If you haven’t yet heard of the transformational potential of GPT-3, you are about to have your mind blown.
GPT-3 is essentially the first artificial intelligence that can pass the Turing Test and write copy that is highly convincing as human. It can write about any topic and can be directed to write in any voice, style, or tone. The best way to think about it is that GPT-3 is essentially a professional-level copywriter that has the collective knowledge of the entire Internet.
What Is GPT-3?
GPT-3 is an AI language model developed by OpenAI. The model uses a revolutionary architecture called a “Transformer.” Transformer models are relatively simple, but what makes them powerful is that their skill and accuracy is directly related to the amount of data they were trained on and the length of that training. This is referred to as “Param Size.”
The above image references GPT-3 (175B parameters as well as two other smaller transformer models. As you can see from the chart, the leap from GPT-2 (13B params) to GPT-3 resulted in a massive improvement in accuracy. Given that there are already plans in the works for a 1 Trillion param model, we may well be on our way to a true proto artificial general intelligence.
What makes GPT-3 special versus its predecessors is that its large size has enabled something called “meta-learning,” where the model does not require “fine-tuning” (additional training) to accurately respond to a prompt. The prompt itself is enough for GPT-3 to figure out what you want.
As you can probably imagine, the implications of this transformative technology are just at the very beginning of being realized. What follows is my understanding of how GPT-3 will impact the many aspects of digital marketing. Get ready!
General Content Implications of GTP-3
GPT-3 will replace the bulk of low-level content creation, eliminating thousands of writers and hundreds of thousands of content marketing jobs, potentially leading to a crash in the content creation market.
GPT-3 will immediately impact the voice, tone, and flow of all copy being written. And by combining GPT-3 with a virtual assistant like Alexa, Apple’s Siri, or Google’s Duplex, we will reach the tipping point when copywriting becomes, for the first time, a fully automated job.
Copywriting as a profession will slowly become extinct, as will creative writing jobs.
Such an advance, if it occurs, would mark the beginning of the end for content marketing as a whole.
The resulting automation of the ghostwriting services market will have a similar impact on marketing agencies and in-house marketing departments.
GPT-3 will enable the creation of passive income assets in a similar manner to what happened with GPT-1, with the difference being that the skill required to write the content is vastly higher.
The GPT-1 model required a relatively high threshold of writing skill. As a result, the majority of the value was created by people who possessed the skill to write unique content. GPT-3, however, requires a much lower threshold of skill to create content.
This means that anyone who can “copy/paste” can do the job of writing effectively. The value of GPT-3 lies in its ability to do the job of a writer so well that there is no longer a need to have a writer.
In-house marketing departments, as well as agencies, will increasingly use GPT-3 to do work that was previously done by people, replacing both of these roles.
To be clear, I don’t believe that all content marketing jobs will be eliminated.
In fact, I believe that there will be a need for people who specialize in understanding the role of content in the larger marketing strategy, and who have the ability to create, manage, and optimize the content that GPT-3 is writing.
But, do not confuse this with the need for someone to write the actual copy.
Everything above in italics was written by GPT-3. The prompt given was simple: I provided the introductory paragraphs written above.
As GPT-3 mentioned in the previous paragraphs, new state-of-the-art language models present an existential risk for any writing with quality at or below the quality of what GPT-3 has been shown to achieve.
GPT-3 has already demonstrated it can easily fool people into thinking it was a human writer and has the ability to write about virtually any topic. It even consistently demonstrates proficiency with niche esoteric topics. Its command of spelling and grammar is impeccable, and it seems to make considerably fewer of these types of mistakes.
Given these skills, GPT-3 could be said to have the ability to write as well as most mid-tier writers, but at 500X the speed and with instant access to knowledge that would take a writer a considerable amount of time to research themselves to become familiar enough with to write an accurate article.
When GPT-3 becomes ubiquitous, it will be possible to get the same quality article you might have paid $200 for written for under $1 (or perhaps $5 for GPT-3 creation and then editing by a human). This will also result in a massive increase in the total amount of text output online — perhaps an order of magnitude or more, increasing the competition considerably, especially in the long tail.
One can also imagine new ways that GPT-3 generated text can be combined with other forms of AI or procedurally generated media. Here are a few things we are likely to see (some of which may proliferate incredibly).
GPT-3 + Text to Speech
AI-generated content can also be made into podcasts/audio on YouTube. Services that do automatic generation of audio content will likely pop up, offering AI podcast creation about any topic, potentially exploding the number of podcasts available.
For instance, it would be easy to set up a podcast where GPT-3 pretends to be, say, Abraham Lincoln, and comments on daily news headlines as an automated GPT-3 to text-to-speech and then uploaded as a podcast each day.
These AI-generated podcasts will generally proliferate in niche markets, with a few rising to notoriety and general popularity.
GPT-3 + Video Generators
Tools like Vidnami take text as input and generate videos made up of stock video footage. Each scene is chosen by Vidnami’s AI based on the context of the text input. Combining this tool (or even more sophisticated versions of Vidnami) with GPT-3 would allow for scalable conversion of GPT-3 generated text, to engaging videos. Here is an early example of what is possible:
GPT-3 + Data Journalism
Surprisingly, GPT-3 also has some fascinating abilities to understand and parse data. For instance, it can generate the code for data visualizations:
I already knew how GPT-2 worked thanks to @JayAlammar blogs and the papers. I also knew GPT-3 was the same architecture with different(huge) size. But still, good to have a refresher.https://t.co/LubAFv6RzD— nutanc (@nutanc) July 22, 2020
GPT-3 also has deep knowledge of where specific types of data may exist and can be used to make recommendations for starting points for data journalism projects beginning with any topic. Below is an example of a GPT-3 assisted data source search:
1. Data Sources
a. Data source and relevance to Sr. Citizen Issues
b. Data source and relevance to Sr. Citizen Issues
US Census Data http://www.census.gov/ This is a great source for US senior citizen data. Figures are available from 1850 to present. The Census is a great place to start to find out where senior citizens live, why and what they do.
Social Security Administration (https://www.ssa.gov/) The SSA is primarily a pension and retirement program for US citizens, but it is also a good source for US senior citizen data for those who qualify for benefits. The data is not available on the SSA website, but you can access it through the US Social Security Administration Freedom of Information Act (FOIA) website (https://www.ssa.gov/foia/)
National Center for Health Statistics http://www.cdc.gov/nchs/nhanes.htm The NCHS is a good source for American health-related data. This source has data for the general US population, as well as data broken down by demographics and geographic location.
US Department of Veterans Affairs (http://www.va.gov/) The VA is a good source for US military data, including data for US veterans. They also have a section of their website that focuses on issues affecting US veterans. There is also a wide range of VA data available on their website.
US Bureau of Labor Statistics http://www.bls.gov/The Bureau of Labor Statistics is a good source for data about the US economy and the labor force.
The US National Network for Caregiver Support (NNCS)http://www.caregiver.org/support/nncs.htm The NNCS is a good source for caregiver data.
** Bolded beginning text was the prompt for GPT-3, the rest of the text was generated by GPT-3.
GPT-3 + Memes/Image Macros
Image macro memes are perhaps the most widely shared and viral form of content that exists online. Their role in communication, persuasion, and propaganda online can not be understated.
In the example below, we can see GPT-3 can proficiently “meme.” Given this, a GPT-3 generated meme explosion seems all but inevitable.
Tired: Making your own memes— Mrinal Mohit (@wowitsmrinal) July 25, 2020
Wired: Asking @OpenAI's #gpt3 to make memes.
Amazed to see how much of cultural subtext and nuance language models can pick up on. cc: @gwern @gdb pic.twitter.com/eBrFAWiZhA
Search Marketing Implications of GTP-3
While content creation itself will be completely transformed by the wider adoption of utilizing GTP-3, other aspects of digital marketing will be dramatically affected, as well.
Nearly every aspect of the way we market brands and products online will be influenced by GTP-3, whether we decide personally to use it or not, because it’ll have such a profound impact on the overall landscape of content, SEO, advertising, and more.
GTP-3 + Search Engines
With Google announcing that nearly 100% of U.S. searches now use BERT (a transformer network like GPT-3), it’s pretty clear the future of search will incorporate this technology to its eventual limits. The biggest potential impact could be in the form of Google significantly improving its ability to generate dynamic answer results and potentially replacing website snippets. Additionally, transformer architecture is not proprietary. Within the next few years, many companies will be capable of creating models that might easily rival Google’s quality with BERT. If this happens, Google could see its historic place as the highest quality search engine erode.
GTP-3 + Search
Content and SEO are inextricably linked. As I’ve already mentioned, GTP-3 can mean the creation of very high volumes of content at a reasonable quality level for a cheap price.
If you pair this content production with a tool like Clearscope.io, which helps you identify what keywords, word count, etc. you need to top the competition in the SERPs, GTP-3-generated content stands a good chance at ranking.
In my opinion, GPT-3’s valid implication for SEO is in its ability as a writing assistant, helping to improve speed on writing high-quality content. Unfortunately, it seems to be that many or even most of GPT-3’s applications might be negative for the overall health of the search ecosystem. Many of its features can help users in a scalable form of spam of all types we have never seen before.The primary GPT-3 black hat SEO uses will be:
Massive blogspam sites (at least in this case the content will be written by GPT-3, so it won’t be useless, but it’ll be riddled with inaccuracies if not edited by a human
Massive human undetectable review manipulation via scaleable creating of reviews
Fake news generation for link building or social media purposes
How will brands build trust and emerge through the endless amounts of content being generated in their industries? These are questions we need to ask now so we can properly prepare.
GTP-3 + Ecommerce
GTP-3 can allow you to create product descriptions at volume for ecommerce websites. Creating thorough, engaging product descriptions is a crucial but time-consuming facet of ecommerce marketing, and being able to drastically cut back on the time and effort invested can mean a boost in search, conversions, and other aspects of your marketing that you’re now able to dedicate your time to.
But there’s a negative side to GTP-3’s potential impact, as well. It’s likely some marketers will generate product reviews en masse, meaning the dilution of true, human reviews and thus the deterioration of trust in any reviews across the web.
The ecommerce sector already struggles with this lack of trust to a lesser degree, so it’s vital that they explore ways to break through these authenticity challenges before the problem gets exponentially worse.
While this article certainly isn't comprehensive or fully predictive, it’s become clear to me that GPT-3 is going to be a major catalyst for change across content marketing and search. It’s important for practitioners to stay informed and to explore the implications of this tech for themselves because they will shortly be encountering it in many different forms.