OpenAI Develops Text Watermarking Tech for ChatGPT But Faces Challenges

TapTechNews August 5th news, according to The Wall Street Journal report, OpenAI has developed a tool that can accurately identify text generated by ChatGPT, but it has not been released yet. In response, OpenAI acknowledged that it is researching text watermarking technology, but stated that there are still many challenges with this technology.

According to TapTechNews, through text watermarking, OpenAI will only focus on detecting text from ChatGPT, not from other company models. It will achieve this by making minor changes to the way ChatGPT selects words, essentially creating an invisible watermark in the text that can later be detected by a separate tool.

OpenAI said that text watermarking is one of the multiple solutions they are exploring, others include classifiers and metadata, aiming to determine the source of the text. Although in some cases, the text watermarking technology performs well, but the effect will decrease when facing tampering behaviors such as translation, rewriting or inserting special characters.

OpenAI stated that due to these complex factors and their potential impact on the entire ecosystem, OpenAI said it will cautiously advance the research of text traceability technology and give priority to the authentication tools for audio-visual content.