@Riedl This kind of stuff is why I can't in good conscience work with ChatGPT, even just to figure out how broken it is. Fueling the hype in any way is wrong.

There is an extremely strong linguistic convention that you cannot tell another person that they are sorry. But apparently to ChatGPT, there's an even stronger convention that professors must use 'he' pronouns.

Obviously, this kind of stereotype-over-logic result pervades all aspects of what LLMs generate, even when they're too subtle for us to notice so clearly.