I wonder what happens to all that 18 century books scanning data. I imagine it stays proprietary and I've heard a lot of the books they scan are destroyed afterwards.

I’m now reading Pride and Prejudice (first edition released in 1813) and indeed there are many em dashes. It also includes language patterns the models didn’t pick up (vocabulary, to morrow instead of tomorrow)

My question is given their satirical association with AI, why haven’t the models been manually optimized not to use them?

I’m amongst those who used to use em-dashes and now seeks to actively avoid them.

In Russian written languages, the quotes for the people speaking are prefixed with em-dash, instead of double-quoted like it would be in typical English book:

Instead of

"The time has come," the Walrus said,

"To talk of many things:"

... it would be spelled as

— The time has come, — the Walrus said,

— To talk of many things:

I wonder how much of russian language content was in training model.

Are people surprised that training biases a distinct style? I'd think it's kind of expected

Because Sam Altman said so

I always figured it was because of training on Wikipedia. I used to hate the style zealots (MOStafarians in humorous wiki-jargon) who obsessively enforced typographic conventions like that. Well I still hate them, but I'm sort of thankful that they inadvertently created an AI-detection marker. I've been expecting the AI slop generators to catch on and revert to hyphens though.

Robert A. Heinlein used a lot of em-dashes and much of the Internet was created by Heinlein fanboys?

The conclusion is really a guess unfortunately.