How do online AI content Detector tools recognize AI content?

[jootess]

How AI content detectors like copyleaks can detect AI content?

 

[flprocess]

It checks if the last paragraph starts with "In Conclusion,"

 

[jootess]

It checks if the last paragraph starts with "In Conclusion,"

No!!!!!!!!

 

[Yohru]

Ok, so bit of a long answer. Text generation AI works by basically guessing what the next word will be based on context. Ai detection tools reverse engineer that. by basically writing the same word and noticing if the one after that is the expected one. Something like this image, for example.

If the ai detection tool notices that the text follows this pattern, it means the content is AI generated. Gives an score between 1 and a 100% usually, doing the math between which one was the next word and comparing how far it was from the expected one.

 

[jootess]

C

Ok, so bit of a long answer. Text generation AI works by basically guessing what the next word will be based on context. Ai detection tools reverse engineer that. by basically writing the same word and noticing if the one after that is the expected one. Something like this image, for example.


View attachment 245981

If the ai detection tool notices that the text follows this pattern, it means the content is AI generated. Gives an score between 1 and a 100% usually, doing the math between which one was the next word and comparing how far it was from the expected one.

Can you explain more?

 

[Yohru]

C

Can you explain more?

I can try.

Imagine this situation

Prompt: "Finish the sentence: I like __"

Chatgpt: "I like trains."

Now someone takes that sentence, "I like trains"; and runs it through an ai detection tool. What the tool is asking is how would an AI finish a sentence that starts with 'I like_' to another, different ai, which gives an chance from 1 to a 100 that an AI would finish a sentence like that. So for example:

Prompt: "Finish the sentence: I like __"

Chagtp (while processing):
 ...food 66%
...money 25%
...trains 8%
...human flesh 1%

Eventually chatgpt finishes evaluating the context and chooses the word "trains."

Chatgpt: I like trains.

What the evaluating ai is doing is telling how close the words are to the expected values. If you give an AI the sentence "I like" it will finish it with "food" 66% of the time. So if you see "I like food" that would be a tell of an ai generated text. It does this with everything single word of your text, analyzing what would be the expected words an ai would throw in given the previous context, then given a % of likeness the sentence is ai generated. After going through the whole text it does the math, normalizes it and tells you the chance any text is ai generated. It's given you a chance based on AI trends, not detecting anything. Basically its saying: yeah, this is something an AI would say.

 

[moonswamp]

It checks if the last paragraph starts with "In Conclusion,"

and more: "when it comes to..."

 

[Elanar Hammer]

through copyscape ?

 

[jootess]

I can try.

Imagine this situation

Prompt: "Finish the sentence: I like __"

Chatgpt: "I like trains."

Now someone takes that sentence, "I like trains"; and runs it through an ai detection tool. What the tool is asking is how would an AI finish a sentence that starts with 'I like_' to another, different ai, which gives an chance from 1 to a 100 that an AI would finish a sentence like that. So for example:

Prompt: "Finish the sentence: I like __"

Chagtp (while processing):
 ...food 66%
...money 25%
...trains 8%
...human flesh 1%

Eventually chatgpt finishes evaluating the context and chooses the word "trains."

Chatgpt: I like trains.

What the evaluating ai is doing is telling how close the words are to the expected values. If you give an AI the sentence "I like" it will finish it with "food" 66% of the time. So if you see "I like food" that would be a tell of an ai generated text. It does this with everything single word of your text, analyzing what would be the expected words an ai would throw in given the previous context, then given a % of likeness the sentence is ai generated. After going through the whole text it does the math, normalizes it and tells you the chance any text is ai generated. It's given you a chance based on AI trends, not detecting anything. Basically its saying: yeah, this is something an AI would say.

I got it thanks.
If we use rewriter tools like QuillBot, Google can detect that we have used ChatGPT for our orginal content?

 

[Yohru]

I got it thanks.
If we use rewriter tools like QuillBot, Google can detect that we have used ChatGPT for our orginal content?

From what I've seen, yes, you can get away with quillbot for now. I don't know long that will last, but as of right now that's the most common solution to the problem.

 

[ravid32]

From what I've seen, yes, you can get away with quillbot for now. I don't know long that will last, but as of right now that's the most common solution to the problem.

I wonder how long until an AI detection tool is available on the market to use by the public.

 

[thesyndicate]

The QuillBot did not do much the text came out as AI 100% as well. Any other suggestions on how to avoid that?

 

[MoonStarz]

That is a good question and I will ask with another one, what makes a script AI content, and I investigated a little bit, repetition, lack of personal oppinion, general information but when writing about a topic you can provide general details about a sub topic since is not relevant, so I believe is inacurate, when you do not know about something I usually provide general information and we tend to repeat a lot, I asked chatgpt if the content tha tis providing is provided by his own neural connection transistors don't know wherever or if it was from a database the response is that all the information provided is from a real article, book, I asked to provide me 100 words of no copyright novels and got no AI detection asked general information about articles immediately got AI detected? I believe those AI detectors are more copyright detectors get me wrong but I being testing with free content and it never gets detected as AI maybe I did something wrong but if you can prove me wrong will be really nice because I'm curious about this now.

 

[thesyndicate]

That is a good question and I will ask with another one, what makes a script AI content, and I investigated a little bit, repetition, lack of personal oppinion, general information but when writing about a topic you can provide general details about a sub topic since is not relevant, so I believe is inacurate, when you do not know about something I usually provide general information and we tend to repeat a lot, I asked chatgpt if the content tha tis providing is provided by his own neural connection transistors don't know wherever or if it was from a database the response is that all the information provided is from a real article, book, I asked to provide me 100 words of no copyright novels and got no AI detection asked general information about articles immediately got AI detected? I believe those AI detectors are more copyright detectors get me wrong but I being testing with free content and it never gets detected as AI maybe I did something wrong but if you can prove me wrong will be really nice because I'm curious about this now.

View attachment 256592

I guess if they have the same Chatgpt and the same connection they have the same info and see what they would have say in the same situation.

 

[Nikolaos]

Ok, so bit of a long answer. Text generation AI works by basically guessing what the next word will be based on context. Ai detection tools reverse engineer that. by basically writing the same word and noticing if the one after that is the expected one. Something like this image, for example.


View attachment 245981

If the ai detection tool notices that the text follows this pattern, it means the content is AI generated. Gives an score between 1 and a 100% usually, doing the math between which one was the next word and comparing how far it was from the expected one.

Is that also how google does this?

 

[sageshark]

I do not trust AI detection tools because in many cases they give false positive. At times they have marked my human written content as AI generated so I have stopped using them.

 

[analytic]

There is an "AI content detector" on rapidapi - you can test how this mechanism works for free

 

[KindleHD]

I do not trust AI detection tools because in many cases they give false positive. At times they have marked my human written content as AI generated so I have stopped using them.

Yep it has no correlation to google bro or rankings. Not sure why people even use it. You’re using Ai Content , that’s the fact

 

[Elvlin]

I do not trust AI detection tools because in many cases they give false positive. At times they have marked my human written content as AI generated so I have stopped using them.

yupo agree, ai detection tools, is basically just like witch hunt. kill all randomly, there will be ton of real human written article that market as ai.
in the end make people that write real article, or even create ai article run it on those detection tool to give result it want.
just like plagiarism checker, it make the whole point pointless. since in the end it jus make people article to certain form that pass this program. not it real purpose

 

[analytic]

In my opinion, the detectors work well. Since the text created by openai - it turns out sucked from the finger. The quality of openai rewriting is higher, but it is also determined