The first rung of Module 3 · the machine that bends space
[M] · v
Forget definitions. A matrix is a machine. Feed it an arrow, it hands back a new one — rotated, stretched, flipped, or sheared. Just four numbers, total control over space. And quietly, this is exactly what an AI’s “weights” are — you’re about to meet them face to face.
Step 1: you’ve already met this machine
You’ve used it a thousand times without knowing its name. Rotate a photo 90°? Matrix. Warp a selfie filter, stretch a game character, tilt a scanned page square? Matrix, matrix, matrix. Your phone is quietly running the same multiply-and-add trick you’re about to learn — millions of times a second, on millions of pixels. Four numbers, total control over space — that’s the whole secret you’re about to hold.
Step 2: a grid of numbers, but make it a machine
You’ve spent this whole module drawing arrows. Time to meet the thing that moves them. Stack four numbers in a little square — [[1,0],[0,1]] — and you’ve built a machine. Feed it an arrow, it hands one back. This particular machine happens to hand back the exact same arrow — the “do nothing” machine. Every other machine in this rung is just those four numbers, changed.
Step 3: how the machine actually chews on an arrow
Here’s the whole trick, no magic. To feed the vector (1, 1) into the machine [[2,0],[0,1]], multiply straight across each row and add:
Out comes (2, 1). That’s it — every matrix, every time. Multiply-and-add, once per row.
Step 4: press the buttons, watch space bend
Below is a real machine. Same four-number grid, five different settings. Press one and watch what happens to a plain little square — and to the arrow riding at its far corner, (1,1).
The plain do-nothing machine. Press a button below.
Step 5: the part that de-fangs the AI
Here’s the reveal. Every time someone says an AI has “billions of weights,” they mean: billions of numbers, arranged in giant versions of that little grid. Your sentence goes in as a list of numbers — a vector. The matrix multiplies it, and a new vector comes out: rotated, stretched, squished through so much space that meaning falls out the other side. Same multiply-and-add you just did by hand. Just enormous.
Step 6: one machine is nice. What about two in a row?
Right now your machine does one move — rotate, or stretch, or shear. Real AI models chain dozens of these machines back to back, each one bending space a little more than the last. Next rung, you’ll run a shape through two machines in sequence — and discover that two matrices in a row is, itself, just one bigger matrix.
Prove it · get one wrong and I’ll walk you through it
Two quick ones and the rung is yours.
Press “Shear” on the machine: the top of the square slides sideways while the bottom stays put. What did the matrix just do to space?
The machine is [[2,0],[0,1]]. You feed in the vector (1, 1). What comes out?
Answer both to unlock the next rung ↓
The Matrix — mastered.
You didn’t memorize a formula — you watched a grid of numbers grab a shape and bend it: rotate, stretch, flip, shear, on command. Anywhere this shows up again, you skip it. Next you find out what happens when you run a shape through two of these machines back to back.
Next rung → Matrix × Matrix