LinkedIn, poke at my logic here.

What I hear from Aline Lerner and Gayle McDowell is that the biggest problem with DS&A interviews is disengaged and/or bad interviewers.

To quote Aline: "I personally believe that the biggest problem with interviews is not the format and the questions but bad, disengaged interviewers. We often conflate all the things we hate about LeetCode interviews with the interviewers who administer them. Imagine if your interviewer didn’t expect you to regurgitate the perfect answer… imagine if the same algorithmic problem were a springboard to see if you could write some good code together and have fun talking about increasing layers of complexity when you shipped that code into the real world."

I have a different take. I agree that open-ended interviews are generally better for getting more signal about the candidate than, e.g., simply asking "did they get the optimal answer or not" (which only contains 1 bit of information).

But I think that there is an inherent trade-off between how open-ended the interview is and how objective/unbiased the evaluation is. For an impartial evaluation, the format must be strict. However, the stricter the format is, the more "gameable" it becomes (through targeted preparation, memorization, or cheating).

So, there is a spectrum ranging from open-ended (higher signal, more biased) to formulaic (lower signal, less biased).

If we accept the premise that this trade-off is intrinsic, then all that remains is to ask where a company (say, Google) should lie in the spectrum.

I'd say that the larger the scale, the more important objectivity becomes. That's why college admissions use closed-answer tests like the SAT (which begets even more practice than LeetCode interviews), while recruiting a cofounder for your startup can happen over a coffee chat.

As the name suggests, the scale for Big Tech is big, so they must prioritize objectivity.

My conclusion is that better interviewers and more open-ended interviews is not necessarily an improvement. It's a trade-off, and one that probably doesn't make sense at the scale of, say, Google.