Thoughts on AI checkers

[MetaBob]

I will repost what I posted a week ago about my own experience:

I recently got my AI-rejected chapter accepted after two failed tries. Happy to share what worked and didn't work, for me.

First try was to simply resubmit with a note. It sat for a couple of weeks and then got rejected with no further info.

Second try was a light edit for the items mentioned here as generic AI flags. Things like rules of three, numbering, etc. Again, that took a few weeks and same outcome.

Finally, I did what I was told NOT to do by most people here. I used AI to actually get around the problem. Here is what I did:

1. I used two AI detector packages (humanize and gptzero) to flag AI trigger sentences. I went to my chapter and highlighted those.
2. I then asked two AIs (ChatGPT* and grok*) to identify the paragraphs/sections of my chapter that they thought were most likely to be AI products. I highlighted those.
3. I had very little confidence in either of the two above methods, but I figured, whatever was causing the flag, it was probably in there somewhere. At this point, nearly half my chapter was highlighted.
4. I simply deleted whole sections that had more blue than white, and then wrote those sections again from scratch.
5. Then I edited the whole chapter again, top to bottom, since it was now quite uneven.
6. I then edited again, and again, until I felt it was BETTER than the original I had submitted, since I was pissed off that I had probably butchered it in the process.

So, I submitted that one, with a note saying I had rewritten half of it from scratch. It was accepted in 24 hours.

** I should add that I buy the premium version of grok and the super-premium version of ChatGPT for work, so YMMV if you are using the free versions. They are a lot less capable.

An excellent and highly practical strategy: when nothing is certain, use what tools with proven merit are available. Which is exactly what I advised 6 weeks ago:
https://forum.literotica.com/thread...-story-anymore.1648707/page-12#post-102275280

 

[metropolinational]

Thanks, @MetaBob. You and I had a pretty vicious argument in that thread, but in the end, the advice you gave @xannatharr was what led me to follow this strategy. So a huge thank you for proposing something that might have worked (at least for me)!

 

[MetaBob]

Thanks, @MetaBob. You and I had a pretty vicious argument in that thread, but in the end, the advice you gave @xannatharr was what led me to follow this strategy. So a huge thank you for proposing something that might have worked (at least for me)!

I don't know that I would use the word "vicious" to describe our interaction there (I don't do vicious and I didn't get the sense that you did, either), but I've been on the Internet for long enough that I recognize what happens in fora where people aren't familiar with each other and can be quick to take offense. I try to stay on topic, to keep a high signal/noise ratio in what I post, recognize that not everyone is the same, and stick to what's verifiable and/or constructive. I saw what you wrote in response to my advice to @xannatharr, which was to take my advice to him, which was honestly the same as I'd offered to everyone else. I would have taken you off ignore before long even if I'd gone through with doing so in the first place, which I never did. Mostly, I prefer to not get dragged into argumentative noise, and admit to getting annoyed by it sometimes, but life's too short.

This whole "AI" theme gets raised here on too regular a basis for my taste, with the same inane "insights" offered each time. I seldom look into threads devoted to it. It's good to see when somebody follows the practical path through their experience.

FWIW, I finally submitted what I offered up as a before/after example (original human-written text vs the same text when run through an AI "rewrite") here. The human-written original, of course. It was accepted, just like everything else I've ever submitted here:

https://www.literotica.com/s/travel-nurse-or-the-final-ward

I feel fortunate to have never been subjected to an AI-related rejection here, but I imagine that it's only a matter of time.

Edit: Disclosure: I can get impatient as well as annoyed.

 

[metropolinational]

I try to stay on topic, to keep a high signal/noise ratio in what I post, recognize that not everyone is the same, and stick to what's verifiable and/or constructive.

No wonder you inspired my ire!!

Heh. Jk of course.

thanks again!

 

[MetaBob]

No wonder you inspired my ire!!

Heh. Jk of course.

thanks again!

I don't always succeed.

 

[jsmiam]

Without disagreeing with other advice and theories, I’ll share one that can happen independently of others. AI checking may look for individual sentences that are suspected of being AI. Too many and/or too high a percent of suspect sentences may be a factor.

Which would answer the claims of quite a few people who swear up and down they didn’t use AI, but who accepted suggestions for sentence completion from their grammar checker. So maybe they really didn’t use AI to compose their story. But nonetheless it has numerous sentences completed by AI.

And it’s not limited to grammarly. People need to turn all AI of in their word processors, browsers, and device settings.

This doesn’t mean you can’t let grammar checkers point out issues to you. But it does mean fix it yourself with your own brain and fingers. And don’t let the grammar checker’s writing style sink into your brain.

 

[EmilyMiller]

Without disagreeing with other advice and theories, I’ll share one that can happen independently of others. AI checking may look for individual sentences that are suspected of being AI. Too many and/or too high a percent of suspect sentences may be a factor.

Which would answer the claims of quite a few people who swear up and down they didn’t use AI, but who accepted suggestions for sentence completion from their grammar checker. So maybe they really didn’t use AI to compose their story. But nonetheless it has numerous sentences completed by AI.

And it’s not limited to grammarly. People need to turn all AI of in their word processors, browsers, and device settings.

This doesn’t mean you can’t let grammar checkers point out issues to you. But it does mean fix it yourself with your own brain and fingers. And don’t let the grammar checker’s writing style sink into your brain.

Many writers have reported using Grammarly and publishing just fine. The difference between them and those who get pinged for Grammarly use is precisely what you highlight. Using Grammarly to identify issues which you then fix, vs accepting Grammarly’s suggested fix.

As you suggest, Word autocomplete could also be problematic. Turning off Copilot in Word is a good idea (though the next update will probably turn it back on).

 

[big_cane_sugar]

Any simple thing we can think of to make a story look like it's not AI can be fed into an AI's algorithms as instructions.

The world is getting pretty crazy now. I have a student who has learned to write from her interactions with AI. I assumed she was just adding some errors to an AI-generated text, but then she proved to me that she really did write that way by having me sit next to her as she typed a short essay on a prompt I'd given her.

I don't know where we're going from here, but the lines are probably going to get fuzzier before they get clearer.

 

[EmilyMiller]

I don't know where we're going from here, but the lines are probably going to get fuzzier before they get clearer.

I think some of our - understandable - concern is this mantra that LLMs will get better and better exponentially. That’s what the AI companies tell us, so it must be true, right?

But what if LLMs are a dead end technology? What if the pace of their improvement is already slowing and the main use cases (CEOs firing people they’ll have to rehire in two years, students cheating, and a lazy summary of web-content) have already emerged? What if the AI companies are desperately trying to monetize their white elephants before anyone else notices?

 

[metropolinational]

What if the pace of their improvement is already slowing and the main use cases (CEOs firing people they’ll have to rehire in two years, students cheating, and a lazy summary of web-content) have already emerged?

this is almost certainly true. but these chickens have not come home to roost yet. the students becoming idiotic zombies is just hitting. the ones with 3-5 years of AI exposure are entering college or the workforce now. they are essentially disabled. wait til the ones with 5-10 years hit us. then the ones with 10+ years hit us.

we only need incremental improvement in self driving technology from here on out for 10 million people to lose their jobs over the next decade. same for robotics. I think all these technologies are coming together and following the diffusion of innovation model, whether there are more breakthroughs or not.

in fact, the lack of breakthroughs will help them get adopted. there are numerous historical examples of this, like the lightbulb.

 

[EmilyMiller]

we only need incremental improvement in self driving technology

I was specifically referring to LLMs

 

[metropolinational]

I was specifically referring to LLMs

I get that. I was just referring to another tech where we are already screwed, whether more major innovations hit or not.

 

[MetaBob]

Without disagreeing with other advice and theories, I’ll share one that can happen independently of others. AI checking may look for individual sentences that are suspected of being AI. Too many and/or too high a percent of suspect sentences may be a factor.

The ChatGPT AI detector that I've linked in this forum on several previous occasions includes a section on the top 100 phrases that it thinks occur MUCH more often in AI-generated writing than in natural human writing, from 22x to 182x as often:
https://gptzero.me/ai-vocabulary

Keep in mind that this seems to be more aimed at business or academic writing than at the sort of creative writing we do here, but it's another thing to watch for; you might want to avoid using these phrases in your human writing so you don't trigger a false alarm from whatever AI detector this site uses.

Edit:
1. I don't think I've ever used any of these 100 phrases in my writing in any context.
2. I only found one phrase in this set of 100 that I think might work in creative writing: "the journey begins".
3. "A serf reminder"? I didn't even know what that means.

 

[metropolinational]

@jsmiam and @MetaBob, I think maybe what you are both getting at is a quantity vs threshold issue. any good algorithm can be tuned to flag some number of sub-threshold events over a defined expanse of text.

when I applied bob's suggestions, I marked whole sections for deletion that had too much blue. I think this is one area where something like grammarly can create problems. Any one of those corrections may not be a problem, but if you accept a whole paragraph's worth of suggestions, it might be enough to trigger a flag.

Frankly, one of the reasons I don't use grammarly is because I LIKE the ridiculous way in which I write. Grammar software flags my writing constantly - too long, too over the top, calm down, consider turning this paragraph into a single sentence, blah blah. I bet if I were to follow all of its advice, I would soon start accepting the neoliberal corporate opinions of the New York Times and Wall Street Journal economists as true because "the experts" said so. Best to stay away from that.

Anyway, what was I talking about?

Yes. If you accept all the grammar suggestions, I bet the issue that gets flagged is that your writing gets bland, and AI is bland. I think that blandness could be a flag too.

 

[MetaBob]

I'm gonna dissect this. Not because I'm nitpicking, but because you make a lot of points that I think need separate treatment.

a quantity vs threshold issue.

I guess? My goal in posting on these AI-related threads is to help authors navigate these stupid damn AI-related rejections that too many authors are getting. "quantity vs threshold" goes deeper than I think it needs to, and/or deeper than I want to think about.

any good algorithm can be tuned to flag some number of sub-threshold events over a defined expanse of text.

I guess?

when I applied bob's suggestions, I marked whole sections for deletion that had too much blue. I think this is one area where something like grammarly can create problems. Any one of those corrections may not be a problem, but if you accept a whole paragraph's worth of suggestions, it might be enough to trigger a flag.

I don't use Grammarly that way. I use the free browser-based Grammarly on their website as a final or almost-final checkpoint, because it's better at catching some kinds of errors than my WP programs or compulsion to agonize over every word, especially doubled doubled words. I don't let Grammarly change ANYTHING beyond single misspelled or doubled words. Truth is, I don't have access to any of those suggestions since I don't have a paid subscription; their suggestions are blurred-out, which is fine by me since I'm not gonna let an AI touch my (sacred!) text anyway.

That said, if the gptzero site flagged significant chunks of your writing as potentially having been written by AI, that's exactly what I was hoping it would do. Pay attention to those sections. Reword or rewrite, re-submit to that site, repeat as often as necessary, then resubmit with a note to Laurel, and now you're getting somewhere.

Frankly, one of the reasons I don't use grammarly is because I LIKE the ridiculous way in which I write.

Amen, brother.

Grammar software flags my writing constantly - too long, too over the top, calm down, consider turning this paragraph into a single sentence, blah blah.

I think 90-95% of what Grammarly gives me, in the limited way that I use it, is crap, and have said so many times before. But that other 5-10% is stuff I hadn't caught before I finally threw it at the Grammarly wall to see what might (not) stick. So I include Grammarly in my process, and might include individual word choices it recommends, but (see above). I have NEVER included ChatGPT in any part of my process, I have only mentioned it in this forum because it includes an AI detector that might help other writers get past pesky AI-related rejections.

I bet if I were to follow all of its advice, I would soon start accepting the neoliberal corporate opinions of the New York Times and Wall Street Journal economists as true because "the experts" said so. Best to stay away from that.

Anyway, what was I talking about?

Yeah, you're in weeds that I will ignore. Hope you don't mind.

Yes. If you accept all the grammar suggestions, I bet the issue that gets flagged is that your writing gets bland, and AI is bland. I think that blandness could be a flag too.

Right now, yes, everything I have sent to an AI rewrite has come back "bland". That's exactly the word I have used to describe it. Another word I have used in characterizing these AI rewrites (all three of which I have immediately discarded with extreme revulsion) is ... "shit".

 

[metropolinational]

But that other 5-10% is stuff that I hadn't caught before I finally threw it at the Grammarly wall to see what might (not) stick.

I think that 5-10% is important. it's just like bacon, butter, LSD, drinking at work, cocaine and unprotected sex with strangers. in moderation, it's ok. but got to that well too often, and then you might find yourself in trouble.

nothing wrong with using it as a flag, but some people feel the need to make all the red underlined sentences go away. thats when you can start accumulating some damage.

 

[metropolinational]

I have NEVER included ChatGPT in any part of my process, I have only mentioned it in this forum because it includes an AI detector that might help other writers get past pesky AI-related rejections.

it was part of my four-tool approach. I agree that individually, its not good. in fact, I used four tools because I had no faith in any of them. but collectively, maybe they helped.

 

[MetaBob]

it was part of my four-tool approach. I agree that individually, its not good. in fact, I used four tools because I had no faith in any of them. but collectively, maybe they helped.

FWIW, I think Grammarly's AI detector is completely useless. Then I tried gptzero, and it seemed to yield more sensible results. Not that I have ever had to use either for my own writing here, thank goodness.