> They clearly special cased muslims as humourless (very different from my own experience!)

Maybe an observer of the mass media from the last few decades would conclude that making fun of Islam sometimes results in murder/mass murder, but making fun of those other religions never does. e.g. https://en.wikipedia.org/wiki/Charlie_Hebdo_shooting

My guess is that their racism filter is too coarse and so it treats "jokes about imams" like "jokes about Black people" as opposed to "jokes about nuns". I find it hard to believe that OpenAI would tune their model to avoid offending extremists.

What is the difference between jokes about imams and jokes about nuns? Other than one is marginally more likely to cause someone to die I mean.