Anyone who claims that they can reliably detect LLM text is lying (probably to themselves). People in general are terrible detectors for this.

Verbatim copies are often not that hard to detect in a large number of cases; there is a tendency to waffle excessively in a way that few people do. This is of course not a fool-proof method, but it's "reliable" in the sense of "you can get it right more often than not".

Edited LLM text is much harder: using a LLM to generate some text and then edit it in shape (often by removing extraneous paragraphs, maybe rewriting a few things slightly). Those are basically impossible to detect reliably.

I don’t disagree with you, though personally I’ve called out two people on Reddit using ChatGPT to farm karma by pretending to be an expert on some topic… and an hour later they deleted all of their GPT generated replies from their profiles.

There’s definitely a pattern for some prompts, where the model uses an obvious format of:

  - Make generalized statement answering prompt question
  - Support that statement with 3 or 4 discrete paragraphs that don’t exhibit any personal experience, give examples, or cite statistics
  - Finish off with a high-school essay style conclusion statement reiterating the introduction with different phrasing, generally beginning with “Overall” or “In conclusion”

It’s also comically obvious when suddenly someone who has a reply history full of low-effort juvenile, zero grammar, curse laden posts on video game and meme subs is suddenly writing mini-theses on topics ranging from Swedish forestry management to the chemistry of dyes used in Cambodian textile manufacturing.

There's definitely a "what are the benefits of" or "what are the reasons for" format that ChatGPT falls into pretty easily. It's usually not really wrong and would probably serve as a passable high school essay. (And I've probably seen worse marketing copy.)

But it's not very good and, at a minimum, lacks nuance and supporting evidence.

The key word from GP is "reliably". Sure, you can probably tell when someone blatantly copied and pasted output from ChatGPT and when the prompt was not very sophisticated (write a comment about X). However for all other cases when someone put even the tiniest bit of effort into the prompt or did some post filtering the reliability of such detection plummets significantly.

Overusing them can be a sign, though obviously it’s not proof

Not really. I think people don't understand that this is kind of chicken and egg problem.

You don't see overuse of linking words because it was generated by LLM. You see them, because every non-native spear is literally taught to link every paragraph with them. And all the texts, blog posts, wikipedia articles, stackoverflow responses and all the other stuff they wrote was then used as training data, from where LLM had learnt to do the same.

What I am saying is that there are just much more non-native English speakers and LLMs are inherently kind of non-native speakers too. So a sign that distincts majority (non-native) speakers from the minority (native speakers) is actually a bad sign (: