Because users of ChatGPT may be detected, OpenAI will not watermark text.

The Wall Street Journal says that OpenAI has had a method for watermarking ChatGPT-created text and a tool to find the watermark ready for about a year. But people inside the company disagree on whether to share it or not. On the one hand, it seems like the right thing to do, but it might hurt their bottom line.

OpenAI’s watermarking is said to change how the model guesses the most likely words and sentences that will come after the ones that came before them, making a pattern that can be seen. If you want to learn more about Gemini’s text watermarking, you can read Google’s more in-depth description.

The company says this was “99.9% useful” at making AI text recognizable when there was sufficient of it. This could help teachers who want to keep students from giving their writing tasks to AI without affecting the quality of its chatbot’s text output. The company settled for a poll, and “people around the world kept the idea of an AI detection tool by a margin of four to one,” the Journal writes.

But it looks like OpenAI is scared that watermarking might turn off ChatGPT users. About 30% of those asked told the company that they would use the software less if watermarking was added.

According to the Journal, some staff members were also worried that watermarking could be easily stopped by doing things like using Google Translate to switch between languages or making ChatGPT add images and then removing them.

Even so, workers are said to still think that the method works. The story does, however, say that some people have suggested trying ways that are “potentially less controversial among users but unproven.” I guess something is better than nothing.

Thank you for reading this post, don't forget to follow my whatsapp channel