OpenAI confirms that it looks for text watermarking of ChatGPT that can expose students who cheat

Following is a report from The Wall Street Journal saying that OpenAI was sitting on a tool that could recognize articles written by ChatGPT with a high degree of accuracy, the company shared little information about its research into text watermarking — and why it hasn’t released its detection method yet. In accordance with The Wall Street Journalreport, the debate about whether the tool should be released has never seen the light of day, except for “right.” In an update published on Sunday in May’s blog post, seen by TechCrunchOpenAI said, “Our teams have developed a text tagging method that we continue to think about as we research other methods.”

The company said watermarking is one of many solutions, including classifiers and metadata, that it is looking at as part of “wider research into the area of ​​text evolution.” According to OpenAI, it was “very accurate” in some cases, but it does not perform well when dealing with certain types of interference, “such as using translation systems, rewriting words with another generating model, or asking the model to insert a feature.” character between all words and removes that character.” And watermarking “could equally impact other parties,” OpenAI wrote. “For example, it may marginalize the use of AI as a useful writing tool for non-native English speakers.”

In a blog post, OpenAI is weighing these risks. The company also wrote that it has prioritized the release of tools to verify audio and visual content. In a statement to TechCrunchan OpenAI spokesperson said the company is taking a “deliberate approach” to wording because of “the complexity involved and its potential impact on the broader ecosystem beyond OpenAI.”