Researcher: AI safety movement 'sidesteps' alignment by being bad at both theory and empiricism
Apoorva__Lal · x · 2026-09-16
Researcher Apoorva Lal argues that AI alignment is a hard problem requiring a rare mix of formalization prowess for disciplined extrapolation and empirical pragmatism to run experiments and learn mechanisms—and claims the current AI safety movement 'cleverly sidesteps' these issues by being bad at both. A pointed, substantive critique of the field's methodology.
More from AGI Musings
- Gary Marcus on AI liability: 'the risks WERE foreseeable; I foresaw them' amid Altman-Amodei warning debate — GaryMarcus · 2026-09-16
- Compute maximalist: no magic 4-6 OOM training gains, so Ant or OpenAI win — anpaure · 2026-09-16
- Danish administrative data finds no effect of AI on earnings or wages — paulnovosad · 2026-09-16
- 150ms model decisions may end fixed-loop agent harnesses — GlenBradley · 2026-09-16
- Silicon Valley Hiring Now Prizes Human 'Agentic' Skills, Says Chinese Tech Observer — vista8 · 2026-09-16
- Scott Aaronson: AI labs sitting on major problem solutions after math community backlash — Tolopono · 2026-09-16