I don’t think that would be very useful, because there would be a 1/2048 chance of getting a NaN, and it would be difficult to convert the numbers to a known di

I don’t think that would be very useful, because there would be a 1/2048 chance of getting a NaN, and it would be difficult to convert the numbers to a known distribution.