OpenAI now reportedly has one of the methods to easily identify content generated using ChatGPT - especially in writing research papers, or essays. However, the firm reportedly does not want to offer it to the public for now.
According to the WSJ report, the functionality to identify ChatGPT-generated content has been available for almost a year – however, it is still being discussed internally whether to offer it or not. It is said that this function has an accuracy rate of up to 99.9% effective in identifying text generated by ChatGPT - through the use of watermarks that can be identified by the system, but unknown to the general user.
Previously, many teaching staff hoped to be able to identify research papers, essays or any works written using ChatGPT, which would make it easier to identify such works, and ensure that students carry out assignments more transparently.
It is said internally that various matters are discussed in offering this function to the public. With an open offering, it is expected that some may find a watermarking method is done to identify the work, and come up with a way to overcome it. While if it is offered on a limited basis to some parties, then the function may not be completely useful.
In addition to OpenAI, Google has already developed a method to identify GeminiAI-generated text, named SynthID – and is currently in beta testing.