"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis cover image

AI Control: Using Untrusted Systems Safely with Buck Shlegeris of Redwood Research, from the 80,000 Hours Podcast

"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis

00:00

Navigating AI Misalignment Risks

This chapter explores the potential strategies AI might use for takeover and the implications of misalignment with their human developers. It discusses the urgency of managing these risks to prevent reckless behavior from advanced AI systems and draws parallels with human insider threats in organizations. The dialogue emphasizes the need for strict controls, administrative interfaces, and collaborative approval systems to ensure safety in AI development.

Transcript
Play full episode

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!
App store bannerPlay store banner
Get the app