Raw LLM Responses

As someone who recently started messing around with LLM APIs in my projects and had my mind blown, the biggest thing about AI for code in my opinion is not its ability to create/fix code, but its ability to easily inject fuzzy logic into your code to make it do things that you otherwise wouldn't even be able to do. I think AI is both overhyped AND underhyped, specifically the direct human interaction is overhyped, but the potential for building innovative solutions around it is underhyped. You can have an LLM decide which functions to call based on some abstract input/data, you can have it generate data on the fly based on some prompts, it can rate/rank objects based on some abstract heuristic that a human could easily do manually, but standard code would not be suited to doing. As far as its ability to code, while the actual models by themselves are still not good enough to code fully fledged products on their own, it's just a matter of building the right scaffolding/structure/framework around the LLMs, adding different layers together, and you could dramatically increase their effectiveness. And obviously, big tech companies around the world are trying to do that, hence the hype. What you can do through normal human interactions with an LLM are not as impressive as the AI hype train would warrant, because that's not really what the hype is about.