[bnejad]

Despite the sentiments of "they just want to see how you think" or "you don't have to get it right", in my experience if you don't do fairly well on the whiteboard you've failed the interview.

[curtis]

I think this partly depends on how hard the coding question is. If they give you FizzBuzz, I think you're pretty much required to ace it, even if you've never seen it. FizzBuzz is really simple though.

At Facebook I was getting I guess what you'd call classic algorithm questions but with an extra twist thrown in. If you combine that twist with excessive time restrictions (the Facebook interviews were strictly 45 minutes), you not only have to code well, you have to do it fast. On an algorithm that they appear to be hoping you've never seen before.

[soham]

Yeah, that hope from them is baseless. Truth is, you're competing with candidates who have practiced a lot. They know it, and candidates know it too.

I'll re-iterate: Technical interviews are a competition. You cannot go unprepared. Testing your "raw" skill is stupid and a myth. It's as stupid as saying you want to see Usain Bolt (and everyone else) compete without practice. Those days are gone.

[dmansen]

"Doing fairly well" to me is: you asked the right questions and converged on the solution, even if it took some time. Doing poorly means: you made stuff up instead of asking questions, you invented a convoluted way of solving the problem when some other data structure would have basically solved it for you, etc. It's also painful when the candidate's mental model of computation is miles away from reality.

Obviously, getting it right the first time would be great. But I'll settle for something not obviously insane.