Practical advice for dealing with AI rejection.

[tripleflip]

I agree, as I dont have any of these in mine, and it still gets flagged, I resubmit a few times and it goes through. I wonder if something else is the main cause of this. I.e. the engine gets overwhelmed and starts to spam reject. ? Who knows because nothing comes out of HQ as an explanation.

 

[tripleflip]

How not to look like A I.


I can't say which I find more stupid; using A I. to write a story, or having to comb through a thick fog of bad writing to find evidence of A.I. to use as an objective basis to reject it. Whoever does the latter deserves to be sainted.

Writing is supposed to be a kick. If you let a machine write your story for you, you’re letting a machine get your kicks for you. Unlike human authors, machines don't get kicks. This one fact perfectly explains why they don't write well.

I am a university professor. My field is Low Dimensional Topology, not A I. But my PhD student’s dissertation is on the application of topology to A.I. I am supposed to be her mentor, but she is way smarter than I am. I try to keep up. She works in a 60-dimensional Hamming space. Every word in a story is a vector in this space. Every story is a peculiar arrangement of words, or vectors. in other words, a "solid object" (tensor) which has a distinct “shape,” This 60-dimensional “shape” is not something a human can usefully visualize, but the mathematics of tensors are well understood.

Large Language Models produce story "shapes" recognizably different from the story shapes typical of.human authors. It is easy to show mathematically that human authors using auto-complete, spelling and grammar checking algorithms also tend to produce story shapes different from those typical of human authors not using those aids A statistic my student has developed is the “innovation variance" in a story. Roughly speaking "innovation variance" is the answer to the question: “Given the story so far, how unexpected is the next word the author chooses?”

LLM’s tend more often than not to choose "safe," "low risk,” or "expected” words, from clichés or memes (low innovation); whereas good human writers tend more often than not to choose “high risk” or “unexpected" words. (high innovation)

It seems that to avoid looking like A I. a writer should turn off auto complete, avoid clichés and stale memes and use imaginative word choices. In other words, write well.

Write like a human!

May your words flow easily, as though you were a human getting his, her, it's or their kicks!

Tripleflip

 

[SkyBubble]

Vary your sentence length. Short is good. Sometimes, though, you need to go longer, espouse on what you're describing, dig into the feeling of the moment to get to the core of the character's motivation. Consistent sentence length is not only boring, it's what AI does, so don't do that.

The same goes for paragraph length. Sometimes a one-sentence paragraph is all you need. Sometimes it runs over several lines. Try not to make them too long (a lot of people read on a phone) but do vary the length.

 

[SkyBubble]

In the core, I'd add the advice to avoid the em-dash, as those are known triggers for detection algorithms.

I agree with this. From conversations with LLMs on other topics not related to writing, I've noticed that they LOVE the em dash. It triggers AI detection because all those sites overuse it. I'd also advise leaving space before and after the dash (presumably an en dash if you must use one), as AI usually jams the words together on either side of the dash. (And use those sparingly. If a comma or a colon or some other mark will do the trick, use that.)

AI: that guy---the one she met last week---
A different way: that guy -- the one she met last week --
Or: that guy (the one she met last week)

 

[BeechLeaf]

Grr. If you're actually writing it, and you actually want to use an em dash, use it. Don't let air-raid warnings scare you into changing your style. If you're okay at writing, you're distinctive enough not to be flagged as monotonous.

 

[SkyBubble]

Grr. If you're actually writing it, and you actually want to use an em dash, use it. Don't let air-raid warnings scare you into changing your style. If you're okay at writing, you're distinctive enough not to be flagged as monotonous.

It's just a little safer to use an en dash.

 

[SkyBubble]

I'd add the advice to avoid the rule of three. An example: "This guide is solid, helpful, and comprehensive."
The AI often writes sentences in such a way.

I find that advice solid, helpful, and comprehensive.

 

[SkyBubble]

I'm sorry and I don't want to turn this thread into an argument (it probably will become one with or without me) but I don't believe any of these writing style mechanics issues are at all responsible for whether a story is flagged as AI generated or not

Then what do you think is responsible?