OpenAI creates a tool to detect texts from ChatGPT

Since we started talking about ChatGPT or even before, with the first news about other artificial intelligence models capable of creating texts made from simple queries, a more than likely use of them has been in the minds of many people, understandable but not defensible. And it is that, in the hands of any student, the temptation to resort to an AI for the elaboration of works that should be written by the same it must be, it is easily imaginable, simply enormous.

Those of us who have gray hair, that is, those of us who had to resort to the encyclopedia at home or the library to do the work that we were asked to do in class, saw the arrival of digital encyclopedias and later the Internet (with services such as Wikipedia, yes, but also with others that are much more specific, such as El Rincón del vago), how suddenly everything was much easier (in some cases even too easy) and, although we disapproved of some of those uses, specifically those based on the sequence of keystrokes control+c and control+v, we probably would have made heavy use of them as well if they existed in our time.

With the arrival of AIs like ChatGPT, everything gets even more complicated, because unlike the aforementioned options, which always start from the same text, artificial intelligence models generate different answers (although with the same sense and meaning) when faced with similar responses, which makes it somewhat difficult to detect texts with this origin. It is true, yes, that AIs have a very “personal” style (paradoxically, since they have nothing of a person) that can be identified, but of course, prior training is necessary for this. In other words, today it is quite likely that many teachers are facing texts that they do not know if they originate from ChatGPT but that are delivered by the students as their own.

This, however, may be about to change, as OpenAI, the company behind the AI ​​trend, has created a tool that allows to identify texts created by ChatGPT and other similar text generation models. The tool, like ChatGPT, is totally free. To use it, you just have to access it, something you can do at this link and, if you are not registered, follow the steps indicated here.

Once inside, its operation is very simple and, although the website is in English, you can also use it with texts in Spanish. To carry out the test, copy the text that you want to analyze in the box that is shown in it, taking into account that it must have a minimum length of 1,000 characters, and press the Submit button. If the text looks like it was created by an artificial intelligence, then the message will be displayed

OpenAI creates a tool to detect texts from ChatGPT

For its part, if the OpenAI tool considers that it is more likely to be a text written by a person, the message you will see is the following:

OpenAI creates a tool to detect texts from ChatGPT

And if it is not clear, what it will indicate is the following:

OpenAI creates a tool to detect texts from ChatGPT

Keep in mind, yes, that its creators affirm that identification errors may occur, that if the texts have been edited it is more difficult to detect their origin in an AI, and that the highest degree of reliability is offered with texts in English. To test its operation and take screenshots of its responses, I have tried it with several texts in Spanish (including the draft of this publication) and the results… well, they leave a bit to be desired. You have perfectly identified several texts written by ChatGPT, but there have been several false positives and some doubts with texts of human origin.

Related Articles

Leave a Reply

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