The problem I see with this-AI-generated versus human-written isn’t a binary, it is a continuum.

One person gives an AI a brief prompt, the AI writes a whole novel, person publishes it without even reading it first

Another person spends weeks tinkering with prompts, producing dozens of outputs for the same prompt and deciding which to keep and which to cut, editing numerous AI outputs together - that’s still partially AI-generated, but with vastly more human input than the first case

A third person does all the writing themselves, but uses an AI for review, copyediting, as a source of ideas or suggestions, as a brainstorming partner… maybe the AI suggested a few turns of phrase here and there, or gave them some story ideas

Where do you draw the line?

No AI period. As soon as you feed your writing into the slop machine it starts telling you how to make it more like slop (I know someone who's cowriting a book with ChatGPT - this is exactly the result.)

I'd rather read things with typos and bad grammar than read something copyedited by AI.

Can I use the sugestions in Google Docs? Is that an expert system or LLM?

If the machine can predict your writing, then so can the reader. Which means the reader is getting bored.

Take the Google Docs suggestion as a sign that you shouldn't be writing that sentence at all. Back off, and find a way to tell the story in a way that every sentence provides something new, exciting, and unpredictable.

If the suggestion is better than what you were going to write, take it as a sign that you're not ready to publish. Treat it as a lesson in how to write better, then start over when you're more skilled at the craft.

Much the same applies to all LLM writing. If it can write it, it probably shouldn't be written at all. If an LLM is writing your boilerplate code, it means that there's too much boilerplate in your system. Solve that not by letting somebody else write it, but by rebuilding the system so that it doesn't require boilerplate.

My errors are mostly missing s in verbs, like "The dog eat[s] a lot of meat." , also wrong prepositions like "I'm still thinking [in->about] the exam." and sometimes idioms. I like it [the free version?] as an advanced orthography/grammar corrector for details, not a whole rewrite.

I agree with you. I remember a recent discussion where someone "cleaned" the comment of other user using AI, https://news.ycombinator.com/item?id=47201796 the cleaned version was soulless or as the comment says wooden.

EDIT: From a recent comment by myself:

> People here is friendlier when the author is in the comment section replaying questions.

This time I notice the error on my own, but I made similar silly mistakes in the past. Google Docs may show a line with a suggestion to fix it.

I hope this is successful. Fake books are flooding online stores and if this was widely adopted I would certainly refuse to buy anything without it.

But unfortunately I think that's unlikely. Most authors will likely never hear about this. I assume there will be some kind of fee to participate, but this will discourage people from using it. And even if it takes off, the fake book authors will just slap it on anyway as I'm not sure what enforcement mechanism would be effective.

Looking to create an "AI detector" is like asking which organelle in a cell contains the "soul". It's a fool's errand and uncomfortable truth that there is rarely a magic watermark of any sort. Honesty and integrity in authorship is a social problem.