To give you an example of what I am talking about:
1. Figure out if a binary tree is a mirror image of another binary tree.
2. Describe X where X is a language specific feature. More amusingly, in machine learning interviews, pick one out of millions of algorithms that are out there that the interviewer knows very well, demand that the candidate answer and derive every single part of that algorithm.
There is a distinct difference between demanding that people rely on memory to solve a problem vs problem solving abilities to solve a problem. While both are certainly useful for a job, I am honestly not sure how exclusively devoting yourself to the former is better than the latter.
One question in an interview was something like: How many possible IP addresses are there per square meter of the Earth?