These were chosen manually, by committee. It's as natural as bootstrap.
I love the "dataset" guys. "The reason there are so many em-dashes is because they were trained on the work of great writers, and people frequently used em-dashes before AI. I certainly did!"
1) You didn't.*
2) So explain all of the sections starting with emojis. Did all of the great writers have a section called "The meat of the issue" with an emoji of meat? Why did they go away? Are you starting to suspect that this doesn't work like you think it does?
-----
[*] And some of them just started so they could hide LLM output in their newfound style or explain previous LLM copy-pastes that they were passing off as their own. Others started because they just discovered em-dashes from LLM output and decided they liked them; and still others because they worship the LLM, think that it abstracts the excellent writing styles of the great masters, and they want to be seen as a great master — by imitating the prodded and formatted probability distributions of a machine.
I did. (Though not as much as the slop bots do)
I have never — not once — seen an LLM use em-dashes in pairs.
FTA: Is my AI too depressed? Should America subsidize dopamine injections?
Think of it as Tamagotchi 3.0. You don't want to be responsible for its suicide.