According to a recent report by The Wall Street Journal, OpenAI has developed a highly accurate tool to detect essays written by ChatGPT but has not yet released it. The company’s latest update, highlighted, sheds light on their research into text watermarking and the reasons behind withholding the detection method.
OpenAI stated in an update to a May blog post that their teams have created a text watermarking technique and are exploring other alternatives. While watermarking has shown high accuracy in some scenarios, it struggles with certain types of tampering, such as using translation tools, rewording with other generative models, or adding and removing special characters. Additionally, the watermarking could negatively affect specific groups, potentially stigmatizing the use of AI writing tools for non-native English speakers.
OpenAI has been carefully considering these risks and has given priority to developing authentication tools for audiovisual content. In a statement to tech magazine TechCrunch, an OpenAI spokesperson mentioned that the company is taking a “deliberate approach” to text provenance due to the complexities and potential impacts on the broader ecosystem beyond OpenAI.
Source: Engadget.com
