LLMs fundamentally can't do math. What OpenAI has started to do is have the LLM write
and execute code to do math for it. Which is clever but also hilarious. If you just think of LLMs as incredibly expansive auto complete and nothing me, then you can figure out where to use it.