The 3 AI "Archetype" Models that Make Up Most Business Use-Cases

In the vast AI landscape that we face today, which is actually much less complicated than it looks as we learned on day one, there are only a couple of "archetype" models you have to worry about for most of the use-cases your company will have.

Our Story Begins with Charles

This is Charles. He's British. He frequents only the most exclusive clubs in London's West End. He drinks the most expensive brandy and smokes the most expensive cigars.

Charles is a Deep Learning model.

 

Charles has been around since the mid-2000s though he is descended from a long line of AI research in neural networks dating back to the 1960s and 70s. He's wonderful at identifying patterns within data that a human might miss. He can often make very accurate predictions on data he's never seen before based on his training. He can be used on all kinds of data (numeric, textual, images, video, sound).

Charles has always been expensive. He is extremely specialized to specific problems. He requires a large and clean data set to train. And he has some complex math at his core that usually means only highly qualified people can figure out how to build him.

The big companies like Google and Facebook have been using Charles for the past 20 years, but generally he has been unavailable to anyone with a smaller technology budget.

Then Some Translation Nerds Came Up with Something That Affected You

In 2017, a Google team working on a German to English translation problem discovered that they could get much better results by applying a new kind of architecture called a transformer. They published their results and soon the AI community began to realize that they could use this same architecture in other areas.

While the new architecture was crucial, another concept began to become more common, the idea that AI models could be "pre-trained" by experts on a massive amount of general data (basically the entire Internet), and then released to the common people to use as they saw fit.

Over the course of 5 years, the AI community worked feverishly on producing new models based on these ideas. The models that affect your business split into two main archetypes.

Archetype 1: Mae, the Image Generation Model

The first archetype, a generative image model called Dall·E, shocked the world in 2021 (pioneered by OpenAI). The now perhaps more famous StableDiffusion arrived in 2022.

We'll call her Mae.

This archetype AI model is great at producing images and editing images. She typically has an insane amount of image and textual data baked in which allows her to generate a wide range of image styles.

But best of all, because she's pre-trained, you don't have to pay Charles prices to be able to get these results.

 

Then in 2022, everything changed again, this time not with images but with text generation.

Archetype 2: Barry, the Large Language Model

OpenAI once again beat the market to release ChatGPT in November of 2022. Within the next year a host of competitors also released similar large language models.

We'll call this Large Language Model (LLM) archetype Barry.

 

For most people, Barry is what they think of when they think of AI. And for good reason. He brought AI to the masses.

Barry is one of those people who has an opinion and a fact about everything. Unfortunately he doesn't know when he's wrong, you so have to watch out for that. But in most cases, he's great at gleaning the meaning of a document, making small logical deductions, and even conducting simple math. He is also usually great at translating between language and checking your grammar.

Just like Mae, he's trained on the entire Internet in most cases. This means that you can plug him into many different use cases in your company and get useful logic deductions and textual summarizations at a very low cost.

Who Else?

You might be thinking with all this buzz about AI, there must be another 15 different archetypes we have to learn. Fortunately, the answer is NO. If you're a smaller company, you should start focusing on the first two above: Mae (image generation) and Barry (large language models). If you're a bigger company, you may have some places where adding a customized Charles (deep learning) makes sense.

Of course there are a few more types of AI models out there, but to be honest, in most business use-cases you won't need them.