Paul Christiano: 1-in-3 chance of government-toppling superintelligence within a year
Dwarkesh Patel · youtube · 2026-11-17
On Dwarkesh Patel's podcast, Paul Christiano (head of ARC, inventor of RLHF, now on OpenAI's board) argues there's a 1-in-3 chance superintelligence powerful enough to overthrow governments arrives within a year absent slowdown. He urges labs to pause algorithmic progress until weights and research secrets are secured, warns even datacenter-trained systems could be uncontrollable, and says no current alignment approach has better than even odds — hope rests on AIs solving alignment during the intelligence explosion.
More from AGI Musings
- Ben Bajarin: enterprise AI adoption is the only conviction that matters for the decade — BenBajarin · 2026-11-17
- LangChain founder: agents are shifting from executing tasks to owning goals — hwchase17 · 2026-11-17
- Sam Altman on stage: "We didn't have AGI. Now we do" — and OpenAI is withholding self-improving AI — imjustnewatai · 2026-11-17
- Vik Paruchuri: Even Experts Struggle to Tell Good Benchmarks from Bad Ones — VikParuchuri · 2026-11-17
- Report: US Has 9x China's Compute, So It Can Afford to Pace the AI Frontier — trevposts · 2026-11-17
- Open-source models now serve 30% of tokens, growing as fast as closed frontier AI — ivan_bezdomny · 2026-11-17