A Layer of Neurons
One neuron produces one number. To get anywhere interesting, you run several neurons in parallel, all reading the exact same inputs but each with its own weights and bias — this is a layer. Instead of one output number, a layer of 4 neurons produces a 4-number output vector.
This should feel familiar: it's the same "several independent computations over the same input, run side by side" idea as multi-head attention. There, each head learned to notice something different about a sentence. Here, each neuron learns to notice something different about the input vector — one might respond mostly to x1, another mostly to x2, another to some blend of both.
Four neurons, same inputs, different weights — a layer
Each neuron reads the same (x1, x2) but has its own weights and bias, so they each respond differently — together they produce a small output vector.
Move the input sliders and watch all four neurons react differently to the exact same change — that's the whole point of a layer. None of them see anything the others don't; they just weight it differently.
A layer is just a list of neurons, each computed the same way:
A layer turns one vector into another vector. Chain several layers together — feeding one layer's output into the next layer's input — and you get real depth.
What distinguishes the neurons within a single layer from each other?