AI Policy Figure Notes Surge in 'Vetting' Term for Model Evaluation
Perhaps there are worlds in which such a function should exist at CAISI
But in today's world, CAISI gets in trouble for making banal announcements, let alone gating deployment
Very strange times
It's interesting that "vetting" became a widely used term for pre-deployment testing/evaluation of AI models in the past week It's not the most common such term, but more importantly, it implies a judgment re: readiness for deployment which is exactly what CAISI hasn't done
To clarify: CAISI evaluates properties of systems, both American and Chinese to date
That is different from judging their suitability for deployment, which is what "vetting" implies (at least to some ears/readers, I suspect)
Perhaps there are worlds in which such a function should exist at CAISI But in today's world, CAISI gets in trouble for making banal announcements, let alone gating deployment Very strange times