How to destroy your own AI writing tool

[WPRipper]

I've read an article written by Justin McGill, founder of Content at Scale, in which he explained some things about AI writing.

At some point, he wanted to demonstrate how "well' the tool is performing and linked to an article - link here

Now, on the website, there's an AI content detector - link here

If you paste the article promoted as an example, you get a 100% fake score.

I have my doubts about these detectors, but we can ignore the fact that the written text from your own tool is showing that 100% fake score.

And this is how you basically destroy your reputation.

-------------

This post was not written by AI.

 

[IM Dude]

i just pasted an ai article i generated today and here is the score

Predicted based upon 425 words.
100% Real 0%Fake

So its a crappy detector.

 

[devinda]

today i generate 10 article

Predicted based upon 550 words.
35% Real 65%Fake

how to fix it?

 

[Dot Mirror LTD]

today i generate 10 article

Predicted based upon 550 words.
35% Real 65%Fake

how to fix it?

Use more detailly prompt

 

[Site Arrow]

The quality of the AI text detectors will improve over time, and so articles that previously passed will start to be flagged by those detectors.

I'm guessing that is what happened here.

 

[randomstrings]

If grammar is perfect, with lot of fancy words AI tool detection will say it AI content, coz humans are dumb,

 

[Anne-Maree]

so what's the best rewirte tools for chatGPT AI content

 

[devinda]

so what's the best rewirte tools for chatGPT AI content

yes im looking to

 

[BowtiesAreCool]

yes im looking to

I find if you just enhance your prompt in CGPT, you can get a 100% Real score. I did it yesterday and no issues. Lots of people like Quillbot as 1-2 punch but I think you just need to work the content a bit more.

 

[malayan]

I once tested a tester and it said my homepage text was 100% AI... So I deleted the words "GPT" and "OpenAI" and retested it and it said it was 5% AI.

These things mean nothing.

- Sincerely
Not GPT-3

 

[zigzagzig]

Pasted my whole AI Generated article. It says it looks great. LOL.

Well, It does look great.

 

[goatsinboats]

Yeah, I'll stick to using AI detector models found on huggingface, copyleaks (a company who specializes in plagerism detectors for EDU) & the recently released AI detector on openai.com:

https://huggingface.co/spaces/openai/openai-detector

https://platform.openai.com/ai-text-classifier

https://copyleaks.com/features/ai-content-detector

(You have to log-in to your openai account to use the text-classifier, also it just says whether or not it thinks it's probable that the text was AI or Human, no actual percentage or indication what part of the text was "detected")