>>1785I wonder if it will catch on, writing specifications in natural language and then debugging the randomly generated code sounds a lot more effort than just writing the code yourself. At least when you are debugging code written by another human you can assume that they had a reason to write the code the way it is, but figuring out what the generated code actually does, is that correct, and if not, how to make it correct, sounds a lot harder.
It makes me wonder what kind of work the people excited about this do, where this is the actual bottleneck. I know that even if the AI could write the same code as I do, it wouldn't drastically change my job, because that's not the actually hard part of software engineering in my field.