Three words show up everywhere in AI — vector, matrix, tensor — and they sound far more intimidating than they are. They are the same idea with one more axis each time. Here is the whole thing in one line, and then we'll walk it slowly:

A vector is a row. A matrix is a waffle. A tensor is a syrup-drenched stack of waffles on a plate.

That's it. Keep the breakfast in mind and the rest is just detail.

Vector — one axis, a row of numbers

A vector is a single row of numbers laid out along one axis. Nothing more. Something like [0.12, -0.03, 0.88, 0.41].

In AI, the canonical example is a token embedding: the model turns a chunk of text — a word, part of a word — into one long row of numbers. A real embedding is much longer than four slots (12,288 in some frontier models), but the shape is the same: one axis, and each position holds a single number, a coordinate. The vector is the word, as far as the model is concerned.

Matrix — two axes, a grid of numbers

Stack many vectors on top of each other and you get a matrix: a grid with two axes, rows and columns. Each row is a vector; each column is one feature shared across all the rows.

This is exactly the shape of an embedding matrix. Every token in the model's vocabulary gets its own row. Line them up and you have a table — "cat" is a row, "market" is a row, "bitcoin" is a row — where each row is that token's vector. That is why a matrix feels like a waffle: a flat grid of pockets, each pocket holding a number.

Tensor — three or more axes, a stack of matrices

Now add depth. A tensor is a stack of matrices — a third axis piled on top of rows and columns. Picture a stack of waffles on a plate. One waffle is a matrix, its pockets are the numbers, and piling waffle on waffle adds the extra axis.

And the syrup? The syrup poured over the whole stack is the operation that ties the layers together. In a transformer, that's attention — information flowing across every layer at once, mixing the numbers between the waffles. That stacked, syrup-linked shape is exactly what a modern model actually computes on. Not a single row, not a single grid, but a stack, glued together.

The one-line recap

• Vector — a row of numbers. One axis.
• Matrix — a waffle: rows of vectors. Two axes.
• Tensor — a syrup-drenched stack of waffles. Three or more axes, tied together by an operation like attention.

Same idea, one more axis each time. Once you can hold the waffle stack in your head, most of the "shape" talk in machine learning — batches, layers, attention heads — is just more axes on the same plate.