Everybody is writing and talking about ChatGTP. For teachers and professors, the question often is: how will we deal with this in education. The same people who made the tool also made the GPT-2 Output Detector. Let’s have a real-life look.

Yesterday I used the tool myself in a way my friend and colleague Casper advised his students to use it: as a spelling checker on steroids. What did I do?

I first used Deepl to translate the post by Jeroen Janssen for the guest post on this blog that I posted yesterday. Next, I asked ChatGTP to check whether the text was written by AI, both for the translation and the updated version by ChatGTP.

This led to some surprising results. First, the original translation by Deepl was regarded as very fake!

But… the altered version by ChatGTP was regarded as less fake!

I think using these kinds of tools to do translations could be okay for people, but I’m worried about the detector tool. Let me explain. We don’t know how this tool works.  It can be a kind of ‘computer says no’-story, and secondly: in the current state, it’s hard to tell the difference between smart use (spelling control 2.0) and evil use (all the examples people share online lately).