The narrowing was a choice

Something happened partway through this journey that was never announced, and it should be, before the ending.

Up to the going-deeper module, the story was about vision. Networks that sorted photographs. A million labelled images. A system that learned dozens of Atari games from pixels. Then it turned toward language and never turned back. Words, tokens, embeddings, attention, transformers, next-token prediction, context windows, agents.

That was deliberate. Following one thread all the way down beats sampling ten threads shallowly, and the language thread happened to arrive somewhere extraordinary.

But it leaves a false impression, and it is worth correcting before the end. Someone could finish this journey believing that AI means chatbots.

It does not. The machine you now understand in detail is one branch of a much larger family, and several of its relatives work on principles genuinely different from anything in the last four modules. Not simpler. Different.

This chapter is a short tour of what was left out. It will not teach any of them to the depth of the transformer, and it is not trying to. The job is to give you the shape of the family, and to hand you one contrast sharp enough to make what you already know clearer by comparison.

Start with the one you have almost certainly used, which produces pictures, and which works nothing like the machine you have spent four modules building.