With more advanced use cases I find the LLMs better at exploring a problem space and possible solutions rather than providing a solution full stop. Being able to refine a problem down into a digestible chunk the AI can take a bite of does require human level understanding that's hard to get if you haven't 'done the work' to have good fundamentals.
The number of times it misled me makes me reluctant to rely on it again. I was wasting time chasing non existent paths.
I think the marketing around chatgpt is more powerful than the tool. It looks impressive to junior devs or people without expertise in a domain, but that’s about it. What bothers me is the amount of spam around this tool - they need to work on improving it and then restart promoting as thus far it’s just annoying noise.
I actually think maybe it is even better with more experience as it allows you to understand what it excels at and be critical of its output to recognize when it might be wrong.
I would usually be able to do whatever it does, it would just take more energy and time, but I am able to almost immediately recognize when it is wrong.
Issue is that my time is more valuable than to be spent improving someone’s product.