Because LLMs are based on the abstract ideas of neural nets from brains. Say what you wish, but some problems were completely unsolvable before we adopted this paradigm. On some level, we must've gotten some ideas close to the right ballpark.