From what I have read from Gayle Laakmaan, they know this throws away many good engineers, but it is acceptable to them because it doesn’t let bad ones through.
I wonder now: for every "bad one" not let through by this sort of process, how many other "good ones" look at it and say "no thanks"? How many hear about this sort of hazing and are dissuaded from applying in the first place? How many experience it one too many times and just quit the industry altogether? How many of those "bad ones" are even objectively bad, and not just having a bad day or intimidated by a process rife with both intentional and unintentional hostility? In other words: are we actually assessing what we claim to assess with _any_ predictive accuracy, and is the collateral damage to company reputation and the pool of available candidates - a damage often hidden to the individual companies that impose this process - even worth it?
And that doesn't address the file drawer effect. I'm 54. I can code a binary tree, hash table or what have you, if I have to, but I am not as practiced as somebody fresh out of school, because that heavy lifting is done by libraries, and I'm busy contributing original mathematical algorithms that no one ever asks about because they don't have the background to understand the explanation. So I don't go on FAANG type interviews. I know I will fail, and if I don't my offer will be predicated on that apparently lower performance. At best I will have an utterly miserable experience. So I self select myself out, mostly on age.
This is not good, for so many reasons.
This made me chuckle, because it's absolutely how it is.
Some other comments here talk about how nobody creates new algorithms these days unless they are a researcher. But it's not true at all! Like you I'm creating new algorithms and other tricky techniques all the time, but I'm not fresh on the stuff I learned at school... or so I thought.
I have been really surprised to find some screening interviews recently asked me shockingly trivial questions. So simple that I stumbled over myself trying to simplify the answers, thinking "I know too much about this topic and need to keep it simple", and "can these questions really distinguish candidates?".
I did the TripleByte online test recently and found the questions much simpler than I expected, including those in languages I've never seen before. That is not TripleByte's reputation. I see people writing about how difficult they found the questions. (Admission: I didn't score all 5s, but I can't figure out why unless it's timing as I think I answered them all correctly.)
So I'm thinking, perhaps it's not so bad, people just make it sound bad because there's a wide variation in people's knowledge, abilities and expectations.
If you are thinking you might apply for something like a FAANG or other hard-reputation company, and feeling put off by the horror stories, I would say, just take a look at the old stuff a bit for a refresh, then give it a try and you might be pleasantly surprised to find their reputation is because people less skilled than yourself found it hard. Their "hard" might not be hard for you.
You'll probably still fail the interview if it's a FAANG because they are so selective, but I would bet my dollars to donuts that it would be more refreshing than miserable if you're not attached to passing.
I know your point isn't about yourself, it's about bias in hiring, including bias due to perception by the candidates, but I wanted to address that side point about feeling there's no point applying. That sounds like anxiety to me. (I have it too, I'm trying to get over it.)
Ooohhhh this is frustrating. Nothing is more frustrating than telling an interviewer about something cool you've done, and realizing they don't understand, but also don't care.
Google is basically using anxiety to filter good candidates and eliminate false positives which works in a sense but is still highly illogical.
Why not use a technical filter to filter for technical candidates? Anxiety seems like a pointless filter.... how does that even eliminate false positives?
People without that niggling sense of anxiety always in the background, the folks who are easily going to ace the programming interview because their anxiety levels are so low, those folks, I theorize, are potentially not going to be so careful. Now you could argue that that's a plus in many situations - a startup that needs to get code out the door right away, for example. But it really depends on the domain. For critical systems in applications like healthcare, avionics or robotic control I think the anxious coder is the one you want.
Companies that completely weed out the anxious programmers, as you imply, will have a different culture with perhaps too much emphasis on risky behavior.
Could this be deliberate? As working with people with anxiety problems is often a pain in the ass as they won't report problems for fear of seeming stupid or ask for help.
Confident people are far easier to work on a team with.
E.g. it's acceptable (but not ideal) not to promote talented commanders. It's catastrophic to promote someone to General who isn't ready or is unsuited.
I'm not sure how much engineer skill matters though. One slightly above engineer might be fine. But if half your org is made up of slightly above average engineers, I feel like there will be a knock-on effect.
Although as a counterpoint, it's also easier to get work, because part of the value I'm providing is that you "fire" me whenever you want and no hard feelings. So nobody does "leetcode interviews", because it's not like we're getting married, like an employee. So there's much lower risk to hiring someone, because a bad hire doesn't infect anything, because you expected it to be temporary in the first place.