Am I the only one that sees the parallels between what these language models do when it pertains to code to what some humans savants do, in relation to many "exceptional" skills some of them seem to posses? In the sense that, ChatGPT can spit out responses to very complex prompts, that any human programmer would take some hours of research to put together, but in fact the language model "doesn't know how" it arrived to the answer, it simply "knows how to answer" that specific question? To me it's a very similar phenomenon looking from the outside. Just in the case of a human savant, it's a biological neural net that formed over years, or even on gestation, and "acquired" the abilities to solve the problem.