Comment by khuey

One obvious reason to have an LLM output a high level language even if you are never going to read the code is because it will require fewer output tokens and therefore be cheaper.