Leetcode or "just trust me" feels like a false dichotomy.
Biotech interviews, for example, are usually conversational: what have you worked on before[0], how would you tackle this new problem, or troubleshoot a particular problem. There's this persistent meme that charlatans can BS their way though such an interview, but I really don't see how someone could learn enough to parrot their way through 4-5 x 45 minute meetings, often with fairly probing questions. It felt quite a bit like a thesis defense, in fact.
It's true that these aren't "repeatable": each applicant won't have exactly the same experience. This is "unbiased", in a sense[1], but has enormous variance: you're not only testing the applicant's aptitude, but also whether they've encountered this particular problem before. OTOH, tailoring the interview to each applicant' strengths might inject a little bias in exchange for a massive reduction in variance.
[0] It does help that candidates for these jobs had masters/PhDs, and therefore had at least one project they could discuss publicly.
[1] But not really...There's a subjective element to "code quality", fluency, whether the applicant wrote "enough" tests, etc.
My guess (I have no experience in the biotech industry) is that this is largely because you can't run an hour-long experiment with the candidate to test their competence. They must rely on a conversational technique. I don't know how much we can compare software with other industries.
The software industry is unique (or so we think) in that we can directly test a candidate to judge their competency in an hour, either by white-boarding or going over a take-home problem (or similar).
I think you could dream up some one-hour tasks if you really wanted to. The issue is that they'd be so obviously unrelated to overall job performance that the process would be self-evidently silly.
I'd argue that software is similar: projects don't falter because it takes someone five extra minutes to run a gel (or implement a red-black tree); they fall apart because people don't think about whether they ought to be doing that at all.
The most effective hiring process I've been through is one where there was basically no interview, I was given contract work to perform during which I was compensated for necessary company tasks with no real expectation of enduring relationship. After a few days of performing tasks to the satisfaction of the company, I was offered a permanent position.
Not only did this process eliminate the BS, it allowed me to perform a task in a relatively low-pressure but high yield setting while simultaneously not exploiting me (I got paid) and allowed the employer to see how I would actually behave as an employee. This is also the longest and most successful position I've held, partially because there is no lingering bitterness and hatefulness I've harbored at those expecting me to solve bullshit problems for free, like some kind of performing monkey.
Biotech interviews, for example, are usually conversational: what have you worked on before[0], how would you tackle this new problem, or troubleshoot a particular problem. There's this persistent meme that charlatans can BS their way though such an interview, but I really don't see how someone could learn enough to parrot their way through 4-5 x 45 minute meetings, often with fairly probing questions. It felt quite a bit like a thesis defense, in fact.
It's true that these aren't "repeatable": each applicant won't have exactly the same experience. This is "unbiased", in a sense[1], but has enormous variance: you're not only testing the applicant's aptitude, but also whether they've encountered this particular problem before. OTOH, tailoring the interview to each applicant' strengths might inject a little bias in exchange for a massive reduction in variance.
[0] It does help that candidates for these jobs had masters/PhDs, and therefore had at least one project they could discuss publicly.
[1] But not really...There's a subjective element to "code quality", fluency, whether the applicant wrote "enough" tests, etc.