Formally, what "humans" are doing is learning the probability distribution of data, via evolution, cultural mimesis, and then, as a cherry on top, from direct experience. Don't confuse "human" intelligence for the intelligence you get from birth onward. Trillions of tokens is paltry compared to the data amassed by biology through four billion years of evolution. As we grow and grow the mountains of training data, it's increasingly obvious that knowing how intelligence is made (i.e., more and more precise distributions over data) isn't going to render machine intelligence less than astonishing.
Formally, what "humans" are doing is learning the probability distribution of data, via evolution, cultural mimesis, and then, as a cherry on top, from direct experience. Don't confuse "human" intelligence for the intelligence you get from birth onward. Trillions of tokens is paltry compared to the data amassed by biology through four billion years of evolution. As we grow and grow the mountains of training data, it's increasingly obvious that knowing how intelligence is made (i.e., more and more precise distributions over data) isn't going to render machine intelligence less than astonishing.