OpenAI’s classifier tool is pretty much worthless, though.
It only flags its own output 25% of the time. And it only takes slight modifications of its own output to fool the classifier even when it otherwise would catch it.
And, it’s easy for students to just keep making small changes then running it through the classifier until it’s “safe” to submit.
There are better tools, theirs is one of the worst (for now, I'm sure they will improve it).
View more
Yes. Thanks for pointing that out.
OpenAI’s classifier tool is pretty much worthless, though.
It only flags its own output 25% of the time. And it only takes slight modifications of its own output to fool the classifier even when it otherwise would catch it.
And, it’s easy for students to just keep making small changes then running it through the classifier until it’s “safe” to submit.
View more
The issue with the majority of these tools is that it is a blackbox. You cannot tell what it measures to determine if a text is ai written.
This is one of the tools that were are using to detect AI texts in articles on Hive.
View more