🚀 go-pugleaf

In article <1992Jul26.185623.10224@msuinfo.cl.msu.edu> mrr@scss3.cl.msu.edu (Mark Riordan) writes:
>However, a cryptographically strong function, such as 
>DES or MD5, probably would be suitable.  In fact, I wouldn't be
>surprised if crypto-based pseudo-random number generators turned
>in strong performances on most or all general-purpose PRNG tests.  

My favorite definition of a cryptographically strong PRNG holds that you
can't predict a future bit with probability better than ((1/2)+epsilon)(*),
knowing all bits which have already been output and having an arbitrary,
polynomial-size machine to do the testing.  Each general-purpose PRNG test
qualifies as such a machine.  So, if any PRNG fails any such test, then
that's grounds for declaring it cryptographically weak.

(*)[epsilon is, for example, o(2^(-k)) where k is a security parameter (eg.,
# of bits of key) under control of the user]
(Definition from Silvio Micali -- any mangling is from my faulty memory)