A fairly good and nuanced guide. No magic silver-bullet shibboleths for us. I particularly like the observation that in reality LLMs are less likely to use unusual words, because they are statistically less likely.

Consequently, the LLM tends to omit specific, unusual, nuanced facts (which are statistically rare) and replace them with more generic, positive descriptions (which are statistically common). Thus the highly specific "inventor of the first train-coupling device" might become "a revolutionary titan of industry." It is like shouting louder and louder that a portrait shows a uniquely important person, while the portrait itself is fading from a sharp photograph into a blurry, generic sketch. The subject becomes simultaneously less specific and more exaggerated.

Doing a quick search, it hasn't been posted here until now - thanks for dropping it.

In a similar vein, there's a guide to recognising AI-extruded music on Newgrounds, written by two of the site's Audio Moderators. This has been posted here before, but having every "slop tell guide" in one place is more convenient.

Although I never use LLMs for any serious purpose, I do sometimes give LLMs test questions in order to get firsthand experience on what their responses are like. This guide tracks quite well with what I see. The language is flowery and full of unnecessary metaphors, and the formatting has excessive bullet points, boldface, and emoji. (Seeing emoji in what is supposed to be a serious text really pisses me off for some reason.) When I read the text carefully, I can almost always find mistakes or severe omissions, even when the mistake could easily be remedied by searching the internet.

This is perfectly in line with the fact that LLMs do not have deep understanding, or the understanding is only in the mind of the user, such as with rubber duck debugging. I agree with the "Barnum-effect" comment (see this essay for what that refers to).