Dice, being physical objects outside of computer networks can be trusted to a pretty good degree.
When it comes to creating something random on a computer, it's actually incredibly hard to guarantee randomness. Most RNGs are black boxes or are so non-transparent that they might as well be black boxes.
You might have an idea how your *nix /dev/random works, but can you actually guarantee your OS is not compromised? Can your guarantee your BIOS is not compromised? Can you guarantee that your hardware is not compromised?
That's why if you want a truly secure Bitcoin cold storage, you generate a wallet offline using something like dice or coinflips (dice are quicker).
Is it more likely that the RNG is compromised than the bitcoin software?
> Is it more likely that the RNG is compromised than the bitcoin software?
1) That's a weird question. It's not like they are mutually exclusive. Both are possible. Dice allows you to eliminate the RNG problems.
2) http://en.wikipedia.org/wiki/Random_number_generator_attack#...
Isn't that my argument? Maybe I should try again to clarify my position.
'Just' an airgapped computer solves neither backdoored RNG nor backdoored bitcoin software.
A trusted airgapped computer comes close to solving both.
Using dice only gives me half a solution; why bother?
It doesn't have to be competely random, but when humans make up passwords they grossly overestimate the entropy levels; it's easier to say not to do it at all. Especially since the longer a password a human makes, the more likely they are to use shortcuts to gut the entropy, such as involving not only words but actual sentences.
Biased random number sources are okay, but you have to understand the bias, and it's easier to use an unbiased source. urandom is nearly perfect (post-boot) so use that.
If you simply use the C rand function that is bad. But he's assuming you're using something with real randomness, like urandom.
1) When he says "dice" is he talking about physical dice or something, or is there some protocol called dice that I am not aware of?
2) I feel like even though a computer generated random string isn't completely random, that it could still be effective. Could someone elaborate on this a little bit more? What if you do tricky things like concatenate multiple random strings together? What about functions like urandom (I'm pretty sure that is truly random.)?