1. 3
    1. 2

      I was expecting this to be about getting a consistent random distribution across the [-1.8*10^308, 1.8*10^308] 64 bit floating point range.

      It would be interesting to account for the increasing ULP as the mantissa consumes more and more of the precision

    2. 0

      “Random float” is itself an ambiguous concept. Here it’s defined as a 53-bit integer, but one could also expect a random value between 0 and 1.

      1. 1

        No, the function in the article returns a number between 0 and 1. It generates a 53-bit int and then divides it by 2^53.

        1. 1

          Oh, right. :facepalm: