ChatGPT: this online tool recognizes texts created via artificial intelligence

Faced with the increasingly diverted uses by pupils and students of ChatGPT, the famous conversational robot that has the answer to everything, text detection solutions created by artificial intelligence are beginning to be put in place.

Draft & Goal is thus a brand new ChatGPT content detector. This Franco-Canadian initiative consists of analyzing any text and evaluating whether it was really written by a human or by any artificial intelligence, the most advanced of which today is ChatGPT. It currently works for both English and French texts.

Although open to everyone, this tool is above all intended mainly for teachers, for whom ChatGPT has become a real nightmare, no longer knowing if the homework returned to them is the work of their pupils/students or of any kind of artificial intelligence.

As a reminder, ChatGPT is a conversational robot that has the answer to everything. He can just as well answer any topical or even philosophical question, argue or comment on a text, write a presentation or even write a poem or a song on a dedicated theme. In fact, he can easily help with his homework, by directly answering the questions asked by the teachers. Constantly in development, ChatGPT is a creation of OpenAI, an American company whose one of the co-founders is none other than Elon Musk.

ChatGPT is not foolproof

The way Draft & Goal works is very simple since all you have to do is go to the site, paste the text to be analyzed and wait for the result, given in the form of a percentage. The lower this score, the more “chance” it has of having been written by a human being. Conversely, the closer it gets to 100%, the greater the possibility that it is a text from ChatGPT.

To find out if this text was written by ChatGPT, this tool actually relies on errors regularly made (for the moment) by artificial intelligence. Indeed, ChatGPT multiplies grammatical errors, repetitions, exaggerations and contradictions, not to mention calculation errors.

Although very powerful, the tool currently focuses on texts of “only” 400 characters maximum. This should change quickly. Finally, an API is being developed so that one day it can be integrated into third-party sites or services.

Leave a Comment

Your email address will not be published. Required fields are marked *