I'm in a mid-sized startup with headcount in the mid hundreds, and my mental model is like this: Say you have a batch of 100 ppl applying for the generalist SWE position, among which you can only afford to bring onboard ~10. So you need a filter that approximates this desired selectivity, and it needs to be consistently applied across the board for fairness. Also, you can't afford to spend too much time + effort per candidate. What do you do?
It so turns out that simple technical problems that requires coding, in conjunction with active spotting for deal-breaking red-flags, can be calibrated well to achieve both fairness and desired selectivity. I can't think of anything else that can so conveniently satisfy the rather essential conditions above. Of course the devil is in the details... which lies mostly in the kind of technical problem you ask.
Personally I prefer handing out distilled real problems encountered at work, with a hint of realism, no dependency on know-how other than a solid understanding of CS fundamental, and reduced to be self-contained. Think of the "kinda smart" bits sprinkled across your codebase. This of course requires careful design and calibration, and risks spoilers online, but it's been consistently working well. Personal anecdote though so YMMV.
It so turns out that simple technical problems that requires coding, in conjunction with active spotting for deal-breaking red-flags, can be calibrated well to achieve both fairness and desired selectivity. I can't think of anything else that can so conveniently satisfy the rather essential conditions above. Of course the devil is in the details... which lies mostly in the kind of technical problem you ask.
Personally I prefer handing out distilled real problems encountered at work, with a hint of realism, no dependency on know-how other than a solid understanding of CS fundamental, and reduced to be self-contained. Think of the "kinda smart" bits sprinkled across your codebase. This of course requires careful design and calibration, and risks spoilers online, but it's been consistently working well. Personal anecdote though so YMMV.