A Single Neuron

Remember that mystery box with knobs we kept talking about? The thing that somehow learns? Well, today we're cracking it open. Watch — light's already leaking out.

And here it is. This little doodle? It's hiding inside every AI you've ever used. Chatbots, photo filters, all of it. We call it a neuron — and honestly, it's simpler than your homework.

Look on the left — those circles glowing pink? Those are the inputs. They're just facts about whatever we're checking. One for whiskers, one for a tail, one for barking. Yes is a 1, no is a 0.

Now see those lines connecting the inputs? Each one carries a weight — that little number on the line. Weights say how much each clue matters. Green means it helps. Red means it argues against. [pause:400] Everything the AI ever learns? It's hiding in these numbers.

Then everything meets at this little circle in the middle. That funny squiggle is the Greek letter sigma — it just means add it all up. Whiskers times its weight, plus tail times its weight, and so on. One single number pops out. See? Three.

But that big number isn't very useful yet. So we squish it. Watch this curve — feed in any number, big or tiny, and it gets squashed somewhere between zero and one. Think of it like a dial from "nope" to "definitely yes".

And right at the end — see that green bar filling up? That's the answer. It's not just "yes" or "no". It's how SURE the neuron is. Point eight seven means it's pretty confident. Point oh three means it really doesn't think so.

Okay, big reveal time. Remember those mystery knobs we kept hearing about? The ones training somehow adjusts? [pause:400] They're the weights. That's it. When people say an AI is "learning", they mean a computer is slowly turning these dials.

Your turn now. Grab those sliders and play with them. Push barks into the green and see what happens — the neuron suddenly thinks barking dogs are cats! That's exactly how a badly trained AI behaves. Have fun breaking it.

Now here's the thing about one neuron. It's kind of a one-trick pony. It only answers ONE question — "cat or not?" — with a number between zero and one. That's the whole gig. Pretty humble, right?

This time the weights are locked — pretend the AI already finished its training. You're the one describing the animal. Tap the buttons. Whiskers? Tail? Barks? See what the neuron decides. Try to fool it!

Let's lock it in. Inputs come in. Each one gets multiplied by its weight. We add them all up. We squish the total. Out pops a confidence. Five steps, every single time, in every neuron ever built. You've got it.

One neuron is tiny. Boring even. Wire up hundreds — millions — and they start recognizing faces, writing stories, driving cars. That's a neural network. And that's exactly where we're going next.