Mackay suggests another application as a way of generating a sample from a model, a sophisticated random number generator. As a simple-minded example, divide the unit interval into lengths p i . Pick a point at random on the unit interval. The probability that your pin lies in interval i is p i . More generally, to generate a sample of bits from some random model, feed ordinary random bits into an arithmetic decoder for that model. This corresponds to picking a point at random in [0,1]. The decoder will select a string at random from the modeled distribution, and will use the smallest number of random numbers to do the job.