New AI risk debate: stop betting that superintelligence can't do real-world physics

daniel_mac8 · x · 2026-09-16

tszzl challenges AI risk skeptics: if your safety argument rests on future superintelligent models being unable to interact with and research new physics in the real world, you need a new argument — implying frontier models' progress in real-world experimentation is already undermining the "models stuck in cyberspace" safety assumption.

Related event: Researcher tells AI risk skeptics: "models can't interact with reality" argument is dead(4 posts)→

Original post →

More from AGI Musings

AGI Musings channel →