Their process is, overall, a tad disingenuous.

They ask you to do a take-home assignment which is supposed to take "2 or 3 hours". Honey, unless you are actively building neural nets from scratch in your day-to-day, their take-home will take you all day if your model does not converge on your first attempt. All of you fast.ai, Keras and PyTorch importers need not apply.

Which is frustrating, because after spending an ungodly amount of time on their take-home, their "phone screen" interview was rather simplistic. They ask questions about your previous work and a "pair programming" (don't ask him/her to code anything, lol) challenge that was rather easy, given the world-class hard problem they asked in the take-home.

I hope their hiring process isn't to reject 100 people just to prove themselves rigorous in their new market. Their potential talent pool is small; they cannot just burn through 100 applicants and expect to hire an H1-B candidate like some other big companies can.

Consider an interval from 0 to 1 with ten uniformly distributed points in it. What is the distribution of the difference between the 6th and the 5th smallest point?