Hacker News new | ask | show | jobs
by laboo 2105 days ago
Create an exam, broadcast an invite to take it, hire randomly from all who pass.

All excuses for why this won't work can be answered with: then make a better test.

As a species, across all cultures, we tell ourselves that the human we're after -- the student, the employee, the promotee -- can't possibly be selected by test alone.

When in fact, if we can't write a test to select the qualified humans, then either we're too lazy to write one, or, more likely, we actually want to leave plenty of room for human bias to do the actual selecting.

And this is ok because we have a special power: we can judge the value of every human, and its future likelihood for success, with a single conversation. If we weren't in a tech company, we could make a very good living reading palms.

We never hire people who don't pass our wonderfully fuzzy exams, so we have no evidence that we're selecting the best people or not. No worries, though, our palm reading is very, very accurate.

The way we look at it is like this. We make a test no one can pass. We always have one more question or one more "level" that there's not enough time for. Because when all candidates fail, we have to fall back on our palm reading, which is just how we like it.

Power and privilege are precious resources to us. We give them out to those most likely to reciprocate. That's what we're poking for with our "culture fit".

In the future, students of our culture will look back on our hiring practices and say, "It was illegal to hire based on race, age, sex, and a million other things, but not beauty??? They didn't start with beauty? And they never realized that beauty needed to be in the mix? I don't understand."

But we understand. It makes perfect sense.

3 comments

That's the whiteboard interview in a nutshell. Standard problems given to everyone and asked to answer them in a standard way. Done in person to lower the chance of cheating.

The same reasons whiteboard interviews are problematic will apply to any test done at scale.

Not standard questions. Companies switch away from a problem as soon a it becomes known they use it. They want a problem the applicant has never seen before. And I'm saying, it's not because they are testing for intelligence -- they will help you through it if they like you -- it's because they want you to struggle with it. Which gives them the right to use their gut to decide.
That's not been my experience at all. Companies keep using the same problems that are on leetcode or some minor variant of it. More to the point, in many places companies don't assign problems, the interviewer has full leeway in doing that. They will keep using the same problem since they have a rubric for how people perform on it.
I have never worked at a FAANG, but people I know who do, and give interviews, tell me that questions get banned.
If that's the case eventually every leetcode question will be banned and the problem will have resolved itself.

     if we can't write a test to select the qualified humans, then either we're too lazy to write one, or, more likely, we actually want to leave plenty of room for human bias to do the actual selecting.
What makes you think that such a test exists? Why is that a given?
Do we ever say, for the code we write, that no such test exists for it? Why is this different for the jobs we hire for? What is this ineffable thing that can't be tested for?
These kinda of “unbiased” exams only work if people can’t cheat (either directly, by learning the questions from someone who already took the exam, or indirectly by preparing in a way that helps them score more on the exam, while not actually improving the skills the exam is supposed to be a proxy of).

Creating such an exam is an unsolved problem.

The test can be open source and dynamic. It can be a test the applicant can take over and over again every day to practice. When she applies for the job, she takes that very same test under controlled conditions so we know she is taking it and not someone else.

Why can't we write a test to see if she can code in Python at an appropriate level of expertise without mixing it in with algorithm gotchas? What are the algorithms she needs to know? Why can't we provide her with a list of those algorithms and show her how we will test for mastery of them?

>The test can be open source and dynamic. It can be a test the applicant can take over and over again every day to practice.

>Why can't we provide her with a list of those algorithms and show her how we will test for mastery of them?

That's exactly what is done, it's called leetcode.com, recruiters send you a link to it and tell you to study. I don't see how anything you're describing isn't covered by the existing whiteboard interview dynamic.

What company gives you the full list of questions, and the acceptable answers? Which companies use only the results of these tests?

leetcode is an open-ended ocean of knowledge you have study, it's not a test.

Most every coding question asked to me by FAANG was on leetcode (I think 1 out of 10+ wasn't but maybe I just missed it). Had I gone through all of leetcode I'd have covered almost every question asked of me. The behavioral questions asked were basically told to me by the recruiter verbatim ahead of time so were not a surprise in any way. The system design is broader but most cover a few common areas that you can google to find.

So it's all standard questions that have answers you can study for ahead of time.

I think we're talking past each other. I'm saying that no company tell you that (1) all questions you will be asked are on leetcode, (2) these are the acceptable answers -- passing the tests does not always count as acceptable, and (3) if you pass all these tests you will be offered a job, or entered into a lottery with all other people who pass them?

leetcode as used by FAANG is a hurdle that you can't ever be sure you actually cleared. it's not a test that returns a pass fail score.