For conversations: I'm not fully up to date on the technology, but last year's text generator versions could still be confused if you suddenly switched topics. Like iirc that published Lamda interview was very careful not to throw any tricky questions into the dialogue.
For just regular text: ChatGPT still has a characteristic "voice", though as far as I heard they are working on toning it down. I know you can already make it "roleplay" though which seems to be successful in taking on very different tones. ("pretend you're a grandma")
I think as time goes on we can be less and less sure if what we're reading are actual humans, and I don't think there's any way we can defend ourselves either against exposure or accusations 🤷♂️