It's very unclear to me why AI companies are so focused on using LLMs for things they struggle with rather than what they're actually good at; are they really just all Singularitarians?