
80,000 Hours Podcast An AI disaster hits. The president calls Xi. Then what? | Jeremie Harris (crosspost from NonZero)
71 snips
Nov 5, 2026 Robert Wright, journalist and founder of NonZero, speaks with Jeremie Harris, AI entrepreneur and safety researcher. They explore why AI progress keeps accelerating while alignment lags, whether only an “AI Chernobyl” could change Washington’s course, and how OpenAI’s safety culture has evolved. They also examine US-China mistrust, crisis-driven cooperation, verification, and the alliances needed to prevent catastrophe.
AI Snips
Chapters
Books
Transcript
Episode notes
Why Yudkowsky May Have Been Directionally Right
- Harris argues that Eliezer Yudkowsky's broad loss-of-control prediction can remain impressive even when specific architectural predictions fail.
- A distant prediction may look like one peak, then resolve into several mechanisms as technology approaches it.
An AI Chernobyl Is Defined By Government Response
- Harris defines an AI Chernobyl by its political consequence rather than a fixed casualty count: it must force the US government to slow or pause AI.
- The resulting domestic slowdown would require negotiations with China because unilateral restraint could simply shift the catastrophe abroad.
Alignment Has Not Matched Capability Progress
- Harris says researchers have made essentially no meaningful progress controlling systems that are smarter than humans.
- Existing safety techniques can fail even on current models despite enormous commercial incentives to make them reliable.





