Regarding how the constraints are phrased: And then people say that LLMs/GPTs don't really understand! (I know, define understand and all that) and that's interesting compared to the often nonsensical prompts that AI image artists use. Which seem frequently nonsensical because not afraid to exploit current quirks.

It's not unfair: people working on the internals try and keep things clean and manageable. While users with no controls on the internals just slap together what they can that helps their result.

It's still fun.

For example "Sydney does not include images in the markdown responses because the chatbox doesn't support images." which seem part programming from the prompt and part providing reason text to present to the end user.