Washington Hones in on AI Safety, What Are AI Doomers Proposing | Diet TBPN

Summary

TBPN hosts discuss a new AI 2040 proposal from AI doomers that calls for slowing frontier AI development through compute caps, data-center audits, and international verification while keeping current inference and capabilities research going. They also cover Bernie Sanders' proposal to ban superintelligence, industry pushback from compute providers, and Tyler Cowen's challenge to make AI doom bets falsifiable. The episode touches on Paul Christiano's leveraged AI-heavy portfolio and short long-bond position, plus lighter segments on simulated fruit flies and AI moral questions.

• AI 2040 proposes pausing frontier training and enforcing compute caps, data-center audits, and international verification.
• The plan aims to delay superintelligence to 2040 while allowing current models and inference to continue.
• Bernie Sanders proposes banning superintelligence with prison sentences and a corporate death penalty.
• Compute providers and semiconductor interests are pushing back on aggressive slowdown regulation.
• Tyler Cowen's challenge asks AI doomers to name market prices that would confirm their pessimism.
• Paul Christiano's portfolio is described as levered long AI, short US 30-year debt, and 5% Tesla.
• The hosts discuss simulated fruit flies and moral questions around digital minds.
• The episode ends with light banter about AI agents playing games.