Hacker News new | past | comments | ask | show | jobs | submit
nradov | on: OpenAI’s head of ethics leaves less than a year after joining
It's not technically possible to prove model alignment. At best you can establish a probability of alignment within certain constraints.
madrox | parent
"Prove" is a shorthand because of course it's a stochastic process. My point is that I would expect ethics to be part of the training and eval criteria.