“Get LLM to generate 80% of code, tweak 20%” just doesn’t make sense. It takes more time to understand others code than to write it yourself for anything comple

“Get LLM to generate 80% of code, tweak 20%” just doesn’t make sense. It takes more time to understand others code than to write it yourself for anything complex. LLM might be good for boilerplate, or for brainstorming architecture, but I still don’t see any indication of it handling novel use cases. The author gives plenty of examples of tech that was underestimated. They all solved problems in such novel and concrete ways that most couldn’t see the application. Compare to something like crypto which are hyped on the potential to solve problems at some point in the future in ways that we’ll figure out. I actually think LLM has the potential to deliver on the hype, but it’s a safe bet to be bearish on hyped tech until actual concrete uses are shown, and it’s usually the boring stuff that ends up changing the world. LLMs are by definition derivative and lack underlying knowledge. Until those issues are solved I can’t see them generating anything minorly complex or novel with minimal, easily discoverable bugs