In theory, the purpose of the admissions test (or job interview) is to find people capable of doing the work, so they won't wash out from the hard, demanding programs. But none of these tests are rewarding people for doing useful or directly applicable work in the first place. So ideally you'd want a test that didn't require wasted zero-sum-game studying marathons, but would also correspond to the type of work involved... again, I don't think IQ tests are this. But such a thing would be useful, if it could be devised. If you still look at grades in addition to the standardized tests, like in the US, you'll wash out the not-willing-to-put-in-any-effort crowd like that.

To steal another idea from the job interview discussions: what you want might be entry-level intro survey type classes that are broadly open to lots of people (maybe online) but have a high bar for completion to gain full admission, and focus on some of the hands-on aspects of the field in question. "Lower bar + weed out courses" is used in a lot of places that couldn't get away with the high bar of MIT or such.

Alternately, I often suspect the best solution would be fully randomized admissions. Does it make sense to stratify by institution, so that you have one single gate into university, instead of looking at the output of four years of work across mostly-evenly-distributed places? Where you can more legitimately assume that what someone gets out of their time will be the result of what they personally put into it?