if you ever did an interview a at one of (Facebook,Apple,Google, Microsoft..) you know this is not true.

They definitely are asking puzzle question that an university teacher teaching algorithm would fail. And the only way to is practice stupid question on leetcode

He's partially right. There's a lot of pattern matching involved. Like he describes, there are several broad patterns. But then there are sub-patterns upon sub-patterns. You need to be able to recognize these when the interviewer throws a problem at you.

You grind leetcode problems over and over so that you can better recognize these patterns and know when to apply their associated solutions.

The most idealistic scenario is you get a problem you've recently practiced verbatim. Then you just put on an act like you've never seen it before in your life.

But if you get a problem that you can match to a pattern you've practiced, then you're still mostly in a good place.

I work at FAANG - my take is, all questions can be solved based on these 10-15 data structures / algorithms. They are just gift wrapped with a lot of fluff, but underlying each of them is one or a combination of one of these data structures / algos.

As an aside, to everyone who has asked me how to get into FAANG - I always say you should be able to code DFS/BFS in your sleep like your life depends on it. 1/5 questions is surely going to be a tree/graph search.