> But if you spot something's off, you need to adjust it. Go directly for the kill, and make that surgical series of edits.

LLMs are still a really immature technology. The hype is about where it could go in future, not necessarily where it is now.

Think about when compilers were immature technology, and the science of parsing/etc and optimizations were not well understood. You could make the exact same argument you have made now about the need of editing assembly or machine code by hand when the compiler doesn't get it right.

It was indeed common practice to do this well into the 1980s. That, and inline assembly is increasingly unnecessary now.