OpenAI has released yet another evolutionary variant of its AI, billed as ‘too dangerous for public release.’
But now, after over 6 months of GPT-3 AI being in our lives and with millions of words written, has the manual copywriting world taken a fatal hit? Or, on the contrary, has GPT-3 become the helping hand we’ve been looking for?
At least you can be sure this creative and self-deprecating introduction was written by a human, can’t you?
What Is GPT-3?
GPT-3 is the third generation of ‘generative pre-trained transformer’ neural network made by OpenAI, the company founded by Elon Musk with a mission to make the future possibility of general artificial intelligence – well, one that still has humans in the picture.
So, what is it again? It’s like a chatbot, which can write natural-sounding text, all from a couple of prompts. It mimics the style of the prompt, whether it’s dialogue, code, or even a formal encyclopedia article.
A general artificial intelligence, by the way, is what many people think of as AI. Think Skynet, Sonny, HAL 9000, The Cylons, Cortana, take your pick.
AI isn’t anywhere near general intelligence levels yet, but we can all agree that it’ll be one for the books when the day comes. For now, generative pre-trained transformers still have some pretty impressive capabilities.
What Can GPT-3 Do?
GPT-3, for example, has been trained on large chunks of the internet, including purportedly all of Wikipedia, and then given new tasks to perform. These include generating news articles, fiction stories, poetry, recipes, and so forth.
And while these are great examples of things that computers could never previously accomplish, they’re still far removed from anything resembling real life, and writers still need to edit them heavily so they make sense.
So why should anyone care about GPT-3 at all? Well, because it shows us where the technology might go next.
GPT-3 is the latest neural network in a long history of previous systems that were notably less capable. For example, IBM Watson famously beat two champions of Jeopardy! back in 2011, and Google DeepMind created an algorithm called AlphaGo that defeated Lee Sedol, arguably the greatest Go player ever, in 2016.
Both of those algorithms had their own limitations, though: They couldn’t generate any content beyond simple facts, nor did they know how to tell a story. That’s where GPT-3 steps in.
The first version of GPT-3 came out last year, and it generated more than 10 million unique sentences. This time around, however, OpenAI says it’s training on 100 times more data, meaning there will likely be much better results.
How Does GPT-3 Work?
To train GPT-3, OpenAI uses a technique known as reinforcement learning. Basically, it gives the system lots of different prompts and lets it figure out how to answer each question based on past experience.
Then, after every batch of questions, it rewards the model if it answers correctly, and punishes it if it doesn’t. Over thousands upon thousands of iterations, the machine learns to do just that.
You may also want to read: How Smart Marketers Are Using AI To Boost Their Businesses
I’m Ready For GPT-3 Now!
If you want to check out GPT-3 now, you can head over to conversion.ai, rytr.me, or writesonic.com. Each of these three popular tools integrates the GPT-3 AI into their tool, in slightly different ways. Because you can customize the prompts to GPT-3, the tool with the best prompts essentially wins.
However, GPT-3 sounds remarkably similar when given the same prompt, regardless of what tool you use.
As far as pricing, most GPT-3 tools will cost you from $25/month for starter plans, good for about 10,000 words or 5 blog posts, and ramp-up to $100/mo for unlimited or premium plans.
If you write enough and don’t mind the caveats, it might be a great option for you. But, let’s look at the downside first, before taking the plunge.
AI Blog Post Writing Is Shockingly Good…But
The grammar of GPT-3 is near perfect.
But GPT-2, on the other hand, would really have trouble sometimes. It could get stuck in loops and randomly would repeat itself or go way off on a different type of text.
In a statement published in the journal Science, the researchers said: “The unicorn’s DNA contains a unique genetic signature that suggests it used to live in a remote valley for millions of years before it was discovered.
“The findings in this study and others are important as they contribute to understanding how the unicorn evolved and how it has evolved over the last 2,000 years.”
Scientists believe that the unicorn has been living in a remote valley for several months.
The researchers believe that the unicorn has been living in a remote valley for several months.
As you can see, the unicorn has certainly been there for months (or perhaps thousands of years- also?)
And GPT-2 thought it was so important, it went on to repeat the same concept many times. GPT-3 gets stuck in a loop much less often, but there’s an important concept to notice here.
GPT-2 would only output 2-3 lines of text at once. While GPT-3 might output 5-6 lines, you still run into the same issue. After this 5-6 line block of text, GPT-3 might just repeat itself. It won’t always say the exact same thing, but conceptually, you’ll have something useful, followed by a block of fluff.
GPT-3 might also have trouble differentiating between writing styles and get itself on a style tangent quite easily.
For example, it’ll jump from narrative dialogue right into a listicle with prices, pros/cons, and even (fake) reviews. Compared to GPT-2, GPT-3 stays mostly in the same form of writing, and if you ask for narrative dialogue, it’ll try its best.
But if a lot of people are writing listicles about your subject, GPT-3 will be strongly compelled to switch to listicle format, even if you wanted it to be in a different format.
Perhaps the most difficult thing for a GPT-3 editor is the algorithm’s struggles with conceptualization.
You might read something GPT-3 wrote and think it sounds perfect until you think about the conceptual story it seems to be weaving. Then, although the grammar side of your brain loves it, the logic side of your brain hopefully realizes that it’s a lump of nonsense.
You may also want to read: Here’s How To Improve Your Content Marketing Strategy
GPT-3’s Achilles Heel
GPT-3 has read the internet (or the most popular pages anyway), so it can only create things about the same depth and voice/tone as most things on the internet. So if you want to write a post that’s significantly more in-depth, or original, you’re out of luck.
Secondly, GPT-3 can lie and misrepresent factual information – or more precisely, it can repeat these if they are written enough. This can be, well, bad. For example, if a lot of people say x, GPT-3 might say x too, even if y is true and can be easily fact-checked.
For example, GPT-3 has been known to lie about how long it takes to train its neural net models (presumably in a shallow attempt at vanity). In reality, the training time varies greatly depending on hardware, software, etc., but GPT-3 says it takes around three days.
That’s not entirely accurate, but it gives some insight into why GPT-3 doesn’t work very well when asked to do anything outside the norm.
It’s worth noting that GPT-3 isn’t actually lying; it’s simply misinformed.
AI Writers Will Never Be Able To Do This One Thing:
Be a thought leader. GPT-3 is good at paraphrasing other people on the internet. So-called generative design is very good at rehashing, even to the point that it seems very organic (in a creepy-cool way).
It’s not the same as contributing original work, including your own research, your own story, and your own emotions.
If you’re a content manager, you know that it’s hard enough these days to find a human writer who can produce quality content without having to worry about plagiarism.
But finding someone who can both write and contribute their own ideas while being able to tell stories and convey emotion is nearly impossible. With GPT-3, it really is impossible. Every output from the algorithm is a rehash of something else, a big-data average.
While there’s certainly a place for ‘averaging’ the information on the web, creating more comprehensive, get-everything-in-one-place kind of posts, it’s not what human writers should be focused on.
Writers should focus more on ‘uploading,’ or creating from their own minds, rather than cloning other content.
GPT-3 Is Perfect For…
Headlines
GPT-3, and probably GPT-4 and beyond are going to be awesome for some really useful things. For example, GPT-3 makes great headlines (figuratively and literally: generating a headline that is catchy and summarizes the intent of an article, from a given description)
Summarizing Text From a Website
GPT-3 is very good at summarizing text. You could use this to summarize articles, news items, blog posts, tweets, whatever.
The results will usually sound like they came straight off the page.
You’ll have to decide whether you want to keep the summary short or long and whether you want to include links back to the source material.
Generating Unique Text For Emails
Email marketing campaigns often involve sending messages with pre-written copy. If you don’t want to spend hours writing up a new email copy every week, then using GPT-3 would allow you to generate fresh, personalized messages quickly.
This also works well for newsletters, where you need to send out regular updates.
Giving Writers Ideas That They Can Riff Off
You can be part of a GAN, or Generative Adversarial Network. This is sort of like a group of friends that you know well, but who might disagree with you a lot. You get a lot more ideas by riffing off someone (or in this case something) else, compared to just producing text on a white page.
The Number One Absolutely Perfect Utility For GPT-3
It kills writer’s block. Writer’s block is no longer… a thing. If you’ve ever been stuck trying to come up with a topic idea, or even worse, a title for your post, then GPT-3 is going to save you time and effort.
Many tools allow you to write your own text, then hit “write,” and GPT-3 will take the reins for a bit. Even if you don’t like what it wrote, it’ll still 100% of the time give you an idea or two about what to write next.
You may also want to read: 3 Ways You Can Utilize AI To Scale Customer Acquisition And Retention
So, Writing Is Over. Now All Writers Will Eventually Be Just GPT-3 Editors?
Well, no. GPT-3 isn’t good at long-form text (yet).
It tends to still go off on a tangent after 200+ words generated downstream from an input prompt. But, as we move forward into the future, we’re probably going to see people start to experiment with ways to make GPT-3 produce better quality long-form text over time.
Don’t most writers become editors at some point in the progression of their career anyways?
We’ve yet to discuss perhaps AI text generators’ biggest flaw:
AI is detectable (if not by humans, then certainly by other AI).
Recently, Adobe trained an AI to detect deep fakes or manipulated images, and MIT has developed an algorithm that can detect AI text generation, that could someday be used to systematically take down GPT-3 written articles from the internet.
There might be a wave of mostly AI content coming, but at some point soon, human-powered content might become even more highly valued, because the AI content will be cheap and plentiful.
In other words, human writers will have to go up the value chain.
What Is Content Marketing Like In 2021 and 2022?
Right now, there’s a gold rush of content marketing. Like many times in the past, it has become somewhat automated or easy to beat Google’s algorithms and make a ton of money by getting your pages to the top of Google, mostly with AI.
Some other blogs and websites have awesome human-made content too:
- Hubspot’s blog – which ranks on the top spots on Google for “ultimate guide” for zillions of different topics.
- Canva’s Design School – which is good enough to get you a certificate, but also includes an article on almost any design concept imaginable.
- 99Designs Blog – another mega design site, which covers every aspect of design in near-encyclopedic detail.
- Vectornator Blog – design-related articles that touch upon art styles, design tutorials, and curated content. A can’t miss for any budding designer
- Animalz – one of our favorite blogs on content marketing. Real thought leaders that get deep into topics almost no one finds the time to get into on the internet.
If I Were Google, What Would I Do?
Every time such a disruptive technology has emerged in the past, Google has course-corrected, penalizing companies that try to cheat and ultimately lower relevancy to searchers.
Google doesn’t want you to use bots to generate content. They don’t want you to pay people to write content.
And they definitely don’t want you to buy ads to drive traffic to your website. Ok, to be fair, they do want that.
But in all seriousness, what would happen if Google did nothing about this new threat?
I think the answer is pretty obvious.
The quality of search results would plummet. It wouldn’t matter how much money you spent on SEO, PPC, social media advertising, etc., as long as companies were using AI to game the system, real content would suffer. Content marketers would end up paying through the nose just to stay relevant.
I’m not saying that Google should ban all forms of automation. But it’s very likely that Google will implement a GPT-3 detector and penalize sites using AI-generated content.
And again, like when the Penguin (2012) and Hummingbird (2015) Google core algorithm updates went live, the shape of the internet might change drastically overnight.
You may also want to read: Topic Clusters In Content Marketing: Here’s What You Need To Know
For Now, What Tools Are Best For GPT-3?
Now that you’ve been warned, check out these tools to jump into the GPT-3 world:
Conversion.ai (now called Jarvis.ai)
Conversion.ai is the market leader of GPT-3 tools and includes long-form content writing, as well as many different marketing frameworks.
It’s the best GPT-3 writer on the market, partially because Conversion.ai has acquired many of the previous leaders, such as Shortly and Headline.
- Pricing for the starter version is $29/month and limits the number of words to 20,000 per month.
- From there, a Pro plan at $109/month provides unlimited words,
- And the Boss Mode at $119/month offers the most customization and long-form writing features, like extended character look back, where the AI ingests more of the previous text to understand how to write the next block.
Neuraltext
Neuraltext is a content optimization platform that integrates many of keyword research software’s content optimization features, including an AI text generator.
It’s free to get started, using keyword research and content optimization.
- If you want to get started with GPT-3, what Neuraltext calls “smart copy,” upgrade to the $49/mo. plan for 200 ‘runs’ per month.
- Then, for $119/mo, jump up to the Pro plan, for more smart copy, a lot more keyword research, and unlimited document management.
- Unlimited smart-copy is an extra add-on for $39/mo.
Rytr
A relative newcomer, and recent AppSumo contender, Rytr is notably inexpensive compared to the alternatives. It produces content in a range of 20+ different use cases, pretty standard compared to the other tools here.
Pricing starts at a free plan with 5000 characters per month and moves up to the premium plan at $29/mo for unlimited usage, which is basically a steal compared to any other GPT-3 writer on the market.
Writesonic
Writesonic has some of the most natural-sounding headline generation we’ve seen with GPT-3. It also has a nice variety of customization tools and even writes full-length blog posts (around 800 words).
- The Free trial gives you 10 credits (runs), where each credit is around 300 characters.
- Moving up to starter for $13.05/mo, you’ll get 75 credits.
- For $44.55/mo, you can unlock unlimited credits with 150 credits on the long-form writer,
- And for $224.55/mo, with the Business plan, you’ll get up to 1200 credits on the long-form writer.
Don’t trust this
The info above is probably out of date already, just due to the sheer volume of changes happening in the AI world. We tried our best!
To wrap up, there are plenty of options to dive into GPT-3 AI writing, and many are maturing to the point to be useful for professionals, so choose wisely.
Keep in mind that GPT-3 isn’t for everything. If you’re planning on making lots of SEO content, you might want to hold off until Google has responded as to how they’ll proceed with the flood of AI-generated text on the web.
As for the tools, I’d recommend starting off with one of the cheaper ones first, then moving onto the others once you have enough experience under your belt.
For now, AI writing is a content creator’s gold rush, full of opportunity, risk, and the whole enchilada that comes with rapidly evolving, disruptive technologies.
It’s up to you whether to take part or sit safely on the sidelines, but in any case, we’re witnessing one of the most fundamental shifts ever to occur on the internet before our eyes today. Stay safe out there!
Sign Up For Our Mailing List
If you’d like to receive more in-depth articles, videos, and Infographics in your inbox, please sign up below. We’ll also keep you abreast of our upcoming soup-to-nuts blogging class.
Sign up for the newest articles from Curatti, delivered straight to your inbox
Featured image: Copyright: ‘https://www.123rf.com/profile_paulfleet‘ / 123RF Stock Phot
CURATTI GUEST