#ChatGPT also the DetectGPT?

20 messages · Page 1 of 1 (latest)

frail hatch
#

So I've just realized that chatGPT can detect whether a certain text/essay etc. was generated by it or not. Thought I should share with you guys.

Screenshot_32.png Screenshot_33.png
dim bane
frail hatch
sinful cloud
#

I would not trust that ChatGPT is not just pretending. I have seen this idea mentioned a few times and never saw anyone saying it works.

#

There is a post on the **open AI **website from January 31 where it announced a classifier that detects the plausibility that text might be AI generated

#

google openai AI Text Classifier

#

It can already be used but it's just measuring likelihood it can not guarantee anything

frail hatch
spiral geyser
#

whatever i enter and ask chatgpt to tell if its generated by itself or not

#

it always says

image.png
vague narwhal
spiral geyser
#

hm

wind oyster
#

The accuracy of the model is pretty good in my opinion. I tried changing the original text. Simply changing words or rephrasing didn't prevent the model from producing accurate results. However, when the overall structure of the text was changed, chat was no longer able to recognize it.

vague narwhal
#

The thing is with new instances and new context they accuracy may be improved

#

like if you let him doo research about expsing ai generated text by gpt 3.5 models etc, then ask him to check your text if its ai generated

#

could potentially increase the accuracy, except the model does like to do what it like and pretend that their info is correct

#

Also the restricted memory of the tool...

cold hornet
#

Try GPTZero. It tests the text for 'burstiness', which indicates the text was 'generated' and not 'written'. ChatGPT is likely to be verbose and repetitive. That's the best (currently) design for plag detection, IMO.

charred fractal