The world was recently introduced to GPT-3, the newest language prediction model created by OpenAI. The artificial intelligence research laboratory released their GPT-3 private beta to select developers, who got a sneak peak into the capabilities of the biggest and most advanced AI text generator ever created.
Developers have already used the technology to create applications that could benefit digital marketers. Need help creating new Google Ads headlines? GPT-3 can write dozens of them for you. In seconds. Need help creating content? GPT-3 can help you do that, and in whatever your preferred style.
Although it will be at least a year before GPT-3 is available as a commercial product, digital marketers and SEOs should familiarize themselves with this amazing technology and anticipate its potential applications to our industry. Here are 5 things SEOs and digital marketers need to know about GPT-3.
At its most basic level, GPT-3 is a text predictor. Give GPT-3 an example of what you would like it to write — whether a business report, a Dr. Suess poem, or any other type of text — and it will identify the language patterns and return a text completion.
The ultimate goal of AI is to be able to understand language to the point where it can perform human-like tasks with very little supervision. Those developers who were able to play around with GPT-3 were wowed by its ability to generate so many different types of text, perform specific tasks, and respond to a wide variety of prompts. When it comes to making a robot that writes, responds, and adapts like a human, the newest model of GPT-3 seems well on its way to accomplishing that ambition.
Although GPT-3 is already really good, it does have some kinks and lacks the reasoning to avoid glaring, simple mistakes. Sam Altman, one of OpenAI’s founders, even tweeted that “the hype was way too much.” OpenAI plans to continue refining the product before making it available on the market.
Until then, the rest of us will have to use our imaginations as to all the tasks that GPT-3 might be able to help us one day complete.
In its debut weeks, we saw programmers and developers use GPT-3 in some amazing ways. For example, GPT-3 can function as a search engine.
It can generate code.
It can do UI/UX design.
From writing poetry to translating, to composing legal documents, the ways GPT-3 can be harnessed are seemingly endless.
GPT-3 is much bigger than any other language model we have seen. It’s over 100 times larger than its previous version, GPT-2, and 10 times larger than its closest competitor, Microsoft’s Turing NLG. With 175 billion parameters, GPT-3 has ingested nearly everything published on the internet. It also cost around $12-million dollars to train.
To put the potential impact of that size into context, think back to Google’s BERT. Although BERT shares a similar architecture to GPT-3, BERT is significantly smaller. Last year, the BERT update impacted 10% of all search queries, providing more relevant results to users because of its more nuanced understanding of complex syntax and preposition use.
But unlike BERT, GPT-3 doesn’t require nearly as large of a training data set. It can parse and predict language from just a few pieces of input. GPT-3’s sheer size is the primary reason why it is far better performing than other models we’ve seen. It is essentially the most literary, well-read robot ever.
If BERT could have such a significant impact on the quality of the search experience, imagine how a model as large as GPT-3 might elevate the performance of those applications that are already relying on AI text prediction — like call centers and chat bots. A GPT-3 commercial product may soon provide those businesses with a newer, far more efficient product.
Although OpenAI was originally founded as a non-profit research center committed to making AI that works for the good of society, OpenAI recently established a for-profit arm. It is ready to commoditize GPT-3, and for good reason. It’s an incredibly powerful technology, and potential customers are already lining up to get their hands on it.
But as with all discussions centered on artificial intelligence, it is the power of the product that has caused alarm for some. GPT-3 has been trained by both the highest-quality content on the internet — think reputable online newspapers — and also the lowest quality content (Forums, blogspots, etc.)
Unlike a human, GPT-3 cannot evaluate whether the text it generates is offensive, discriminatory, fake, or even just flat-out illogical.
OpenAI has certainly created a model that sounds more human-like, but actual human oversight and curation is still necessary for the product to function correctly and “for the good.” OpenAI is aware of these limitations, and has made it clear they want to do their best to resolve them prior to allowing their product to be purchased by just anyone.
At least for now, their LLC is staying committed to their non-profit’s mission.
GPT-3’s direct application to the SEO industry is abundantly clear. No one has monetized machine learning and NLP as well as Google, and it’s been a long time since Google had an equally-matched competitor. A commercial GPT-3 product could change that.
Advancements in the field of natural language processing are the reason why when you prompt Google with any question or keyword phrase, it understands your intent and finds the web page it believes will best serve your needs. But it is not only Google’s NLP capabilities, but its immense indexing power, that catapulted it to it’s 92% share of the search engine market.
When GPT-3 (or some other equally sizable model) becomes available as a commercial product, it could equip other innovators and brands to develop search engines of their own. The search engine market could finally open up to new competitors who have similar technological advantages and cloud services, like AWS, Microsoft, or Apple.
As the field of natural language processing continues to progress, digital marketers could see entirely new search engines and advertising platforms emerge. GPT-3, and more products like it, could create seismic shifts in the realm of search.
Until then, Google’s role as the market leader is unlikely to change anytime soon.