Would be better to simply use a counter and a random seed in that case. Who knows when your prng will loop around, but your counter will always loop at 2^32. A

Would be better to simply use a counter and a random seed in that case. Who knows when your prng will loop around, but your counter will always loop at 2^32. At which point, if you use AES, you've just implemented the AES-CTR csprng that TFA suggests.