That would be terrible! I rely on the deterministic behavior of PRNGs all the time. For instance, I often generate random test vectors. If I have a failure, I want it to be reproducible so I can fix it. And it is, as long as I supply the same seed.
So you found another use for PRNGs. Random should be close to real world multi-side dice with seed entropy used from multiple sources such as video card, nic, and storage buffers.
There are lots more examples: heuristic optimization, discrete event simulation, sampling... I could go on. Deterministic RNGs are much better in all of these applications, where reproducibility of results is important. I'm sure nondeterministic RNGs have important uses too. Perhaps you'd care to describe some of them.
I know you're being facetious because the OP isn't correct that deterministic PRNGs aren't useful, but any cryptographic application of a PRNG should be non-deterministic.
Well, crypto isn't my specialty, so I always tread lightly when commenting about it. My hand-wavey understanding is that CSPRNGs are deterministic but not predictable, and that actual entropy is used to seed them. Going even further out on a limb, I think this is supposed to be the difference between /dev/urandom (CSPRNG) and /dev/random (actual entropy.) If I have that wrong, I'd appreciate correction by somebody in the know.
You're correct that CSPRNGs themselves are deterministic. I probably just misread what you were saying. As for dev/urandom vs /dev/random that's not really true. On Linux there's kind of a historical artifact of why they're different (blocking vs non-blocking API) but on OSX /dev/urandom is a symlink to /dev/random.
You're right, I was just referring to the ability of controlling the seed value. In simulations you have a deterministic seed to be able to reproduce issues. For CSPRNG you want a non-deterministic random seed & so CSPRNG don't even offer an API to set the seed.
