Raw LLM Responses

As it stands right now, it's a cool party trick, but AI's ability to engineer completely breaks down as soon as you try to get it to do anything even remotely complex. While minesweeper seems complex to a lay-person, it's one of the simplest, most well-documented games of all time. Some people say "it will only get better, and then it will replace software engineers," except you still have to be able to communicate to the AI what you're trying to do. Software engineers themselves have to go to school for YEARS to be able to even communicate to OTHER software engineers what they're trying to do, and entire languages have been developed specifically for that purpose (e.g. UML), and it still isn't easy even with that. Before AI can do real engineering projects, what will have to happen is effectively a new kind of declarative framework that AI can use to deterministically output code, essentially making it a new ultra-high-level programming language that the AI "compiles" down to high-level code, which in turn compiles down to machine code. But this will still require software engineers to make anything that is actually good, because they have to know how to communicate to the AI with the proper language. Otherwise, it's like a baby trying to tell Shakespeare about a story they want him to write. Shakespeare might indeed come up with something, but it's not going to be what the baby is imagining, because the baby doesn't have the required vocabulary to express the intent.