Comment by embedding-shape
20 days ago
Yeah, I mean, ultimately, aren't the LLMs actually trained to look like human language? So whatever particular "quirk" you have as a writer, there is probably an LLM emulating that either wholesale, or like 50% of the times.
LLMs use em-dash because people (in their training data) used em-dash. They use "You're absolutely right" because that's a common human phrase. It's not "You write like an LLM", it's "The LLMs write kind of like you", and for good reasons, that's exactly what people been training them to do.
And yes, "pun" intended for extra effect, that also comes from humans doing it.
The LLM output isn't an unfiltered result of an unbiased model. Rather, some texts may be classified high-quality (where the em-dash, curly quotes, a more sophisticated/less-everyday vocabulary are more expected to appear), some low-quality, and some choices are driven by human feedback (aka fine-tuning), either to improve quality (OpenAI employs Kenyans, Kenyan/Nigerian English considered more colonial) or engagement through affirmative/reinforcing responses ("You're absolutely right. Universe is indeed a donut. Want me to write down an abstract? Want me to write down the equations?"). Some nice relevant articles are [1],[2].
[1]: https://marcusolang.substack.com/p/im-kenyan-i-dont-write-li... [2]: https://www.nytimes.com/2025/12/03/magazine/chatbot-writing-...