I know LLMs are pretty wild technology, but you're wrong about this. LLM's are just models of language. To some extent they can do some of the things you're talking about, but almost by the very nature of the problem other models could do the same better and with much less by way of resources.
