Recognizing AI Writing - jmadison222/knowledge GitHub Wiki

The discussion here is inspired by the following video. And the bullet list is taken almost entirely from it, then mixed in with my own thoughts, particularly my grouping and ranking around the bullet items. The vido is here:

• I Can Spot AI Writing Instantly — Here’s How You Can Too - The video source for the list that follows.

Great thanks to Evan for his insights and production. Please leave a like on the video if you find it useful.

These are problems I believe will be inherent in the AI for a long time. They stem from the AI having no direct physical understanding of the world—​only what it can infer from text or from physical sensors that are far behind our senses.

• No Personal Specifics - The AI almost never tells you personal specifics because it has none. Humans not only have them and tell you, but they will often lead with the personal aspects. My son and I rode on the train into Boston. His favorite part of the ride was all the balls he saw on the side of the tracks where kids hit them in but couldn’t retrieve them. The AI would likely tell you how many millions of riders per year that train carries.

• Bad Metaphors - Mixed metaphors are the black sheep of the literary toolbox that derail the train of thought before it even leaves the harbor. The AI mixes metaphors because it doesn’t understand the subtle nature of the interaction between the literal truth and implied meaning. They’re not usually as bad as the first sentence, which I got from ChatGPT by asking "what’s an example of a mixed metaphor about mixed metaphors", but they are frequently subtly off.

These are problems that I believe can be fixed in the near-term years with a bit of fine tuning. Even just getting rid of so many em dashes feels like a simple enough programming fix. Same for overuse of triples. Etc.

• Parallel Sentence Structure - Not only is AI not intelligent—​sometimes it hallucinates terribly. See that sentence. It’s parallel structure. Read it again. Feel the cheesy salesman feel to it. Humans rarely write that way. The AI uses it frequently, often at the start of a response.

• Praising the Reader - If you’ve read this far into my article, you clearly have a great interest in AI and are doing a wonderful job investigating it so you can understand it for yourself. See what I just did there? Why am I kissing your butt? I don’t really mean it. Neither does the AI since it doesn’t really mean anything. But it loves to butter you up.

• Use of Triples - Recognizing can be done if you study the writing style carefully, read closely, and detect the several signals. See that? That’s a triple. It is true that triples are powerful. That is backed by research. Humans should use them occasionally. But the AI uses them constantly.

• Big Words - You should abstain from the inclination to employ grandiloquent or ostentatious vocabulary in lieu of clear articulation of your ideas. That is, don’t use too many big words. Humans may have some at their disposal, but tend to write without them unless only the big words will work. AI tends to use bigger words than are needed and often in slightly awkward ways. (I used AI on that first sentence—​it’s very useful when you ask it to profess its own guilt!).

• Empty Words - TODO.

• Use of the Em Dash - TODO.

• Repetition - TODO.