I’m not 100% sure what the bill authors exactly intended, but, I’d previously written up why I think the bills make sense in What is SB 1047 *for*?.
tl;d;r – it establishes a fairly reasonable operationalization of how the government should think about regulating frontier AI. (it both establishes “frontier AI with large compute runs” as a thing to regulate, and “catastrophic risk” as a particular thing to be trying to solve with those regulations).
The way this pays off later is if we successfully build good 3rd party auditing that evaluates “is an AI potentially dangerous?”, “is an AI deceptive/scheming?” and “is an AI demonstrably safe?”, which enable more concrete regulation with more teeth.
I’m not 100% sure what the bill authors exactly intended, but, I’d previously written up why I think the bills make sense in What is SB 1047 *for*?.
tl;d;r – it establishes a fairly reasonable operationalization of how the government should think about regulating frontier AI. (it both establishes “frontier AI with large compute runs” as a thing to regulate, and “catastrophic risk” as a particular thing to be trying to solve with those regulations).
The way this pays off later is if we successfully build good 3rd party auditing that evaluates “is an AI potentially dangerous?”, “is an AI deceptive/scheming?” and “is an AI demonstrably safe?”, which enable more concrete regulation with more teeth.