The “em-dashes” (—) come up a lot in online translations of books like Bible and Quran.
Normal keyboard “-” and “–” are different from “—” but microsoft office auto-formats “–” to that.
I kinda assumed it was ALL microsoft word data that caused training to include that.
I am only now realizing AI stole from even the religious texts and influenced by them as well.
Software converts human-typed comments to use fancy quotes, dashes and other punctuation. Even this platform does that with the Markdown extension Fancypants - look at the quotes in your post.
That’s where LLMs get this from.
E.g.: to get an em-dash here:
---=> —I think that highly depends on the client you use:
But I see your point