The development of solutions capable of detecting text created by artificial intelligence does not stop. Researchers at Stanford University have announced the creation of a new toolknown as DetectGPT, which is capable of detecting if a text has been created by an AI such as the popular ChatGPTcreated by Open AI.
With increasing development, fluency, and knowledge of AI language models, students can make use of these tools to carry out written tasks. This has caused several academic centers to show preoccupation with the use of technologyeven in some centers the use of the solution created by Open AI has been prohibited.
As mentioned by the group of researchers who have created DetectGPT, text generators using artificial intelligence “make it difficult for teachers to accurately assess whether the student has really learned.”
The tool, using probability and statistical data, was created with the aim of recognizing and distinguishing the texts created by AI, so as to detect fraud in the authorship of the texts.
The operation of this tool is based on a “likelihood curvature” statistical method, which suggests that text developed by an AI tends to occupy regions of negative curvature of the model’s log-likelihood function. That is to say, ands able to detect patterns in a text that reveal that there is a machine behind the authorship.
From this method, the team of researchers developed a new criterion based on this curvature to determine if a text has been generated from an AI or not. Thus, as the developers explain, DetectGPT uses only the log odds calculated by this model, and does not require training a classifier or collecting a data set of real or generated passages.
This is not the first solution dedicated to recognizing text created by AI. Also, DetectGPT is currently not available for use by any Internet user. This is something that does not happen with GPTZeroanother tool that has been on the Internet for weeks and that works similarly to the software created by researchers at Stanford University.
Likewise, Open AI itself has acknowledged that it is working on its own solution aimed at preventing ChatGPT from being used to copy within the academic field.