We have all seen a liftman. In the mid-’80s and ‘90s, you would usually find them in large shopping complexes or residential apartments manually operating the elevator. Today they are a rarity because lifts have gotten smarter. We all know that someday someone will replace us at our current jobs. If you are fortunate, that will happen via a promotion or retirement. But the way the world is progressing that could also be a result of technology disruption.
The content writing business hasn’t seen much innovation in the last few decades. Yes, we have advanced workflow and content management tools to manage the process. But writing still requires the actual expertise of human beings.
Lately, the world of content marketing was abuzz with the news of Open AI launching GPT-3. This new language model uses deep learning to produce human-like text. So what is so significant about GPT-3 considering language models have existed for a while?
Why is GPT-3 a big deal for Content Marketing?
The answer lies in the model’s ability to get better each year. GPT-1, the earlier version of the model, was based on Transformer from Google, which had over 110 million weights. These weights are also referred to as parameters. They are matrices, arrays of rows and columns by which each vector is multiplied. The more weights you add to your model, it is tunned better to close the error gap.
GPT-3 covers 175 billion weights making it one of the largest neural networks the world has ever seen. Eclipsing its predecessor GPT-2, which had over 1.5 billion weights. For comparison, this new model beats the competition by a ginormous margin.
So if you keep the math aside, can this model really write human-like text? Earlier this year, Liam Porr, produced a fake blog post with GPT-3, which ranked at the no.1 spot in Hacker News. Porr was trying to convince others that the content produced by GPT-3 could easily fool people into believing that a human wrote it. If you read Porr’s blog post (with a pseudonymous author), titled “Feeling unproductive?” I’m confident you could easily pass this off as something that was written by humans.
GPT-3 can create any type of content you want with minimal tunning, as Aaron Jack shows here. You could simply type in your article’s title, and GPT-3 would write the article for you.
What Does GPT-3 mean for Content Marketing?
Let’s address this from the context of a content marketing agency if you look at a typical resource who works at a content marketing agency. Writing is only 50% or less of a specialist’s role. They also have to excel at SEO best practices, design, user experience, and outreach. Agencies train the resources well to ensure that they deliver maximum value to the clients. So what does the emergence of GPT-3 mean for content marketing? We cover the impact on both short-form and long-form content.
GPT 3 for Short-form Content Creation
With nearly 50% of the time being spent in writing content, content marketing agencies could leverage GPT-3 to create short-form content at scale. GPT-3 model really excels at writing low-value content at high volume. The applications of GPT-3 for short-form content could include the following:
- Product Descriptions
- Headlines
- Short-forms Ads
- Captions
- Link Descriptions
- Social Media Posts
Any short-form content that is repetitive in nature, like product descriptions or headlines, could be easily automated with GPT-3. Copy.AI is one of the many tools that has been recently launched that uses the GPT-3 model from Open AI to help with your short-form content needs.
GPT-3 for Long-Form Content Creation
When it comes to long-form content, the GPT-3 model still needs to get better. In an experiment done by the Australian Agency, Animalz found that the model cannot form a narrative thread.
As a writer, we are well aware that you have to thread together all component parts of an argument to form a logical narrative when writing long-form content. GPT-3’s writing meanders and skirts around the topic but fails to build a logical narrative.
One way to build a compelling argument is to show proof, which GPT-3 lacks in its current form.
How should you use GPT-3 for Content Writing needs?
Suppose you choose to use GPT-3 for your content writing needs. It is important to understand how the model works. GPT-3 is an auto-regressive language model that attempts to identify the next best word in a string of words based on the ones that come before. To do this, researchers fed the model with a vast trove of information, from popular Reddit posts to articles on Wikipedia. On its own, GPT3 is an impressive proof of concept.
But it is formulaic, only sounds great because of the data and parameters used.
GPT-3 or AI Generated Content always suffers from “sameness.”
Most tools that deliver AI-generated content suffer sameness. If you check for the headlines, subheads, and core points, they are all the same. That’s because to begin the process; you have to provide the tool with a seeding paragraph that you could have borrowed from the web or written on your own. In many ways, the model simply rewrites the article for you to make sure it doesn’t violate any copyrights.
Several tools already use the GPT-2 model to help you create AI-generated content; they include KafKai, Articoolo, and Article Forge. If you choose to use one of these tools or the GPT-3 model, your work doesn’t stop at generating the content. It starts by leveling the content to make sure that it forms a compelling argument and is unique to your industry.
Today the most common usage of AI-generated content is in the content outreach process. The agencies find it economical purely from a cost perspective, considering it improves the margin they earn from seeding guest blog content.
Sensitivity & Implications for YMYL Content
As pointed earlier, if you are writing content for industries like health, finance, or wellness, it is essential to support your content with facts. The content from GPT-3 also comes across as biased, so it is important to make sure you edit the content to make sure the final output is worth publishing.
Use GPT-3 as a starting point for your ideas
Most agencies have a minimum of five people working on any piece of content: a strategist, researcher, writer, editor, and copyeditor. Producing content also requires other skills, which is difficult for any AI model to replicate completely.
GPT-3 model could be a sparring partner when you have those brainstorming sessions with your team. You could type in some of your seeding ideas, and the tool will help you frame your ideas better. The model will provide you a new perspective based on its large trove of information.
Final Thoughts
As content marketers, I’m excited to see what capabilities GPT would bring to its fold in the next few iterations and what marketing choices it will present. Content writing is a time-consuming activity that requires help, whether it comes to researching facts or addressing different perspectives. GPT-3 could be your partner in the entire process.
I don’t see AI replacing content writers because writing will always need a human touch. You will always need exceptional writers to tell your stories. Stories that blend emotions into your content to stretch your audience’s imagination. I don’t see the content marketer turning into liftman anytime soon.
