ChatGPT texts detectable - model reliably recognizes AI plagiarism

[Redaktion]

a program at the University of Kansas claims to be able to reliably unmask 99% of all AI-generated texts. This is huge progress compared to previous attempts to automatically separate artificially created texts from human ones.

https://www.notebookcheck.net/ChatGPT-texts-detectable-model-reliably-recognizes-AI-plagiarism.765958.0.html

[Ednumero]

Promising findings. I wonder how much the detection rate would vary depending on how much a human guides the AI in producing the fictitious work.

Linguistically, I'm cautious towards extending the term Plagiarism to this, as the word already has a history of misappropriation in certain educational spaces (see: "self-plagiarism"). Words as strongly-connotated as such should be adopted only with sufficiently thorough forethought so as to ensure the actual definitional criteria is met, lest they lose their impact.

We certainly need tools to combat this, whatever the proper term may actually end up being.