The purpose of this article is to aid in understanding generative AI in a way that almost anybody can understand.
Forget the Chatbot: Let’s Look at the Hardware
Ask anybody how much they understand about AI and the answer will probably include the name of the 3 top Chatbots. Push for a deeper explanation and the person answering might refer to a couple of individual parts of a computer – video card, CPU, memory, etc. And while they are starting to move a little closer to a correct answer, they are still nowhere near the ballpark in their attempt to explain how AI actually works.
While it’s understandable a typical layman really doesn’t know, and is unable to explain how AI works, it’s also a shame. Because when it comes right down to it, understanding exactly how an AI chatbot accomplishes the amazing task of communicating with a human, or vibes up written computer code, AI is actually very easy to understand. Not just from a superficial perspective, but right down to its bare bones operation.
Join me into a shallow dive into the world of linear algebra (remember high school Algebra 1?), GPUs, and a few basic computer functions, and you will possibly walk away with a detailed understanding of generative AI, or what most of us know simply as ‘AI’, version 2026.
Where to begin?
When taken as a whole, AI is incredibly complex. If you aren’t already familiar with the subject, you might not even know where to start the learning process. So rather than explaining AI in terms of tokens, embeddings, and parameters, we’re going to visualize AI as a 3D video game.
With that in mind, it is important to understand that yes, AI does run on a computer, but the bulk of the magic of AI happens on the GPU.
Setting up a baseline that most everybody already knows: CPU (Central Processing Unit) is basically the brain of a computer. Among other things it performs calculations, makes logic decisions, assigns memory addresses, and manages the flow of information across all of the computer’s hardware components.
GPU is an acronym for Graphics Processing Unit, something that many recognize simply as ‘video card’. Specifically, it is a type of video card ideal for 3D gaming.
Think of the GPU as a “CPU” except it’s for video generation. It’s the central brain of a video card. A GPU is ideal for 3D gaming because of what it’s capable of doing. While a video card has many different component parts (resistors, diodes, cooling fans, etc.), a GPU is very good at solving a certain type of algebraic linear equation.
Specifically:
A + B = C
And to reiterate, it’s very, very good at it. Far better than a CPU. Why? A standard CPU handles tasks as if being presented with packages arriving on a conveyor belt, solving for C each time a new package arrives. A GPU, on the other hand, is a massive team of thousands of workers, each working out a single instance of A + B = C. Just as the thousands of workers solve for C, all roughly at the same time, a single GPU also solves for C simultaneously (in ‘parallel’). The parallel nature of a GPU allows it to address billions or even trillions of these calculations every single second.
At this point your brain should be screaming BUT THAT ISN’T LINEAR ALGEBRA, THAT’S BASIC MATH!
And you are correct, however, this being a basic explainer article, I don’t want to muddy the waters by trying to describe its older cousin, something known as Fused Multiply-Add (FMA), or:
(A + B) * C
…which is linear algebra. But I don’t really need to get into FMA at this point because both A + B = C and (A * B) + C are treated nearly identically when fed through a GPU’s CUDA core.
CUDA core?
CUDA cores are the traditional workhorses of the NVIDIA GPU. They are excellent at handling general, standard math calculations such as A + B = C or a basic (A * B) + C.
The final technical concept you’ll need to understand (but this one is fun)
I am now going to direct you to a video produced by Branch Education. It explains how GPUs work, mostly through pictures and animations. You don’t have to sit through the entire video, although I totally recommend watching all of it because a) it’s a very good video and b) the concepts it presents build up from one to the next. But you’ll absolutely need to walk away with an understanding of the 2 major points covered between 8:48 to 10:37 (CUDA Cores), and from 16:52 to 20:52 (especially SIMD).
You will find the Branch Education video at: https://youtu.be/h9Z4oGN89MU
Are you ready for an aha moment?
Generative AI works exactly like the 3D image render described in the Branch Rducation video, except instead of vertices that appear at different X,Y,Z coordinates, we’re operating on words (or lists of words). That’s it. That’s everything you need to know! That is basically how generative AI works. Of course, there are many steps involved converting words to ‘coordinates’, but for the purpose of this article, there is literally no difference between the 3D image vs. how AI interprets the sentence that you type in to your favorite chatbot, and how the chatbot comes back with a reply. It’s all math and it all traces back to A + B = C.
Words As Spatial Coordinates?
The Branch Education video uses a cowboy hat to explain how a 3D object exists in world space, how sets of 3 vertices comprise a fragment of each 3D object, and how other 3D objects within the same world space can be made to appear in front of the hat, or behind it, or off to its left or right side. In an AI Chatbot, instead of tracking the 3D points of a cowboy hat, the GPU is tracking the 3D coordinates of language.
To draw an analogy, imagine looking up at the night sky. In a dark sky environment, you might see several thousand stars. Now imagine that each of those stars is a single word. Words with similar meanings are clustered close together in the same constellation. “Apple,” “Orange,” and “Banana” sit in a tight little cluster. “Engine,” “Wrench,” and “Piston” sit in a completely different corner of the galaxy.
To find the path between these word-stars, the AI draws invisible lines connecting them based on grammar and context. (And thus just like the cowboy hat in the Branch Education video, we can scale up, rotate, and move words around so that they connect in different ways.) When you connect trillions of these stars/words together with lines of logic, you get a massive, glowing web called a neural network. (We won’t get too far into neural networks in this explainer article).
Wrapping Up
The Branch Education video used a cowboy hat to explain how a 3D object exists in world space, how sets of 3 vertices comprise a fragment of the cowboy hat, and how thousands of individual vertices comprise and entire 3D object. The video also explained how other 3D objects within the same world space can be made to appear in front of [the hat], or behind it, simply by modifying the Z axis (basically by modifying the object’s depth in world space).
By introducing you to the A + B = C mathematics, and the analogous linear equation (A*B) + C, and letting you know things work exactly the same way in an LLM, you now have a basic understanding of the magic that happens under the hood.
(But Shane, what about grammar and context and all the stuff that makes up natural language?!?) Fear not! (or more accurately, be very afraid!) We’ll cover some of those concepts in a future article where we’ll discuss matrix math and the mystical neural network.