There is a massive difference between an LLM and a compiler. Compilers work deterministically, an LLM does not. You cannot ever trust an LLM to actually produce

There is a massive difference between an LLM and a compiler. Compilers work deterministically, an LLM does not. You cannot ever trust an LLM to actually produce working code and must review it all yourself, as all models (yes, even $latest_model) will happy insert hallucinations into random places in the code. But once you take the time to do that, you have not saved any time and could've just written it yourself. That is why LLMs are not actually useful for programming, contrary to the hype, unless you want to just YOLO stuff out there without caring if it works.