ChatGPT Detection App Built by Princeton Student

[irisXE]

"The app was so popular that it crashed "due to unexpectedly high web traffic," and currently displays a beta-signup page. GPTZero is still available to use on Tian's Streamlit page, after the website hosts stepped in to increase its capacity."
Source:
https://www.entrepreneur.com/busine...t-detection-app-to-fight-ai-plagiarism/442253
Source:

https://twitter.com/x/status/1610067688449007618

The websites don't load on my end:
http://gptzero.me/https://etedward-gptzero-main-zqgfwb.streamlit.app/

 

[praetserge]

well... how can we test it if the site is down?

 

[MoonStarz]

This is really interesting, it will be nice to test it but is unfortunate that a tool to make everything easy now has a nemesis well the happinest lasted just a little bit.

 

[moonshine7000]

All you need to do is to spin your essay and proof read it.

 

[Deleted member 1501362]

All you need to do is to spin your essay and proof read it.

Exactly, and don't be lazy.

 

[irisXE]

All you need to do is to spin your essay and proof read it.

Exactly, and don't be lazy.

I tried spinning numerous articles created by ChatGPT. I used Quillbot and Wordtune and they still get detected. Sometimes the articles had even higher AI content percentages after running them through spinners. The thing is that those spinners are also AI's, therefore I don't think they will help.


LE: https://etedward-gptzero-main-zqgfwb.streamlit.app/ works for me now

 

[Jev_bh]

If he has built that tool and detect plagiarism, that means google will be able to soon detect people that have used ChatGPT to right articles. I take it this would give you a negative SEO score…?

How the hell does it detect if you have used ChatGPT lol?

 

[dvc]

Yeah, amazing tech, saw the video where the tool analyzed a reply generated by ChatGPT. The reply felt 100% human as it features an agressive/troll tone. Even with that it took the tool few seconds to mark it as AI generated content.

 

[buhari4u]

Tested https://etedward-gptzero-main-zqgfwb.streamlit.app/ with two paragraph of purely direct gpt generated article and the result is "likely human written".

 

[Jev_bh]

Tested https://etedward-gptzero-main-zqgfwb.streamlit.app/ with two paragraph of purely direct gpt generated article and the result is "likely human written".

Oh really! haha. So its not that good then by the looks of it?

 

[DanielX]

Tested https://etedward-gptzero-main-zqgfwb.streamlit.app/ with two paragraph of purely direct gpt generated article and the result is "likely human written".

I would not be surprised if some human-generated keyword-stuffed texts will be flagged as AI-generated. I believe there will be a lot of false positives and false negatives.

 

[Justin Pettingill]

Really is it work?

 

[Zera Api]

The line is getting thinner between human readable and GPT , these tools will only help improve the the content quality of Chat GPT in future

 

[RealTraffic1]

I'm almost sure that google already has some sort of tool to find which contact is AI-generated.
but it is nice to know there is such a tool for public users, It could help to test "copywriters"

 

[GoodDogWoof]

I would not be surprised if some human-generated keyword-stuffed texts will be flagged as AI-generated. I believe there will be a lot of false positives and false negatives.

Yeah, I guess that's a good thing.

Quality content should be unique and until this point, that hasn't been rewarded by Google despite theior pledges over the years. The whole .edu baclink debacle was evidence of that.

If the tool picks up some ****** human content in the crossfire and demotes, so be it.

Maybe we'll actually see the end to KW stuffing

 

[Eternalman]

I have tried it, I think it is as accurate as a tool can be. It is only saying a certain input is all ai. It only alims one part of the input is likely to be ai, while the other part is all human, it is only going paragraph by paragraph, as I understand. That at the same time might be its weakness as well. Anyways, seems to be good enough for now

 

[DanielX]

Yeah, I guess that's a good thing.

Quality content should be unique and until this point, that hasn't been rewarded by Google despite theior pledges over the years. The whole .edu baclink debacle was evidence of that.

If the tool picks up some ****** human content in the crossfire and demotes, so be it.

Maybe we'll actually see the end to KW stuffing

The main problem I see is that Google is a robot. So it will always rank content written for robots higher than content written for humans. If they check with their own AI for AI content, there will be a better AI whose content pleases Google AI and passes, regardless if it's the best content for humans.

 

[irisXE]

Tested https://etedward-gptzero-main-zqgfwb.streamlit.app/ with two paragraph of purely direct gpt generated article and the result is "likely human written".

I wish I'd get the same results, but...

Paraphrasing/ rewriting it with different other tools (Quillbot and similar ones) doesn't help either. On the contrary, sometimes the text gets even higher detection scores after it has been paraphrased.

Does anyone know a working method to make the ChatGPT content undetectable?

 

[Sartre]

I paraphrased ChatGPT output using Quillbot and passed:

 

[MatrixR00T]

Mh.... So the war between AI vs AI finally begins.... Make a AI that can't be detected by AI