> How can we possibly make use of these breakthroughs if we don’t understand them?

I mean, arguably the development of LLMs matches that description.