Why do AI models frequently use em dashes (–) in their writing?

Because skilled writers have started using in place of semi-colons and parentheses in modern writing - it just feels right!

AI models are trained off the work of these writers, so they use them now. Except that the difference is that a writer will just use the - next to the `0` key, expecting that Word, Google Docs, or whatever is turning their markdown into rendered text will automatically turn it from - to an em dash, which is a different character.

The models are then provided that rendered or autocorrected text, and return text with the em dash, instead of the one we type.

One of my degrees in college was in Technical Writing, back in the early 1990s. I've always used em dashes when writing--it only make sense in many cases. And I still use them regularly, just typing two hyphens, one after the other.

And so many emoticons. Wonder what training set has these many characters