
Simulation Hypothesis Musings

A mind-expanding conversation on how models actually work | Emmanuel Ameisen (Anthropic)
Topic 13 of 18
Published Oct 18, 2026 · Clip 55:43–59:01 (3:18)
Emmanuel says looking inside models reveals that independently trained models converge on surprisingly similar internal structures, suggesting there's one efficient way to solve important problems — a fundamental structure of information. This makes him open to the simulation hypothesis (a simulated world might land in a similar 'basin'), though he doesn't fully buy it because he can't answer who's running it or why. The host cites Elon Musk's point that simulations would be run at interesting times in history.