What is GPT-3 and how will marketers be impacted?
Note: A version of this story ran in Episode 3 of my Marketing & The Machine podcast.
Explain GPT-3 to me like I’m 5
The term AI gets thrown around a lot. You could think of Siri or Alexa as Artificial Intelligence. So is a chatbot. You could say that the technology behind a Google search, Google Ads, and Google Photos is AI. All these technologies can fall under the big umbrella called Artificial Intelligence, which, at its essence, is building smart machines that are capable of performing tasks that typically require human intelligence.
You’ve probably used at least one of the technologies above, and I bet you’d agree that we have a long way to go before they fully mimic human intelligence. For me, Siri is just a $1,000 kitchen timer 😀. But a program by the name of GPT-3, which launched in June, might be the closest we’ve come yet to real human intelligence from a machine.
GPT-3 was created by OpenAI, an artificial intelligence research lab out of San Francisco that has funding from Elon Musk. This is the third iteration (and first commercial application) of the machine learning model the lab has produced...hence the three at the end.
GPT-3 is a neural-network powered language model. Now, stick with me here for a minute. I promise to ease into this. And the language model has trained itself on gobs and gobs of unlabeled data crawling across the web, including stuff like Wikipedia, Reddit, news articles, fan fiction, you name it. Something to the tune of 45TB of data and nearly one trillion words!
Now, in GPT-3’s case, the language model essentially predicts and weighs the likelihood of a sentence existing in the world. A typical example I’ve seen thrown around lately is that GPT-3 can label the sentence: “I take my dog for a walk” as more probable to exist than the sentence: “I take my banana for a walk.” In all that data, it looks at big chunks of content to make inferences on what is likely to come next. In this particular sentence, it’s learned that “dog” has a stronger connection and is more likely to exist.
In the same vein, if you were to give it just a single input word of “fire,” it’s more likely to follow back with the words “truck” or “alarm” over some other gibberish because it’s already made those connections in the data.
If you want, you can loosely think of it in comparison to how a baby learns to speak - it takes language cues and patterns from all different inputs, which initially probably feels like gibberish to them, but the wheels start spinning, and the brain starts connecting all those itty bitty neurons and synapses until full words and sentences form.
Now, GPT-3 has made connections and can do this for sentences and phrases and, generally, any sequence of characters. It can be used for all sorts of things that require, well….language. Things like answering questions, summarizing text, conversations, and more.
GPT-3 can do all this because in all the data it's ingested. It has made something like 175 Billion different connections! GPT-2 only had 1.5 Billion connections, and GPT-1 had 175 million.
How can marketers use GPT-3?
One day with GPT-3 you might be able to just write a few lines of an article or piece of content, and the program automagically fills in the rest. It could help with writer’s block and lower the barrier of entry of creating media even lower.
Recently a college student generated an entire blog, and one of his posts ended up hitting the top of Hacker News, a popular social site focused on computer science. The post wasn’t groundbreaking, but it did create buzz, and very few people caught on that the article was AI-driven!
It could write a trendy headline or even a viral tweet. It could help with your strategy or writing briefs. It could write your ad copy or even translate pages of your website to various languages.
GPT-3 can write functional code or query a database for you based on plain English directions.
“GPT-3, please make a green square button that is 100x100 pixels and has the words “Buy Now” written in yellow.” That’s crazy to me!
OpenAI is keeping access to it’s API invite-only for now, so common folk can’t just go download it from the internet.
But researchers who have been granted access have built imaginary conversations between historical figures, summarized movies with emojis, and written Dr. Seuss poems.
This is just the tip of the iceberg, my fellow marketers. It’s exciting and fascinating and a bit scary, so watch this space!
