OpenAI Develops Text Watermarking Technology to Identify ChatGPT-Generated Content
According to a report by The Wall Street Journal, OpenAI has confirmed that it is working on a high-precision tool to identify text generated by ChatGPT, although it has not yet been officially released. The company acknowledges that while the development of text watermarking technology faces numerous challenges, they remain committed to researching this innovation.
The essence of text watermarking lies in its ability to specifically detect text attributable to ChatGPT, as opposed to content produced by other models. OpenAI plans to accomplish this by subtly adjusting the choice of words generated by ChatGPT to embed an invisible watermark, which can later be detected with specialized tools.
OpenAI has indicated that text watermarking is just one of several solutions they are exploring, alongside classifiers and metadata, to determine the origins of text. However, the efficacy of text watermarking could diminish significantly in cases of translation, rewriting, or the insertion of special characters. Additionally, this technology may disproportionately affect certain groups, particularly non-native English speakers.
Given these complexities and the potential implications for the broader ecosystem, OpenAI has stated that it will proceed cautiously with the development of text tracing technology, prioritizing the advancement of verification tools for audiovisual content.