Alex Karp on real enterprise advantage in the AI era — Olivier__OG · 2026-08-29
- AI infrastructure's high IRR may push traditional businesses into the abyss by raising capital costs — elonmusk · 2026-08-29
- AGI definition diluted; cognitive levels met long ago, argues user — RileyRalmuto · 2026-08-29
- DeepMind CEO: AGI likely just a few years away, marking a pivotal historical moment — Ghost_Pilot_MD · 2026-08-29