In a further future that is inevitable, but in this case I'm referring to an intermediate state where it's probably LLM + one more thing. Some of the inevitabi

In a further future that is inevitable, but in this case I'm referring to an intermediate state where it's probably LLM + one more thing. Some of the inevitability is optimization. Computers intrinsically admit of certain optimizations that biology doesn't. There's an effect in technology you can notice if you look where a lot of times the hardest part is getting the thing working at all. Once you have a working thing in hand, optimizing it becomes possible. It's hard to optimize something you don't have in hand. Steam engines show this, and I expect that fusion will follow this trajectory too. In this case, for instance, I expect once we have high-functioning computer vision we can start optimizing individual components of it. Eventually it may not even resemble the current architecture, but it's hard to get from here to there without the intermediate phases. The cerebellum provides an interesting case study in biology, more clearly than the language center of the brain. It's still made out of neurons, but it is structured in a highly stereotypical way for the task it is doing, and does not resemble the "everything connected to everything" architecture used by our artificial neural nets and a lot of the rest of the brain. It is very, very optimized for its particular task. In some ways a lot of it looks more like a specialized and exceeding parallel DSP (albeit, of course, analog) rather than a neural net architecture.