OpenAI, creator of the famous AI ChatGPT, just revealed a tool capable of detecting if an AI was used for writing a text. Quite practical for teachers, who see more and more homework written by a machine. Moreover, the use of ChatGPT has been prohibited in certain schools, including Sciences Po on our territory. The tool is there mainly to address this concern.
AI Text Classifier looks like a web page in which you just have to copy/paste the suspicious text (at least 1000 characters). The tool will then give you a ranking ranging from a "Most likely NOT written by an AI" to "Probably written by an AI", passing through some uncertainties, because OpenAI specifies that its tool is not reliable at 100 %.
The tool is powered by a model trained on texts written by humans and texts written by AIs. OpenAI hopes its tool will have an impact on journalists and disinformation researchers.
Is AI Text Classifier effective?
I was able to test it since this article is intentionally written using ChatGPT itself, then edited by myself. A quick copy/paste in AI Text Classifier gives the following result: “Classifer considers the text to be possibly generated by AI. " Well seen !