It seems like you could solve this by asking a question no one has ever been exposed to before, or by exposing each candidate to the theory relevant to the question yourself.
Except that the first strategy is guaranteed to fail some of the time, because there are no questions like that.
Why not go the other way?
If I could guarantee that every candidate would read my blog, I'd happily point them to an essay like this, and then in the interview we could work on something meaty, like pattern matching.
That would suit your second recommended strategy. But I wouldn't want to write an essay like this, and then favour those candidates who like to read Hacker News or follow me on Twitter. We're trying to select for talent, not for membership in a tribe :-)
So we're retiring this question, and we already have a few replacements that we use for many candidates. But I won't blog about them, in an attempt to hew closely to the first strategy you suggest.
We only use this for people at the beginning of their career, mind you. We have completely different expectations of someone with ten years of experience than we do of someone applying for a CAP (aka "work-term") or SDE I (aka "entry-level") position.
It's not that good people can't work it out from first principles, but it's clearly a lot more work for those who have to figure it out from first principles than for those who know immediately that a stack is involved and apply their focus to the loop and the stack and handling follow-up questions.
The ideal question or problem from our perspective is is one that is equally challenging for everyone, that way when we're having a candidate review, we won't need a lot of, "...But then again, the candidate clearly knew nothing about parsing, so they actually did quite well, considering."