AI is math you already know

You have probably done this math before, in college algebra or a linear algebra class, without anyone calling it artificial intelligence. The vocabulary sounds intimidating, but the building blocks are ordinary. Let's walk up the ladder.

Stack a few matrices and you have something you see every day. A color photograph is three matrices, one each for red, green, and blue, holding the intensity of that color at every pixel, from 0 to 255. Hover or tab through the image below. Each pixel is one number from each of the three grids.

Words work the same way. A model turns each word into a vector, a list of numbers that places its meaning as a point in space. Words with similar meanings land in similar directions, and the angle between them measures how related they are. Click a word to see what sits near it.

None of this is new. What changed is the scale. Asking how every word in a sentence relates to every other word is one matrix. Then you stack many of those, and many of those, until a single layer holds more than a hundred million numbers, in a model with dozens of layers.

Knowing the shape of the thing also shows you where it breaks. A model learns from what it is trained on, and that is mostly the ordinary, everyday web, with very little of the specialized material at the edges of any field. So it can sound equally fluent on everything while being least reliable right where your expertise lives. More on that in "fluent everywhere, wrong in your field".

The operation is matrix multiplication. You have done it before. AI is organized numbers at industrial scale, not magic. The moment you recognize the math you already know, the rest of it starts to make sense.