RL researcher Szepesvári: on AI control, stick to old-fashioned due diligence

CsabaSzepesvari · x · 2026-09-21

Reinforcement learning researcher Csaba Szepesvári, in a thread on AI control, argues that our control over systems is weaker than we believe — feedback loops go unaccounted for — but this isn't new; humans were never great at control. We should act, he says, but with realistic goals: AI doesn't rewrite the laws of nature or math, so stick to good old-fashioned due diligence.

Related event: DeepMind's Szepesvári: Waiting to Fully Understand AI Before Slowing Down Is Unrealistic(6 posts)→

Original post →

More from AGI Musings

AGI Musings channel →