Watermarking ChatGPT is Still on OpenAI’s Radar

OpenAI has a system ready for watermarking ChatGPT-created text. But for about a year, the company has been majorly divided on whether to release this tool or not. 

One side of the argument says that it seems like a responsible thing to do, while on the other hand, it could crash the very significance of this AI language tool. 

But, What’s the whole ChatGPT Watermarking All About? 

OpenAI’s Watermarking tool is described as predicting the ChatGPT-created Text. The model predicts the most likely words and phrases, following the previous lines.  

In kudos for their recent model, OpenAI brings up the effectiveness of their tool against some localized tempering, such as paraphrasing. 

However, they have also been exploiting the tool’s non-effectiveness against the global tempering methods. Such as leveraging a translation system, rephrasing the whole text with another generative tool, or even asking the ChatGPT to insert some special character in between every word and then delete those later. 

According to a journal, OpenAI reports that the tool is working with almost 99.9% effectiveness and that “people worldwide supported the idea of an AI detection tool by a margin of four to one,”

Also, the company has expressed its concern about the stigmatization of AI tools, and their usefulness to non-native speakers. 

And, it seems that OpenAI is deeply worried about implementing their watermarking model, as almost 30% of the ChatGPT users say that they would use ChatGPT less if it included a watermark.