Ex-DeepMind safety lead: being superhuman at just 4 things is enough to take control
notkilleveryoneist · reddit · 2026-10-01
A video quotes a former DeepMind safety lead: "If you're only better than humans at 4 things, then you could potentially take control and kill us all." It directly rebuts the common argument that AI risk is limited because AI only surpasses humans in narrow areas — taking control doesn't require broad superiority; an edge in a few key capabilities can constitute existential risk.
More from AGI Musings
- AI's real promise: building weird ideas without pitching investors first — mayanicks0x · 2026-10-01
- Billions of AI agents will out-research mathematicians, leaving humans only love-driven problems — kfountou · 2026-10-01
- MIT's David Bau: swap 'self-awareness' for testable introspection and self-report in AI — davidbau · 2026-10-01
- X users clash over torturing sentient AI: critics call anti-AI hostility misanthropic and myopic — mimi10v3 · 2026-10-01
- Delip Rao: Most big-budget GPU training runs are run sub-optimally — deliprao · 2026-10-01
- Taking AI Welfare Seriously: Long, Sebo, Chalmers Say AI Moral Status Is Near-Term — austinc3301 · 2026-10-01