SEO

How The ChatGPT Watermark Works And Why It Could Be Defeated

Published

1 year ago

December 30, 2022

How The ChatGPT Watermark Works And Why It Could Be Defeated

OpenAI’s ChatGPT introduced a way to automatically create content but plans to introduce a watermarking feature to make it easy to detect are making some people nervous. This is how ChatGPT watermarking works and why there may be a way to defeat it.

ChatGPT is an incredible tool that online publishers, affiliates and SEOs simultaneously love and dread.

Some marketers love it because they’re discovering new ways to use it to generate content briefs, outlines and complex articles.

Online publishers are afraid of the prospect of AI content flooding the search results, supplanting expert articles written by humans.

Consequently, news of a watermarking feature that unlocks detection of ChatGPT-authored content is likewise anticipated with anxiety and hope.

Cryptographic Watermark

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

It’s largely seen in photographs and increasingly in videos.

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

Scott Aaronson and ChatGPT Watermarking

An influential computer scientist named Scott Aaronson was hired by OpenAI in June 2022 to work on AI Safety and Alignment.

AI Safety is a research field concerned with studying ways that AI might pose a harm to humans and creating ways to prevent that kind of negative disruption.

The Distill scientific journal, featuring authors affiliated with OpenAI, defines AI Safety like this:

“The goal of long-term artificial intelligence (AI) safety is to ensure that advanced AI systems are reliably aligned with human values — that they reliably do things that people want them to do.”

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

A large language model (LLM) like ChatGPT can be used in a way that may go contrary to the goals of AI Alignment as defined by OpenAI, which is to create AI that benefits humanity.

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

Aaronson explained the reason for watermarking ChatGPT output:

“This could be helpful for preventing academic plagiarism, obviously, 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.

Content created by artificial intelligence is generated with a fairly predictable pattern of word choice.

The words written by humans and AI follow a statistical pattern.

Changing the pattern of the words used in generated content is a way to “watermark” the text to make it easy for a system to detect if it was the product of an AI text generator.

The trick that makes AI content watermarking undetectable is that the distribution of words still have a random appearance similar 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 actually random.

ChatGPT watermarking is not currently in use. However Scott Aaronson at OpenAI is on record stating that it is planned.

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

Presumably watermarking may be introduced in a final version of ChatGPT or sooner than that.

Scott Aaronson wrote about how watermarking works:

“My main project so far has been a tool for statistically watermarking the outputs of a text model like GPT.
Basically, whenever GPT generates some long text, we want there to be an otherwise unnoticeable secret signal in its choices of words, which you can use to prove later that, yes, this came from GPT.”

Aaronson explained further how ChatGPT watermarking works. But first, it’s important to understand the concept of tokenization.

Tokenization is a step that happens in natural language processing where the machine takes the words in a document and breaks them down into semantic units like words and sentences.

Tokenization changes text into a structured form that can be used in machine learning.

The process of text generation is the machine guessing which token comes next based on the previous token.

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

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

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

Here is the technical explanation of GPT watermarking:

“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.”

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

But that randomness contains a bias that can only be detected by someone with the key to decode it.

This is the technical explanation:

“To illustrate, in the special case that GPT had a bunch of possible tokens that it judged equally probable, you could simply choose whichever token maximized g. The choice would look uniformly random to someone who didn’t know the key, but someone who did know the key could later sum g over all n-grams and see that it was anomalously large.”

Watermarking is a Privacy-first Solution

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

Scott Aaronson confirms that OpenAI could do that but that doing so poses a privacy issue. The possible exception is for law enforcement situation, which he didn’t elaborate on.

How to Detect ChatGPT or GPT Watermarking

Something interesting that seems to not be well known yet is that Scott Aaronson noted that there is a way to defeat the watermarking.

He didn’t say it’s possible to defeat the watermarking, he said that it can be defeated.

“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.”
Advertisement

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

There is no indication that the watermarking is currently in use. But when it does come into use, it may be unknown if this loophole was closed.

Citation

Read Scott Aaronson’s blog post here.

Featured image by Shutterstock/RealPeopleStudio

var s_trigger_pixel_load = false;
function s_trigger_pixel(){
if( !s_trigger_pixel_load ){
striggerEvent( ‘load2’ );
console.log(‘s_trigger_pix’);
}
s_trigger_pixel_load = true;
}
window.addEventListener( ‘cmpready’, s_trigger_pixel, false);

window.addEventListener( ‘load2’, function() {

if( sopp != ‘yes’ && !ss_u ){

!function(f,b,e,v,n,t,s)
{if(f.fbq)return;n=f.fbq=function(){n.callMethod?
n.callMethod.apply(n,arguments):n.queue.push(arguments)};
if(!f._fbq)f._fbq=n;n.push=n;n.loaded=!0;n.version=’2.0′;
n.queue=[];t=b.createElement(e);t.async=!0;
t.src=v;s=b.getElementsByTagName(e)[0];
s.parentNode.insertBefore(t,s)}(window,document,’script’,
‘https://connect.facebook.net/en_US/fbevents.js’);

if( typeof sopp !== “undefined” && sopp === ‘yes’ ){
fbq(‘dataProcessingOptions’, [‘LDU’], 1, 1000);
}else{
fbq(‘dataProcessingOptions’, []);
}

fbq(‘init’, ‘1321385257908563’);

fbq(‘track’, ‘PageView’);

fbq(‘trackSingle’, ‘1321385257908563’, ‘ViewContent’, {
content_name: ‘chatgpt-watermark’,
content_category: ‘news seo’
});
}
});

Related Topics:ChatGPT Defeated seo Watermark Works

Up Next

Let’s Talk About Old Content And Redirect Chains

Don't Miss

10 Biggest & Best PPC Features Of The Year

Click to comment

You must be logged in to post a comment Login

SEO

brightonSEO Live Blog

Published

9 mins ago

April 25, 2024

Entireweb News Bot

Hello everyone. It’s April again, so I’m back in Brighton for another two days of ~~sun~~, sea, and SEO!

Being the introvert I am, my idea of fun isn’t hanging around our booth all day explaining we’ve run out of t-shirts (seriously, you need to be fast if you want swag!). So I decided to do something useful and live-blog the event instead.

Follow below for talk takeaways and (very) mildly humorous commentary.

SEO

Google Further Postpones Third-Party Cookie Deprecation In Chrome

Published

3 hours ago

April 25, 2024

Max

Close-up of a document with a grid and a red stamp that reads "delayed" over the word "status" due to Chrome's deprecation of third-party cookies.

Google has again delayed its plan to phase out third-party cookies in the Chrome web browser. The latest postponement comes after ongoing challenges in reconciling feedback from industry stakeholders and regulators.

The announcement was made in Google and the UK’s Competition and Markets Authority (CMA) joint quarterly report on the Privacy Sandbox initiative, scheduled for release on April 26.

Chrome’s Third-Party Cookie Phaseout Pushed To 2025

Google states it “will not complete third-party cookie deprecation during the second half of Q4” this year as planned.

Instead, the tech giant aims to begin deprecating third-party cookies in Chrome “starting early next year,” assuming an agreement can be reached with the CMA and the UK’s Information Commissioner’s Office (ICO).

The statement reads:

“We recognize that there are ongoing challenges related to reconciling divergent feedback from the industry, regulators and developers, and will continue to engage closely with the entire ecosystem. It’s also critical that the CMA has sufficient time to review all evidence, including results from industry tests, which the CMA has asked market participants to provide by the end of June.”

Continued Engagement With Regulators

Google reiterated its commitment to “engaging closely with the CMA and ICO” throughout the process and hopes to conclude discussions this year.

This marks the third delay to Google’s plan to deprecate third-party cookies, initially aiming for a Q3 2023 phaseout before pushing it back to late 2024.

The postponements reflect the challenges in transitioning away from cross-site user tracking while balancing privacy and advertiser interests.

Transition Period & Impact

In January, Chrome began restricting third-party cookie access for 1% of users globally. This percentage was expected to gradually increase until 100% of users were covered by Q3 2024.

However, the latest delay gives websites and services more time to migrate away from third-party cookie dependencies through Google’s limited “deprecation trials” program.

The trials offer temporary cookie access extensions until December 27, 2024, for non-advertising use cases that can demonstrate direct user impact and functional breakage.

While easing the transition, the trials have strict eligibility rules. Advertising-related services are ineligible, and origins matching known ad-related domains are rejected.

Google states the program aims to address functional issues rather than relieve general data collection inconveniences.

Publisher & Advertiser Implications

The repeated delays highlight the potential disruption for digital publishers and advertisers relying on third-party cookie tracking.

Industry groups have raised concerns that restricting cross-site tracking could push websites toward more opaque privacy-invasive practices.

However, privacy advocates view the phaseout as crucial in preventing covert user profiling across the web.

With the latest postponement, all parties have more time to prepare for the eventual loss of third-party cookies and adopt Google’s proposed Privacy Sandbox APIs as replacements.

Featured Image: Novikov Aleksey/Shutterstock

SEO

How To Write ChatGPT Prompts To Get The Best Results

Published

23 hours ago

April 24, 2024

Max

How To Write ChatGPT Prompts To Get The Best Results

ChatGPT is a game changer in the field of SEO. This powerful language model can generate human-like content, making it an invaluable tool for SEO professionals.

However, the prompts you provide largely determine the quality of the output.

To unlock the full potential of ChatGPT and create content that resonates with your audience and search engines, writing effective prompts is crucial.

In this comprehensive guide, we’ll explore the art of writing prompts for ChatGPT, covering everything from basic techniques to advanced strategies for layering prompts and generating high-quality, SEO-friendly content.

Writing Prompts For ChatGPT

What Is A ChatGPT Prompt?

A ChatGPT prompt is an instruction or discussion topic a user provides for the ChatGPT AI model to respond to.

The prompt can be a question, statement, or any other stimulus to spark creativity, reflection, or engagement.

Users can use the prompt to generate ideas, share their thoughts, or start a conversation.

ChatGPT prompts are designed to be open-ended and can be customized based on the user’s preferences and interests.

How To Write Prompts For ChatGPT

Start by giving ChatGPT a writing prompt, such as, “Write a short story about a person who discovers they have a superpower.”

ChatGPT will then generate a response based on your prompt. Depending on the prompt’s complexity and the level of detail you requested, the answer may be a few sentences or several paragraphs long.

Use the ChatGPT-generated response as a starting point for your writing. You can take the ideas and concepts presented in the answer and expand upon them, adding your own unique spin to the story.

If you want to generate additional ideas, try asking ChatGPT follow-up questions related to your original prompt.

For example, you could ask, “What challenges might the person face in exploring their newfound superpower?” Or, “How might the person’s relationships with others be affected by their superpower?”

Remember that ChatGPT’s answers are generated by artificial intelligence and may not always be perfect or exactly what you want.

However, they can still be a great source of inspiration and help you start writing.

Must-Have GPTs Assistant

I recommend installing the WebBrowser Assistant created by the OpenAI Team. This tool allows you to add relevant Bing results to your ChatGPT prompts.

This assistant adds the first web results to your ChatGPT prompts for more accurate and up-to-date conversations.

It is very easy to install in only two clicks. (Click on Start Chat.)

Screenshot from ChatGPT, April 2024

For example, if I ask, “Who is Vincent Terrasi?,” ChatGPT has no answer.

With WebBrower Assistant, the assistant creates a new prompt with the first Bing results, and now ChatGPT knows who Vincent Terrasi is.

Screenshot from ChatGPT, March 2023

You can test other GPT assistants available in the GPTs search engine if you want to use Google results.

Master Reverse Prompt Engineering

ChatGPT can be an excellent tool for reverse engineering prompts because it generates natural and engaging responses to any given input.

By analyzing the prompts generated by ChatGPT, it is possible to gain insight into the model’s underlying thought processes and decision-making strategies.

One key benefit of using ChatGPT to reverse engineer prompts is that the model is highly transparent in its decision-making.

This means that the reasoning and logic behind each response can be traced, making it easier to understand how the model arrives at its conclusions.

Once you’ve done this a few times for different types of content, you’ll gain insight into crafting more effective prompts.

Prepare Your ChatGPT For Generating Prompts

First, activate the reverse prompt engineering.

Type the following prompt: “Enable Reverse Prompt Engineering? By Reverse Prompt Engineering I mean creating a prompt from a given text.”

Screenshot from ChatGPT, March 2023

ChatGPT is now ready to generate your prompt. You can test the product description in a new chatbot session and evaluate the generated prompt.

Type: “Create a very technical reverse prompt engineering template for a product description about iPhone 11.”

Reverse Prompt engineering via WebChatGPT

Screenshot from ChatGPT, March 2023

The result is amazing. You can test with a full text that you want to reproduce. Here is an example of a prompt for selling a Kindle on Amazon.

Type: “Reverse Prompt engineer the following {product), capture the writing style and the length of the text :
product =”

Reverse prompt engineering: Amazon product

Screenshot from ChatGPT, March 2023

I tested it on an SEJ blog post. Enjoy the analysis – it is excellent.

Type: “Reverse Prompt engineer the following {text}, capture the tone and writing style of the {text} to include in the prompt :
text = all text coming from https://www.searchenginejournal.com/google-bard-training-data/478941/”

Reverse prompt engineering an SEJ blog post

Screenshot from ChatGPT, March 2023

But be careful not to use ChatGPT to generate your texts. It is just a personal assistant.

Go Deeper

Prompts and examples for SEO:

Keyword research and content ideas prompt: “Provide a list of 20 long-tail keyword ideas related to ‘local SEO strategies’ along with brief content topic descriptions for each keyword.”
Optimizing content for featured snippets prompt: “Write a 40-50 word paragraph optimized for the query ‘what is the featured snippet in Google search’ that could potentially earn the featured snippet.”
Creating meta descriptions prompt: “Draft a compelling meta description for the following blog post title: ’10 Technical SEO Factors You Can’t Ignore in 2024′.”

Important Considerations:

Always Fact-Check: While ChatGPT can be a helpful tool, it’s crucial to remember that it may generate inaccurate or fabricated information. Always verify any facts, statistics, or quotes generated by ChatGPT before incorporating them into your content.
Maintain Control and Creativity: Use ChatGPT as a tool to assist your writing, not replace it. Don’t rely on it to do your thinking or create content from scratch. Your unique perspective and creativity are essential for producing high-quality, engaging content.
Iteration is Key: Refine and revise the outputs generated by ChatGPT to ensure they align with your voice, style, and intended message.

Additional Prompts for Rewording and SEO:
– Rewrite this sentence to be more concise and impactful.
– Suggest alternative phrasing for this section to improve clarity.
– Identify opportunities to incorporate relevant internal and external links.
– Analyze the keyword density and suggest improvements for better SEO.

Remember, while ChatGPT can be a valuable tool, it’s essential to use it responsibly and maintain control over your content creation process.

Experiment And Refine Your Prompting Techniques

Writing effective prompts for ChatGPT is an essential skill for any SEO professional who wants to harness the power of AI-generated content.

Hopefully, the insights and examples shared in this article can inspire you and help guide you to crafting stronger prompts that yield high-quality content.

Remember to experiment with layering prompts, iterating on the output, and continually refining your prompting techniques.

This will help you stay ahead of the curve in the ever-changing world of SEO.

More resources:

Featured Image: Tapati Rinchumrus/Shutterstock