Chat GPT is watermarking its content. Here’s how you can escape it.
From copywriters to students, everyone is using Chat GPT for writing sales copy, assignments, and much more. However, Chat GPT is planning to watermark the content. But you can still use it with a straightforward trick.
Who Broke The News?
Scott Aranson, a computer scientist working for OpenAI for AI safety and alignment, reported at one of his talks at the University of Texas that the team is working on cryptographic watermarking of the AI-generated content from Chat GPT to develop a distinction between human-generated and AI-generated content.
But Why Do We Need It?
Although Chat GPT has been a heavenly solution for copywriters and students from around the globe, many individuals and institutions are still repulsive of it. Talented and experienced copywriters, who produce high-quality content by investing their precious time in copywriting, can face substantial challenges. While Chat GPT cannot create high-quality content like many skilled writers in the industry, AI still allows individuals and agencies to develop a vast range of content and storm the internet with similar keywords in no time. This would lead to undermining the efforts of skilled and talented copywriters.
Educational institutes, on the other, are facing the issue of academic integrity. Students can generate content for their assignments, thesis, and many other tasks that are required to be solely written by them. Since plagiarism detectors cannot identify which content is AI-generated, this presents institutions with a massive issue of validating the authenticity and ownership of content.
However, Aronson also pointed out other ethical dilemmas associated with the ability of Chat GPT to generate unique content in bulk. He highlighted that:
“This could be helpful for preventing academic plagiarism, obviously, but also, for example, mass generation of propaganda…”
Yes, propaganda is another major issue. Previously, institutions or individuals were required to produce content independently to spread political and social propaganda. However, with Chat GPT, one can write thousands of words of false and misleading information that influence many others around the globe with simple one-line prompts.
Therefore, with the help of watermarking, institutions and platforms like Medium and Quora can identify and block AI-generated content and take necessary actions against it.
How Will This Work?
This can get complex, but I will make it simple and precise. Nevertheless, you can certainly check out Araonson’s blog for a detailed view of the process.
AI generates content in the form of tokens, which can be words, punctuations, or even parts of words. Manipulating these can provide a solution for watermarking. Currently, there exist 100,000 tokens, and Chat GPT is continuously developing a probability distribution regarding which token should be generated next. The system checks the randomness of each token, and if it is random, it shows it as the output. That is why a similar prompt can yield different outcomes.
For watermarking, Aaronson and his team worked on selecting the next token in a pseudo-random fashion with the help of a cryptographic function rather than being truly random. The pseudo-random process will create a score, and summing it covers a particular value of tokens in a series will give a result. Those who know the score and the key can determine the likelihood of the content being AI-generated. The essential part is that the output would be different for the end-user.
This is how Aranson puts it:
“For GPT, every input and output is a string of tokens, which could be words but also punctuation marks, parts of words, or more — there are about 100,000 tokens in total.
At its core, GPT is constantly generating a probability distribution over the next token to generate, conditional on the string of previous tokens.
After the neural net generates the distribution, the OpenAI server then actually samples a token according to that distribution — or some modified version of the distribution, depending on a parameter called ‘temperature.’
As long as the temperature is nonzero, though, there will usually be some randomness in the choice of the next token: you could run over and over with the same prompt, and get a different completion (i.e., string of output tokens) each time.
So then to watermark, instead of selecting the next token randomly, the idea will be to select it pseudorandomly, using a cryptographic pseudorandom function, whose key is known only to OpenAI.”
OpenAI engineer Hendrik Kirchner has also developed a working prototype for it.
So Is It The End? Or Do We Have A Solution?
The interesting thing is that Scott Aaronson himself highlighted that there exists a way to defeat watermarking. Aranson maintained:
“Now, this can all be defeated with enough effort.
For example, if you used another AI to paraphrase GPT’s output — well okay, we’re not going to be able to detect that.”
So, unless OpenAI has not changed the watermarking plan, as Aranson explained it, all those using the AI-generated content are in luck, but they would need a bit of effort.
New to trading? Try crypto trading bots or copy trading on best crypto exchanges
