Because people never tried to effectively utilize the tool. If you spend an hour learning how to effectively prompt your chances of getting higher quality gens go up significantly. It still happens obviously but I can confidently say ~80% of the outputs I get are servicable enough for my usecases. The remaining 20% require some manual tweaks and reprompting but I can usually get it there. It works best for things where you know how to ask an effective question already I find. Like just asking a basic question without giving enough information will give you lower quality responses than a more detailed prompt. Its kind of the opposite of what we’ve learned to do with search engines. More information is actually better than less in this case.