The thing is that interviewing is hard; it is a full-time job, really, and one that could make or break the company. At the same time, interviewing is not something that current employees are remunerated for, materially or even in terms of time. It is very rare for deadlines to move because half the team is conducting interviews.
So, what ends up happening is that nearly everyone in the interviewing process ends up half-arsing it - from the recruiters who prefer to contact nearly everyone eligible based on some keyword searches - to the engineers who have no time to be thorough - to the hiring manager who usually has a bunch of fires to extinguish.
A few companies do tend to put quantifiable metrics around their processes but they're usually too reactive and inconsequential.
The interviewers have a vested interest in being more productive by hiring the best people, but at the same time, there is personal friction against hiring someone good/better when your own promotion is hanging barely by the 'meets expectations' thread.
Now the assignments - they're almost always totally useless. Like, build a new app based on a third party API and then make sure to test it. Useless because of multiple reasons - the API is nothing like what you use at work; your current product team has coalesced around some archaic architecture that no Greenfield app would use; it's the take home assignment of the month (like the favorite 'build a movie app using the OMDb API'). A better alternative would be to open up your own API and perhaps a module of your current production app and add something to it, create a PR, or even just fix a bug or two. I actually think asking candidates to document a module is much more useful than asking them to write code you would never use, even if you pay for such an assignment.
Onsite interviews are even more useless - it's usually the most vocal people asking esoteric questions about older technology or algorithms that they've had success with candidates stumbling upon. Again, almost none of the whiteboard exercises or questions are from real-world scenarios.
You could pay candidates for the interview process, but that doesn't even address why interviewing is so fundamentally broken. I bet we would have amazing software products if we just interviewed better, especially when companies have been able to lobby for right to work laws nearly everywhere.