I just read a Quanta article from 2015 and literally had to check the date after the third "X not Y" I encountered. But the models have learned that from somewhere, too.
Why do you think AI "talks" like that? It is pervasive in a lot oof the training material.
Most people writing scientific literature are not English majors nor even great communicators, their primary competencies are elsewhere. Expecting everyone to write literature is just odd & reductive.
Yes, it helps to be a good writer, but the primary objective for science papers is to convey information; that it is well written or enjoyable is a nice bonus but not essential.
I'd counter that AI usually writes way better than typical person, definitely way better than any commenters here (myself included).
What's jarring about AI content is seeing the mastery of English applied in mundane contexts, where people are used to see more basic, pedestrian language.
(That, and overuse of patterns, which is more of an artifact of how people use LLMs - specifically that 90%+ of the AI output you see is the first response to the prompt in a fresh session. For AI, each time is the first time it's using this pattern, for user it's the 10th time they saw it within the span of an hour.)
Agreed except for that last bit. Even if the session drags on endlessly the AI keeps using the same repetitive speech patterns (at least IME). Not only that but they stand out to me in that they don't match the nature of the conversation. It's a bit like receiving highly polished sales copy in response to an offhand forum post.
