[Jensson]

> This is called a “work sample interview.” (Leetcode problems don’t count, because most jobs don’t look like that. Neither do whiteboard interviews, for the same reason.)

They do count, job sample tests doesn't need to be exactly like the job just similar. There is correlation between performance on those tests with performance on the job, far from perfect but its much better than the zero correlation you get from these cultural fit tests or free form discussions.

[d0mine]

There was found a reverse correlation (on a small sample at Google) the better you are at leetcode-like problems, the worse you are at the job.

The correlation is gone on a larger sample https://news.ycombinator.com/item?id=25426329

[hooverd]

They're good for seeing how someone approaches a problem. Although with Leetcode everyone is just testing to see if you remember the One Simple Trick for that case instead of being able to talk through it.