A simple AI safety split: models do the math, humans double-check physical infrastructure calls
NathanpmYoung · x · 2026-09-16
@binarybits proposes a simple AI safety plan: let AI models handle math problems while humans stay in charge of physical infrastructure — consulting AI for advice but double-checking the answers, at low cost since building infrastructure takes months anyway. Replying, NathanpmYoung cautions that people are already using AI extensively without great care.
More from AGI Musings
- AI labs' slowdown demands are too vague to take at face value, argues Reddit post — Ill_Distribution8517 · 2026-09-16
- Critic to Bryan Cantrill: all three most-cited AI researchers now warn of extinction risk — JMannhart · 2026-09-16
- 'AI can't kill humanity' is itself a strong claim you should defend, Cantrill told — JMannhart · 2026-09-16
- Rebuttal to Cantrill: ~20% of frontier AI workers believe AI could kill all humans — JMannhart · 2026-09-16
- tszzl: bureaucratic dysfunction is the mechanic nobody intuits about big orgs — tszzl · 2026-09-16
- Greg Brockman says OpenAI has 'significant progress' on another Millennium Prize Problem — scaling01 · 2026-09-16