There is an average flood of content coming to the internet the likes of which we haven’t seen. What if you could produce 10 times the amount of content and at 10x the cost savings, what would you do? Even if the content is mediocre, will you still be tempted to benefit from being able to throw content over the well and see what gets stuck?
What does this mean for websites, link farms, private blogging networks, link generators, SEO and search engine algorithms? What could the deluge of poor quality, reliable original content mean?
What is GPT-3 and how does it work?
GPT-3 stands for Generative Pre-Trained Transfomer. According to Wikipedia:
GPT-3 is a self-regressive language model that uses deep learning to produce human-like text. It is a third generation language prediction model in the GPT-n series (and successor to GPT-2) created by OpenAI.
As a natural language processor and generator, GPT-3 is a language learning engine that crawls existing content and code to learn patterns, recognizes syntax and can produce unique outputs based on prompts, questions, and other input.
But GPT-3 is more than just used by content marketers, as OpenAI’s recent partnership with Github to generate code with a tool dubbed “Copilot” attests. The ability to use auto-regression language modeling applies not only to human language, but also to various types of code. Outputs are currently limited, but their potential use in the future could be extensive and influential.
How is GPT-3 currently being kept at bay
With current experimental access to the OpenAI API, we have developed our own tool on top of the API. The current application and application process with OpenAI is rigorous. Once an application has been developed before it is released to the public for use in any commercial application, OpenAI requires a detailed submission and use case for approval by the OpenAI team. Among the approval requirements are limitations on the types and lengths of output that are allowed to be pulled from the API.
For example, the company currently prohibits the use of OpenAI on certain social platforms, including Twitter, with the belief that massive tweets being mass-produced can be used for nefarious or political purposes and to influence or form public opinion may not be accurate.
In addition, OpenAI also limits any tool that uses the API from output greater than 200 characters. With a mission to serve a purpose much higher than producing mediocre content that humans likely won’t read.
Maintaining strict controls over an outrageously usable trial product is more than clever, but that doesn’t mean potential violators won’t find a way to circumvent the rules.
Examples of large-scale GPT-3 content
Since we developed our own tool on the OpenAI platform, we have used it extensively in-house, and tested it on some of our projects and client projects. Here are some examples where we’ve found it most useful in creating content that may cost more and require more resources to implement:
- Landing pages at scale. While the tool isn’t very talented at creating blog-type content, it’s actually rather smart in its ability to create landing pages for things like the ‘sites’ and ‘industries’ being served. We recently tested this by creating over 1,100 city and state landing pages for an in-house project at BIKE.co where we trained several third-party helpers on the tool and mentored them on how to deliver instant GPT-3 output to a refined Elementor core design on WordPress.
- podcast intros. We’ve found podcast introductions – for ourselves and clients – that can be more easily produced with GPT-3. To make it even scarier, we tested the AI-powered audio technology of the podcast’s audio itself. Imagine that, an entire podcast show where no humans create any content!
- Social media. While there are some current limitations on the length and type of format where GPT-3 can be used, there is a real possibility
- Spam. Spam algorithms currently pick up patterns in email messages, particularly with regard to copying. This is one way that AI/ML is used to filter out spam emails, but if it is not controlled, a large number of unique emails can be sent separately with less potential to be flagged as spam.
- Yarn content. Since an API can produce longer, unique outputs with simpler and shorter inputs, the ability to rotate and recreate similar content for use in online publishing is a real temptation, even if you have to put them together to make it happen.
These represent only a small potential for the uses (legal and illegal) of GPT-3. While we are currently scratching the potential surface of how this AI tool will affect us, there are those whose motives, while not per se negative, will still use the tool to create a flood of content that adds little or no value other than just providing content Online for content.
Why large-scale content will destroy the current state of the Internet
20 years ago we joked that you should be careful about the facts you thought were pulled off the web. New technology may actually take us back to a bygone era when facts were more fluid and the quality of content was worse rather than better. In fact, it is estimated that 7.5 million new blog posts are created every day. Imagine if machines could do this in the cloud using just a simple algorithm?
The content will be similar to how Syndrome in the Disney movie “The Incredibles” describes its plan for a post-superhero world where it will provide machines that will make everyone special:
When everyone is great, no one is.
This is exactly what is happening with GPT-3’s ability to provide content at scale.
When anyone can create content at scale for little or no cost, the only thing that will differ in the future is quality. In short, I agree with OpenAI’s feeling that strict controls should be placed on the quantity and purpose of content produced by GPT-3. Otherwise, we will have a lot more when it comes to content written on the web.