I didn't assume either that the LLM is to guess the weather. I said that using LLM for parsing the Met Office's data is maybe not such a good idea if you can do it deterministically.
The idea was "forward $@ to a LLM to make a structured request", not to parse a structured response.