alt Hacker NewsThis heuristic is commonly believed — and expressed on this website — but is idiotic. Plenty of people have known how to use em dashes when writing on a computer since the 1980s. (This was true even before UTF-8 was invented to make extended characters work across platforms; instead of ⌘⇧-hyphen, we'd just type "--" to approximate it.)
The reason LLMs tend to use a lot of em dashes is that the human-authored texts they were trained on used a lot of em dashes.
Replies
> Plenty of people have known how to use em dashes
Plenty? I don't think it's that many. And knowing how to use them, doesn't mean they would actually use them. In my experience, em dashes are mainly found in academic literature. Chances that an em dash in a new text comes from an LLM are pretty high.
Your use of it is dubious, I think. Commas would serve equally well --or better-- since it's a simple conjunction, not an interjection.
This is untrue and a naive understanding of modern model training.
Opus 5 does not speak like a normal, average, or common English user. For example, it says “stated plainly” or “worth stating plainly” (two very uncommon English phrases) every two or three responses.
The text also includes en-dashes, I struggle to believe that someone actually manually typed those.