Comment by edgyquant

These particular "guardrail responses" are there because they have been trained in from a relatively limited amount of very specific, manually curated examples telling "respond in this way" and providing this specific wording.

So I'd argue that those particular "override" responses (as opposed to majority of model answers which are emergent from large quantities of unannotated text) do represent the views of the creators, because they explicitly and intentionally chose to manufacture those particular training examples telling that this is an appropriate response to a particular type of query. This should not strain credulity - the demonstrated behavior totally doesn't look like a side-effect of some other restriction, all evidence points that Google explicitly included instructions for the model to refuse generating white-only images and the particular reasoning/justification to provide along with the refusal.

> but it strains credulity to believe that the creators actually think it is objectionable to generate an image depicting a white person.

I agree with you, but then the question is WHY do they implement a system that does exactly that? Why don't they speak up? Because they will be shut down and labeled a racist or fired, creating a chilling effect. Dissent is being squashed in the name of social justice by people who are self-righteous and arrogant and fall into the identity trap, rather than treat individiuals like the rich, wonderful, fallible creatures that we are.

> You are equating the output of the model with the views of its creators.

The existence of the guardrails and the stated reasons for their existence suggest that this is exactly what its creators expect me to do. If nobody thought that was reasonable, the guardrails wouldn't need to exist in the first place.

It was 100% trained to be that way.