The interviews aren't designed to be representative of the work that you do, its more about whether or not you are willing to put in the time to study and get

The interviews aren't designed to be representative of the work that you do, its more about whether or not you are willing to put in the time to study and get to know some topic to be effective at it, and integrate that into your general knowledge. This is why Amazon doesn't base the interviews for specific teams in terms of technical knowledge, because the understanding is that some Amazon SDE working in web service should be able to take a position doing ML and be an effective contributor in a reasonable amount of time. Furthermore, the ranking isn't on whether or not you even figure out the problem. One of the things that gets looked at is coachability - if you are given a hint about the solution, can you immediately recognize it and run with it and implement code fast? Many people fail at this step - they can't put thoughts into code even when told almost explicitly on how they should approach the problem.