A local minimum is a place where a system gets stuck with an answer that works. Small adjustments stop making it better. A better answer exists, but finding it means leaving the approach that got you this far.
I think LLMs are already better than the average person at most of the tasks we ask them to do. They have obvious weak points, and areas where they are far better. Their range is the extraordinary part. A single person can draw on enough competence to attempt work that would once have demanded several different specialists.
But above average starts to become its own average. When we all draw from roughly the same range of answers, our work can converge on the same competent result. A model keeps refining what it knows how to produce. The answer gets polished. The approach doesn't change.
An excellent human can interrupt that pattern. Someone who knows one part of the work deeply can spot a possibility the model keeps missing. They raise the standard for that part. The model helps them carry it through the rest of the project.
That combination matters because having one strong skill has never guaranteed you could finish anything. A project can fail on a task you barely care about. You have the idea and the judgment to develop it, but some supporting job exceeds your ability, time, or budget. LLMs can cover more of those gaps. They still make mistakes. You still have to judge the work. But fewer of your weaknesses have to become the limit of what you can accomplish.
This is why I'm optimistic about specialization. An otherwise average person can spend more time getting good at the part of a mission they care about. They may only need to become slightly better than the model at that one part to contribute something it would not have produced alone.
The size of that advantage matters less than what it changes. A small improvement at a decisive point can make the whole result more useful. And the model can help turn that improvement into something finished, something other people can actually use.
You could be ordinary at most of what a project requires and still be the person who makes it worth doing. If your judgment improves the part that matters, the whole project can be better because you were there.