Here is the twist the 'dendrites lost' story misses: the frontier has been quietly rebuilding them. A point neuron sums; it cannot multiply its inputs. But multiplication — one signal gating another — is the signature dendritic operation, and it is everywhere in modern models. Attention is input-dependent routing. Mixture-of-experts fires different sub-networks for different inputs, which is exactly context-gated dendrites. Gated linear units multiply a value by a learned gate. None of these were derived from neuroscience; they were found empirically because they work, and they are the dendritic instinct re-expressed in dense, GPU-friendly form. Here you build the smallest one — a gated unit — and give it a task that needs a product, sign of x-one times x-two, which a dense linear layer provably cannot represent. The gate solves it. The two camps were never rivals; they are converging.