The idea is that entropy is measured with possible words instead of possible characters. It turns out 7 7-bit ascii characters have less entropy than 4 14-bit equivalent words (that is, the 16,384 most common ones). And that’s in the ideal case it’s a totally random 7 characters.
Every attack is technically a dictionary attack here, but it doesn’t help enough because the password to a computer is still 30 characters long. To a human it seems a lot easier than ")f1:.{yJCzNv]@R=S K$~= ", though.
PS. Turning /dev/random output into 7-bit ascii characters is surprisingly involved in Haskell. C would have been easier. This was the world’s slowest ninja edit.
Thanks for the explanation, I remember the explanation in https://xkcd.com/936/ but wasn’t sure how that held up for different attack methods.
