You wouldn't use an LLM to solve a big Linear Programming problem, because it would cost way more than using the Simplex Method, and you'd be worried that it might be wrong.