7. Examples and Applications
Finally we ask whether the machinery still feels natural after leaving the two-layer MLP. A ResNet introduces spatial layouts and skip connections. GPT turns text into token streams and causal masks. A Fourier neural operator connects a learned map to PDE data. Diffusion and reinforcement learning force us to keep randomness and state in view. The examples are organized as runs a reader can reproduce: what to execute, what artifact appears, what to inspect next, and which claim that artifact supports and where that support stops.