How The ChatGPT Watermark Works And Why It Could Be Defeated

Posted by

OpenAI’s ChatGPT presented a way to instantly produce material but plans to introduce a watermarking function to make it simple to identify are making some individuals nervous. This is how ChatGPT watermarking works and why there may be a way to beat it.

ChatGPT is an amazing tool that online publishers, affiliates and SEOs concurrently enjoy and dread.

Some marketers love it due to the fact that they’re discovering brand-new methods to utilize it to generate content briefs, lays out and complex articles.

Online publishers are afraid of the possibility of AI content flooding the search results page, supplanting specialist articles composed by human beings.

Consequently, news of a watermarking feature that opens detection of ChatGPT-authored content is also prepared for with anxiety and hope.

Cryptographic Watermark

A watermark is a semi-transparent mark (a logo design or text) that is embedded onto an image. The watermark signals who is the original author of the work.

It’s mostly seen in pictures and significantly in videos.

Watermarking text in ChatGPT involves cryptography in the form of embedding a pattern of words, letters and punctiation in the type of a secret code.

Scott Aaronson and ChatGPT Watermarking

A prominent computer scientist called Scott Aaronson was worked with by OpenAI in June 2022 to deal with AI Safety and Alignment.

AI Security is a research field concerned with studying ways that AI might present a damage to people and creating ways to prevent that kind of unfavorable disturbance.

The Distill clinical journal, featuring authors connected with OpenAI, specifies AI Security like this:

“The goal of long-term expert system (AI) safety is to ensure that innovative AI systems are reliably lined up with human worths– that they dependably do things that individuals want them to do.”

AI Alignment is the artificial intelligence field concerned with making sure that the AI is aligned with the designated goals.

A big language design (LLM) like ChatGPT can be utilized in such a way that may go contrary to the objectives of AI Alignment as defined by OpenAI, which is to develop AI that advantages humankind.

Accordingly, the factor for watermarking is to prevent the misuse of AI in a way that harms humankind.

Aaronson described the factor for watermarking ChatGPT output:

“This could be handy for avoiding academic plagiarism, clearly, but also, for example, mass generation of propaganda …”

How Does ChatGPT Watermarking Work?

ChatGPT watermarking is a system that embeds a statistical pattern, a code, into the choices of words and even punctuation marks.

Material produced by expert system is generated with a fairly predictable pattern of word option.

The words written by humans and AI follow an analytical pattern.

Altering the pattern of the words used in produced material is a way to “watermark” the text to make it simple for a system to discover if it was the product of an AI text generator.

The trick that makes AI material watermarking undetected is that the circulation of words still have a random look comparable to normal AI generated text.

This is referred to as a pseudorandom distribution of words.

Pseudorandomness is a statistically random series of words or numbers that are not in fact random.

ChatGPT watermarking is not presently in use. Nevertheless Scott Aaronson at OpenAI is on record stating that it is prepared.

Right now ChatGPT is in previews, which permits OpenAI to discover “misalignment” through real-world usage.

Probably watermarking may be presented in a final variation of ChatGPT or quicker than that.

Scott Aaronson discussed how watermarking works:

“My primary project so far has been a tool for statistically watermarking the outputs of a text model like GPT.

Generally, whenever GPT generates some long text, we desire there to be an otherwise unnoticeable secret signal in its choices of words, which you can use to prove later on that, yes, this came from GPT.”

Aaronson discussed even more how ChatGPT watermarking works. However initially, it’s important to understand the concept of tokenization.

Tokenization is a step that occurs in natural language processing where the maker takes the words in a file and breaks them down into semantic units like words and sentences.

Tokenization changes text into a structured form that can be utilized in artificial intelligence.

The procedure of text generation is the maker thinking which token comes next based on the previous token.

This is done with a mathematical function that identifies the possibility of what the next token will be, what’s called a possibility distribution.

What word is next is anticipated but it’s random.

The watermarking itself is what Aaron describes as pseudorandom, because there’s a mathematical factor for a particular word or punctuation mark to be there however it is still statistically random.

Here is the technical description of GPT watermarking:

“For GPT, every input and output is a string of tokens, which might be words however also punctuation marks, parts of words, or more– there are about 100,000 tokens in overall.

At its core, GPT is constantly generating a probability circulation over the next token to produce, conditional on the string of previous tokens.

After the neural net produces the distribution, the OpenAI server then actually samples a token according to that circulation– or some customized variation of the distribution, depending on a criterion called ‘temperature.’

As long as the temperature is nonzero, however, there will typically be some randomness in the choice of the next token: you might run over and over with the very same timely, and get a various conclusion (i.e., string of output tokens) each time.

So then to watermark, rather of selecting the next token randomly, the concept will be to choose it pseudorandomly, utilizing a cryptographic pseudorandom function, whose key is understood only to OpenAI.”

The watermark looks entirely natural to those reading the text since the choice of words is mimicking the randomness of all the other words.

However that randomness consists of a predisposition that can just be identified by somebody with the key to decipher it.

This is the technical description:

“To illustrate, in the special case that GPT had a bunch of possible tokens that it evaluated similarly probable, you could just choose whichever token optimized g. The choice would look evenly random to someone who didn’t know the secret, however somebody who did understand the key might later sum g over all n-grams and see that it was anomalously big.”

Watermarking is a Privacy-first Solution

I’ve seen conversations on social media where some people suggested that OpenAI might keep a record of every output it generates and use that for detection.

Scott Aaronson verifies that OpenAI might do that however that doing so presents a privacy problem. The possible exception is for police scenario, which he didn’t elaborate on.

How to Identify ChatGPT or GPT Watermarking

Something interesting that appears to not be popular yet is that Scott Aaronson noted that there is a way to beat the watermarking.

He didn’t state it’s possible to beat the watermarking, he stated that it can be beat.

“Now, this can all be defeated with enough effort.

For example, if you utilized another AI to paraphrase GPT’s output– well okay, we’re not going to have the ability to detect that.”

It looks like the watermarking can be defeated, at least in from November when the above declarations were made.

There is no indication that the watermarking is currently in use. However when it does enter usage, it may be unidentified if this loophole was closed.


Read Scott Aaronson’s article here.

Included image by SMM Panel/RealPeopleStudio