We shouldn't be too quick to jump to "AI did it." People write redundant paragraphs and sentences in articles all the time because they're led to believe that every article needs a conclusion that sums up what's already been said. Ironically, including one in this article showed a lack of good judgment, which isn't confined to AIs.

It does show an interesting second order downside of publicly using an LLM for anything: it raises everyone else’s suspicion that the rest of your work is LLM generated.

A certain amount of redundancy can be helpful. If you explain the same thing two different ways, one way might make more sense than the other to a particular reader. The judgement comes in because you should not just be doing this by rote or because you were "lead to believe" it was necessary, but in consideration of who you are writing for.

Where do you think LLMs learn to write “uselessly redundant conclusion”?