A huge percentage of developers never bother applying to any FAANG (or similar) because they either don't believe they could pass so it'd be pointless, or they don't want to put hundreds of hours of study and go through several very painful days of soul-sucking interviews & travel for something that's not a sure thing, when they're already fairly comfortable as far as income goes (but probably not FAANG-comfortable). That's a pretty significant filter.
It filters for several things, including a very high level of confidence in technical abilities, and a strong enough desire to work in a FAANG(-alike) to put up with a whole lot of dull/painful unpaid bullshit, but the point is that the "hat" doesn't contain a random sample of developers. I think the notoriety of the interviews probably does do a good job of ensuring that the pool almost entirely contains people that, so far as tech chops go, are good enough to do the job—it excludes a bunch more of those from bothering to apply, of course, but the point is that I don't find it surprising that choosing randomly from their already selection-biased "hat" yields results of similar quality to actually conducting the interviews. The problem is that, if they loosen up the interviews, that will (I expect) stop being true.
Look at it like this: if you have a club that is notorious for only letting very pretty people in, I bet a random sampling of the line outside skews incredibly good-looking, but they can't just start randomly choosing people from the line without regard for their appearance, or that will change in a hurry. Meanwhile, it's also true that some very pretty club-goers don't get in that line in the first place, for a bunch of reasons possibly including that they don't want to be part of a scene like that, or a lack of confidence or anxiety over being harshly judged, or not thinking it's worth standing in line for an hour when they could go down the block to another club with lower standards and no line, and nearly all people who might be able to get in on an exceptionally weak night don't even think of trying, since more often than not it's pointless—but despite the loss of those folks, the line is still 95+% very pretty people, so it looks like they could do away with the high standards and harsh judgement and just let in a random sample by lottery, if you consider only that in isolation and not the effect the process itself had on the composition of the line.
(FWIW, this is a non-FAANG perspective—I'm 100% certain I'd struggle to even get past the phone screen without prep; about 95% sure I could do the actual work to an acceptable level; but judge it only about 50% likely I could eventually pass an interview with one or another after a great deal of practice and many, many tries, given probably 18-24 months of on-and-off effort and tens of extremely unpleasant days; so, I'm in the "unwilling to spend weeks of my life doing nothing but unwrapping Wonka Bars, when I'm not very sure there's a Golden Ticket in one of them" crowd)
I worked at a FAANG for a decade. though I've never done well in a tech interview. I've turned down offers and interview opportunies at some of the others--in some cases because I didn't want to bother with the interview process. Somewhere a long the way, after going through the hazing process a few times, I decided that it wasn't worth it, and stopped doing it.
Some of the offers I turned down--and one that I accepted--came after I performed quite poorly in interviews. Then again, as I've said, I have never performed well in them. I have the wrong kind of personality. I have to rely on other means.
Surely it's a little more specific than that? I'm used to winning in high-stakes standardized tests (which may explain why I'm favorably disposed to them), and in employers' assessments of my performance (and willingness to rehire me), but I'm used to losing in tech interviews.
I don't think so. Beyond that you are looking at the noise of the system which is why even over/perfectly qualified candidates have an interview failure rate of up to 50%.
