When LLMs become better than humans at the following:

1. Knowing what you don't know

2. Knowing who is likely to know

3. Asking the question in a way that the other human, with their limited attention and context-window, is able to give a meaningful answer