What a Dot Matrix Printer Can Teach You About AI

Author

Robert Hangu
October 21, 2024

Remember those old dot matrix printers?

The ones that spit out long, continuous paper, with the really annoying noise?

Well, believe it or not, that clunky old printer can actually help you understand how generative AI works.

Wait, what? A printer? AI?

Stay with me here. It’ll make sense in a second.

Here’s the deal: When an AI generates text, it doesn’t magically “know” what the next word is going to be.

Instead, it’s like it’s reading off a long piece of paper—just like that dot matrix printer.

This “paper” is what we call context.

It’s all the information the AI has read so far, including the words you typed and the text it’s already generated. It’s your entire current chat history!

And just like that printer paper, the AI looks at what’s already there on the paper… and chooses the most likely next word.

Every word it generates is based on probabilities.

At any given moment, the system has a few different word choices it could go with.

Some have a high chance of being the “right” word…

Others? Not so much.

So how does it choose?

It takes a look at the entire “context”—all the text up to that point—and calculates the next word based on the highest probability.

Let’s have a look at this short context consisting of a single prompt:

From the five options in the list it chooses the word “redefine”. This has the highest probability based on the prompt at hand.

Overall it’s a continuous process that repeats over and over until the entire answer is produced.

You probably found yourself waiting for the completions to finish. This happened often in the earlier days of ChatGPT. Now you know why.

But here’s where it gets interesting...

Context isn’t just what you type.

There are all kinds of things that can add to this “paper.”

Not just your words, but also:

  • Uploaded documents

  • Data connections

  • APIs

  • Even other AI models!

Yup, you can feed your AI all sorts of information, and it’ll consider that as part of the context.

It’s just like people and machines writing to the paper coming out of the printer.

In fact, you can automate entire workflows with AI by feeding it the right data from multiple sources.

Let’s say you were generating ads for architecture offices. You could pull in customer reviews from a database.

Or connect the AI to your CRM, and have AI draft personalized outreach emails automatically based on actual client data.

Here's what I mean:

The context (in the middle) begins with your prompt written by hand. Then the actual customer reviews are copy-pasted into the chat by outside systems.

Well… they’re not copy-pasted like done by a human – it all happens behind the scenes. But ultimately they land in the context.

The more context you provide, the better the output.

Here’s another fun fact…

You might’ve heard that ChatGPT has a “context window” of around 128,000 tokens—which is about 90,000 words.

That’s a LOT of space for context!

So if you’re thinking, “Oh, I need to keep my prompts short…” Nope.

You can actually feed it a ton of information and the AI will keep it all in mind as it generates.

And that’s where things get really powerful.

When you combine human inputs, machine data, uploaded documents, and APIs, you can guide the AI to generate almost anything you want. 

AND you don’t have to waste time copy-pasting stuff between your CSVs and ChatGPT.

So, next time you’re typing out a prompt, think of that dot matrix printer…

Imagine your context rolling out, line by line.

And remember:

The more context you give, the better the AI’s output will be.

If you want to know more ways of how to fill up the context, I’ve got a full course that dives into this, along with strategies for integrating AI into your business workflows.

Cheers,
Rob