What I realized, after thinking about the statement “AI doesn’t care where it is trained,” is that modern generative AI is essentially a kind of magic box to store up know-how. It can take a tremendous amount of energy to get know-how into that box. But once that has been done, new applications of that know-how can emerge from that box an infinite number of times — all at low cost.
Another time that this kind of thing happened was the creation between the late 1970s and the mid-1980s of desktop publishing. Before then, if you wanted to produce a document with high quality type-setting, you needed to hire a professional to do it for you.
But starting with TeX in 1978, and software that followed, the knowledge of experts could be encapsulated in software that you could run at home on your personal computer. Essentially, that was the beginning of “expertise in a box.”
We are now entering an era in which all sorts of expertise can be stuffed into that box, given enough prior examples, after which it can be cheaply accessed anywhere and everywhere.
More tomorrow.