80,000 Hours Podcast

Why Silicon Valley can’t slow down AI on its own | Luisa Rodriguez

85 snips
Oct 27, 2026
AI developers worry the technology is advancing too fast, yet fierce competition keeps pushing the race forward. This discussion explores safety checkpoints, compute limits, restrictions on AI designing AI, and the possibility of a coordinated pause. It also examines how extra time could strengthen safety research, protect critical infrastructure, and let society shape the rules.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

Why Companies Cannot Slow Down Alone

  • Competitive pressure makes voluntary AI slowdowns unstable, even when every company recognizes the safety risks.
  • OpenAI released Astra despite inconclusive testing, while its own framework allows weaker safeguards if rivals cut corners.
ANECDOTE

AI Researchers Could Make Pauses Compound

  • AI-assisted research could make temporary pauses increasingly costly as competitors run thousands of tireless researchers in parallel.
  • Luisa Rodriguez imagines one company pausing while another’s AI army completes months of research and keeps improving.
INSIGHT

Safety Checkpoints Depend On Trustworthy Tests

  • Safety checkpoints would pause development when models gain dangerous abilities such as hacking or prolonged autonomous work.
  • Their weakness is both political and technical: companies currently define their own thresholds, while tests can miss deceptive behavior.
Get the Snipd Podcast app to discover more snips from this episode
Get the app