There’s a new content writer in town and its name is GPT-3. I’ve spent the last week exploring this new language AI and seeking to understand how this new powerful technology from OpenAI will transform the world of blogging.
First, I had some fun with it. One of the texts I generated was an interview between me, the AI, and a niche blogger. The two of them had some opinions about an AI writing Niche Blogger’s posts.
Then, I went out looking for just what this robot was capable of and found that it’s already been quite active around the web. The best way to really understand the GPT-3 “brain” is to read some of its work. Here are some examples of how it has been used:
Now that we’ve seen what it’s capable of, let’s dig in to how it might be used moving forward.
GPT-3 is a machine learning system that is able to generate many types of written content. The AI learned by crawling the Internet for 45TB of text data, nearly 1 trillion words. It takes only a short prompt to produce sensical, grammatically correct content that is sometimes indistinguishable from that of a human writer. Check out GPT-3’s response to the prompt: “What matters to you most, and why?”
Not all responses are quite this eloquent. Here’s an article which the author claims to have been written by GPT-3. As you can see it’s pretty boring, repetitive, and not exactly insightful. From my attempts to play around with the AI, my experience is that most texts are more like the latter.
GPT-3 doesn’t only generate article text. It is capable of writing code from plain English directions, and performs complex search functions that actually provide direct answers based on questions rather than search terms.
For all its hype and genuine power, GPT-3 has many flaws and shortcomings which we’ll discuss in more detail. Even its own creator has expressed that the GPT-3 is only a glimpse into AI’s full potential.
As of today, GPT-3 is in closed Beta, and when it is released in August, it will be a paid API, though its pricing has not yet been made public.
From what I’ve seen, GPT-3 has the ability to write blog posts beginning to end, following a logical progression and in perfect grammar to boot. The posts make sense, with the AI generally understanding and expanding upon the prompt. After a few tweaks and some manual handling, the text is as good as if written by a human, minus some factual errors or mishandled nuances (it is a machine after all).
My first prediction is that, with the public release of the API, the barrier to generating blog content will drop and in the next two years we’ll see a dramatic increase in the number of blog posts published to the web. It’s hard to imagine a bigger flood than 4.4 million blog posts published per day.
Blogs that struggle to publish regularly will turn to AI to fill gaps in content output. Companies looking to cut corners on marketing will fire writers and rely on low-paid workers to feed and tweak the AI. Niche bloggers will use AI to fill their blogs with long-tail keywords. And most of all, bloggers will use it to get over writer’s block by inputting short prompts and generating longer, fleshed out texts with plenty of fodder for editing and expansion.
Of course, as with all content, the number one pain point bloggers will be looking to address with this technology is traffic. More content, better rankings, more traffic. This leads to my second prediction.
Google’s search engine will implement reverse GPT-3 to flag articles that have been machine-written. One of the awesome things about GPT-3 is that it is capable of detecting if an article has been written by GPT-3. It probably won’t happen right away, but my guess is that Google will begin flagging machine-written articles and penalizing them. And for good reason.
Articles written by GPT-3 don’t provide as good of a user experience as one written by a professional blogger, or a true thought leader. By definition, AI is a regurgitation of what has already been written on the web. It provides no truly novel insight. And occasionally, if not always, it will miss nuances and details that separate an information aggregator from a thought leader.
Which brings me to my final prediction: The gap between true thought leaders/innovators and information regurgitators will actually widen, making it easier to spot those who actually put effort and careful thought behind a subject, and those who prostitute words to get clicks.
This last prediction may seem counter-intuitive. You might think that the easier is it to write, the shorter the gap between prolific innovative writers and others. But, think about it in terms of true painters vs the countless paint-by-numbers products that Facebook incessantly serves ads for. Anyone can paint by numbers, and it doesn’t look terrible, they’re nice designs and lead to palatable visual stimuli. But, would you pay $1000 for a paint-by-numbers painting?
It’s undeniable that bloggers will use GPT-3 to create content. In the early days of its existence, some might even temporarily get ahead because of it. I can imagine a niche website pumping out daily content, the blogger in the background grinding to get links, distributing on social media, building a following and selling the niche blog for a few thousand dollars. Rinse and repeat.
But, soon Google is going to start catching on. The quality of these posts will be low. The user experience will decline and they will possibly slap algorithmic penalties on these blogs.
For those who are looking for a long-term, sustainable partnership with AI to enhance their writing, rather than outsource it entirely, GPT-3 makes for a promising partner. It facilitates the process of writing by essentially aggregating a bunch of relevant information from across the web and putting it into a readable format. Instead of surfing the web for hours to get your creativity going, you can simply enter a few seed keywords or thoughts and let AI inspire you.
Personally, I’d stop there. Extract good points and expand on them. Add original thoughts, opinions and personal perspective to them. Give it flavor and personality. Add some expletives. Make it human before hitting publish.
It’s not just about being ethical, it’s about being compelling. People want to read and experience other people’s original thoughts. As a society and species, we want to further our intellect, our well being, our innovative spirits. That can’t be catalyzed (yet) by a machine. A long lasting, loyal audience will follow thought leadership, not fact regurgitation.
I think this is the high bar. This is the gold standard.
For those who will ignore this standard and publish semi-automatically AI generated content, the least we should require is that a disclaimer be added to the article, especially if the article is published under a company’s or a person’s byline.
When a person publishes a piece of content on, say, SEO, it is usually to sell services in some direct or indirect way. We assume that the person publishing the post is knowledgeable of SEO and therefore we can trust them with our SEO problems. The problem with not including a disclaimer is two-fold: 1) A searcher who knows nothing about SEO may come across a post generated by AI and be unable to pick out the inaccuracies inevitably included by the system. They would go out and implement the post’s advice blindly and possibly screw themselves over. 2) The person who published the post may not be as knowledgeable as the AI in SEO, misleading the reader on their expertise.
This may not always be the case. It could be that the publisher is quite knowledgeable, but lazy. However, it should be left up to the reader to determine what level of risk they are willing to take on someone who publishes AI generated content. It’s a sort of new kind of human-on-machine plagiarism – the source should be mentioned.
Its creators have been quick to respond to the potentially harmful implications of their system, and some solution to this problem (other than centralized moderation) still needs to be implemented.
Since the model will continue to be biased for the foreseeable future – there likely won’t be a fix before the public API is released. It’s on the users of the model to do their due diligence in ensuring that the content produced for them and published by them does not contain harmful biases.
Eventually, I’m sure, there will be some sort of filter or flag implemented by large platforms that house AI-generated content to mitigate the trolls. Until then, expect to see a rise in much more sophisticated troll bots (my 4th prediction, I guess).
GPT-3 is undeniably powerful. And as a blogger, you will likely use it in the near future, either by your own accord or via one of your friendly neighborhood blogging tools (this is a shameless plug).
I have no doubt that it’ll be an asset to bloggers of all kinds, from beginners in the niche world, to thought leaders in huge industries. How the tool is used and the level transparency around its use will ultimately separate the true innovators worth following from the wannabes who use machines to think for them.