You assume the interviewer is ideal: infinitely competent, infinitely great at judging other engineers' skills etc. This is practically never the case.

I am not one for LC type of problems, but at the least I have to admit they have the potential of being more objective than everything else during the interview process. You get the specs, you write the code and it gets tested via multiple test cases. It does not matter if the interviewer agrees your solution will work or not, and the interviewer won't have to copy&paste your code, compile and run it (like someone else mentioned in this thread).

On the other hand, what I particularly dislike about the LC problems is that many of them are essentially trick questions and brain teasers. Can't we just stick to problems that are relevant to an SWE?