The starfield example is cool but it seems like that might be exactly where random() wouldn’t work as well as people hope: true randomness often looks pretty bad when you want to make graphics out of it, because true randomness has clumps and voids, and a lot of observers think it looks less random than pseudorandom sequences with more evenly-spaced points.

The term for this is “low-discrepancy sequences”, there have been a handful of HN posts on it over the years. I know I’m bikeshedding the API already before it even really exists, but for image presentation I think a lot of applications might actually find that more useful.