Gary Marcus: AI firms self-testing safety is like tobacco companies grading themselves
GaryMarcus · x · 2026-09-18
Gary Marcus revived his jab that tobacco companies shouldn't just run their own tests and use their best judgment about whether their products harm users — a pointed analogy aimed at AI labs self-evaluating their models' safety. He added that with enough money and power, people start believing their own statements. A notable voice in the AI safety policy debate.
More from AGI Musings
- agi.fyi Launches as the 'Wirecutter for AI Risk Content' With Five Curated Guides — KatjaGrace · 2026-09-18
- Turing Post Publishes RSI 2026 Guide: Anthropic, Recursive, Sakana AI Show First Real Steps — TheTuringPost · 2026-09-18
- "FOOLS, you don't understand SUPERINTELLIGENCE": the hype-posting debate over domain expertise — tekbog · 2026-09-18
- AI removes the reading friction that protected 'first to publish' norms, researcher argues — jd_pressman · 2026-09-18
- Is recursive self-improvement just PR? Redditor questions the compute bottleneck — Additional-Spray-976 · 2026-09-18
- Noam Brown's Dwarkesh podcast: a capabilities researcher 'freaking everyone out' on safety — Apprehensive_Sand951 · 2026-09-18