OpenAI is taking a deliberate approach to watermarking text from ChatGPT. This can potentially help teachers catch students who make ChatGPT write their assignments. But, as reported by The Wall Street Journal, the company is still debating whether to release it or not.
According to a TechCrunch blog post, Open AI says, “We have developed a text watermarking method, but we are continuing to weigh the risk while researching alternatives.” This watermarking tool will be a different approach from all the previous AI detection tools, as it will be more accurate.
But OpenAI will only focus on detecting AI-generated content from ChatGPT and not from other models. By making small changes to how ChatGPT selects words, it will create an invisible watermark in writing, which could later be detected by a different tool.
OpenAI is taking a deliberate approach to watermarking text from ChatGPT but is also considering other methods like classifiers and metadata. The company is contemplating this because many users say that they will reduce usage if text watermarking is introduced. Watermarking the text could also “stigmatize the use of AI as a useful writing tool for non-native English speakers.”
