OpenAI is exploring the possibility of using a cryptographic watermark in ChatGPT text output. This will make it easier to spot text created by the AI chatbot. It is important to ensure that the tool doesn’t get misused in any way, such as writing papers for college or school.
Table Of Contents 👉
How ChatGPT Watermarking Works?
A cryptographic watermark can be described as a type of digital watermark embedded in a digital signal or file using cryptographic techniques. Digital watermarks can be used to identify ownership of a digital signal or file or to protect against digital content’s unauthorized distribution or use. OpenAI is working on a cryptographic watermark that can be integrated into ChatGPT’s text.
You may think it would work without affecting coherence in the text output. But the truth is that text is so wide and infinite that it’s possible to include non-noticeable cryptographic codes without affecting its readability.
The truth is that AI-generated content follows a predictable pattern when it uses words. This means it is easy to identify if you know what to search for. OpenAI is looking to add an element of predictability subtly. This process is known as pseudorandom distribution, according to Scott Aaronson, Head of AI Safety and Alignment at OpenAI.
He said Pseudorandomness refers to a sequence of numbers or words that appears random but is not. This allows us to mark AI-generated content without affecting the appearance of the text.
Aaronson wrote in one of his blog: “My main project has been to statistically watermark the outputs of text models like GPT. GPT generates long text. We want to see a subtle signal that is not noticeable. This secret signal can later be used to prove that this thing was created using GPT.
How To Detect ChatGPT Watermarking And Remove The ChatGPT Watermark?
There is an easy way to remove this ChatGPT watermarking. You can run the text through another generative AI tool and ask it for a rewrite. Although it might seem silly, it would allow the user to generate text without needing to bypass the safety feature of having another protocol create the cryptographic watermark.
This is a work in process, so OpenAI isn’t actively watermarking ChatGPT’s text output. However, be cautious if you intend to use ChatGPT to write some important text. As we said above, if you know what to look for, you will be able to spot AI-generated text, even without having a specific watermark to look for.
Related Stories: