Github’s dev assistant thing tries to fill in the code based on what you’ve written, by regurgitating prior knowledge almost blindly — but as a human I can process that regurgitation, and maybe extract a relevant subset of it for my actual use.

I’ve recently realized I do the exact same thing with speech — I’ll talk without thinking, forming sentences out of heuristic systems, and process them after the fact. Then tune the sentence slightly, and continue.

In this fashion, I produce sentences without really knowing what I’ve said, until I think about it later. In much the same way that I drive long distances on auto-pilot, and can’t even remember anything of the route by the end of it.

And those sentences often have ideas that I’ve never actually put together intentionally, but I completely agree with, and upon review, are correct to the best of my knowledge.

I don’t know what speech is, but it’s definitely more than vocalization of my thoughts.