Basically people that are going to do reinforcement learning to ensure the model generates output according to a set of pre-defined values. So, this people defi

Basically people that are going to do reinforcement learning to ensure the model generates output according to a set of pre-defined values. So, this people defines what those values and principles are, and then they make sure the model generate stuff aligned with those. We would usually call this moderation, but for branding, marketing and financing purposes it is far more interesting to call it safety because it reinforces on the lay mind an idea that those models are far more powerful than they really are and that we are on the cusp of General AI. It is like lowering the suspension, painting the wheels black, and putting rear flaps and badges that say "Sport" and "Turbo" in a unmodified Toyota Corola.