Voices
Emad Mostaque calls embedded evaluators "structurally hollow" because they can be politely ignored
The Stability AI founder's objection is mechanical rather than ideological: evaluators granted access but no enforcement power cannot constrain a company's decisions, making the pledge a plan whose only teeth belong to people Amodei designed to be overridable. He extends the argument in a companion post, "Intelligence isn't a crime," contending raw capability growth is not itself the risk vector worth regulating. Sourced from aggregator coverage rather than the original post, so treat the exact wording as approximate.
Source
↳ Follow the thread