I said nothing of LLMs. I don't compare the process of LLM training to any form of human education because they are fundamentally dissimilar. LLMs are engineered mathematics and cannot self-train yet.
There are of course parallels. Anyone who's done ML should be able to see it at least. But yes, there are fundamental differences in methodology due to the mode and level of access: with a human, the only way is to go through the senses and there's no - easy - way to affect the learning substrate; with a LLM, data is directly injected into the substrate, which is also human-architected.