[remoteorbust]

That's an excellent idea. I'm in no way saying the existing method is perfect. Just that some of the things it tests around communication and being put on the spot and analyzing a problem in a way that is understandable to the rest of the room is actually a really good engineering skill. There can be other great ways to measure those skills.

I know my opinion is unpopular, but I sometimes do think I've figured out something other people miss. I think when it comes to whiteboard inteviews, candidates are often playing the wrong game. They think it's about gotcha questions and they think they fail because they didn't leetcode hard enough. I don't think that's true, they are just trying to game the thing that's easy to measure.

In my experience with the "terrible" FANG companies, it's not about gotcha questions. I get the offers even though I don't often find a non-naive solution. The people I'm in the room with really do want to see my thought process and they really do want me to communicate the tradeoffs with them. People don't fail the gotcha questions because they don't know trivia or because they forgot a detail from their CS classes. They fail the whiteboard interview because they see an unfamiliar question and say: "I don't know that trivia" instead of drawing out possible solutions and having a conversation with their interviewers.

[WDCDev]

They think it's about gotcha questions and they think they fail because they didn't leetcode hard enough.

But ... if you read some of the feedback from interviewers at "those companies" that rely on these interviews, they say the reason they failed someone is exactly because they "didn't leetcode hard enough". It is manifested as:

"Well other candidates got the same solution as you, they just did it 10 minutes faster"

or

"You missed an edge case, even though your core algorithm was correct".

This is a huge issue with these interviews. It's all too easy for interviewers to evaluate candidates based on how fast, correct, neat or "complete" you answer was. It's easy and takes no time.

[redisman]

Agreed, thats a disingenuous point. "It's not about getting the right answer but the way you think". I've never found that to be true. If you don't get to the right answer, you're gone. If they planned to ask two questions and you only got through one, you're gone no matter how you "thought" about it. A huge part of this is Leetcode practice. If you can't solve most algorithm questions on whiteboard in less than an hour (because you haven't practiced) then you won't pass any interviews.

[mrep]

> "You missed an edge case, even though your core algorithm was correct".

That may have been a reason why I have failed some interviews but I feel like most interviewers are actually good about this and will say something along the lines of "what about this input?" where my code does not work and then I have that "oh shit, that won't work for that" and then I fix it.

[KhanMahGretsch]

Thanks for elaborating so well on your initial premise, I think you bring up some good points about how this kind of interactive problem-solving can be an effective tool available to the interviewer. For better and/or worse, there's a reason why it's so prevalent now and controversial.

I'm inclined to describe it as a sort of interrogation technique, you're offering a stimulus (the problem) and aggregating a number of reactions to form an opinion, and open up new lines of questioning. Very delicate work.

I think problems arise when unskilled interviewers present an excessively obtuse problem to the candidate, then read far too much into their responses (this approach is also easily "hacked" by those who've memorised common problems of this ilk). To stick with my interrogation analogy, it's the equivalent of screaming in the suspect's face, noticing that they gulp before shifting in their seat and scratching their face, then deciding that "they must be lying".