Yeah. That makes a lot of sense. The general inaccuracy/untrustworthiness of LLMs makes me very uncomfortable in their use for data processing and transformations. I’d rather take a while to get it right than to potentially hand off a CSV with glaring problems due to use of an LLM.