Interesting. It only seems to work in English. English texts that it successfully detects as AI generated are classified as human generated if you ask ChatGPT to translate them into a different language (or make the initial prompt in a different language). Perhaps they want to clarify this limitation, either through an FAQ or by stating that it doesn't support it when other languages are supplied in the prompt.
Detect ChatGPT Generated Content
41–50 of 109 posts
Re: Detect ChatGPT Generated Content
#42The accuracy just isn’t there, which isn’t surprising. I’m feeding it Star Wars reviews from both ChatGPT and IMDb, and for sure the answers are correlated. The false negative rate is OK, although I’ve hit plenty of them. But man, it sure does think a lot of people were using ChatGPT to write reviews back in the 2000s.
Again, it’s definitely correlated, even strongly correlated, but that’s not good enough for plagiarism detection. You can’t go accusing students of academic dishonesty based on a tool that gets it wrong multiple times on a couple dozen samples.
Re: Detect ChatGPT Generated Content
#43The problem with such things is that you would need 100% accuracy in most cases for it to be useful. For example many schools and universities fear that students use ChatGPT for homework. If such a plagiarism checker has false positive results in just a small percentage of cases, the consequences for honest students would be too severe to actually act on the results of the check. But perfect accuracy can never be rea…
That’s a very naïve outlook in reality like with every system the burden of proof would end up falling on the accused not the accuser…
What could that burden of proof look like?
Not sure how a student could prove that the negative, apart from videotaping themselves writing the essay, or only being allowed type it on an airgapped machine owned by the university.
Doesn't feel very feasible to implement.
Re: Detect ChatGPT Generated Content
#44The problem with such things is that you would need 100% accuracy in most cases for it to be useful. For example many schools and universities fear that students use ChatGPT for homework. If such a plagiarism checker has false positive results in just a small percentage of cases, the consequences for honest students would be too severe to actually act on the results of the check. But perfect accuracy can never be rea…
That’s a very naïve outlook in reality like with every system the burden of proof would end up falling on the accused not the accuser…
The reality in my experience (as a university lecturer) is that the burden of proof very much falls on the accuser rather than the accused.
Re: Detect ChatGPT Generated Content
#45So, what prevents chatGPT to use detection tool and fine-tune its response accordingly?
Nothing, this is how adversarial training works, but it also works both ways.
It works both ways, but generation is advantaged in the long run. There has to actually be a statistical difference to detect, and AI outputs without statistical differences from human output are obviously possible, since humans make them all the time.
Re: Detect ChatGPT Generated Content
#46The problem with such things is that you would need 100% accuracy in most cases for it to be useful. For example many schools and universities fear that students use ChatGPT for homework. If such a plagiarism checker has false positive results in just a small percentage of cases, the consequences for honest students would be too severe to actually act on the results of the check. But perfect accuracy can never be rea…
> The problem with such things is that you would need 100% accuracy in most cases for it to be useful. Why? If there is 75% confidence a students report was generated using ChatGPT, then that’s enough to sit down with the student and discuss the content in person and see if they actually know it. A tool such as this could help the teacher having to avoid doing this with every student, and also reinforce to students t…
This is not what is going to happen tho. Flagged content will not be accepted and the student failed for the class, or worse for the year. Expulsion is on the card as the war against generated content becomes bitter.
Re: Detect ChatGPT Generated Content
#47Earlier quoted context omitted.
That’s a very naïve outlook in reality like with every system the burden of proof would end up falling on the accused not the accuser…
What could that burden of proof look like? Not sure how a student could prove that the negative, apart from videotaping themselves writing the essay, or only being allowed type it on an airgapped machine owned by the university. Doesn't feel very feasible to implement.
and if the student is not a complete tool, he can definitely learn the output and replicate it very, very closely
Re: Detect ChatGPT Generated Content
#48Here’s and alternate approach to detecting GPT that has an actual benchmarks and research paper available:
Re: Detect ChatGPT Generated Content
#49A nice approach - estimates the randomness in text. I tried with various cases - blog posts, wikipedia, scientific docs, and with a few examples from chat, and it figured it out accurately. Respect. Even though I disagree with a need for such tools - it doesn’t matter if content was written by a human or by a machine. What matters is whether it’s easy to read and worthwhile.
Disagree, this is very much needed. The problem with AI-generated content is that it looks superficially worthwhile and plausible, but in fact it often says things that are not correct.
What I find fascinating is that aspect of AI-generated content actually captures what humans do all the time, confidently making incorrect statements, "bullshitting" with filler text that only pretends to be meaningful, making illogical statements that contradict what was stated earlier, etc.
Re: Detect ChatGPT Generated Content
#50I just fed it an email I'd personally written.
The start and end were apparently written by AI.
What does this mean? Is this proof we're in a simulation and I'm actually a mouth piece for an AI that's reached the singularity?