Except that there's a difference between what people can produce and what they can google up. When you let people out of the room to do something, you have no way of knowing if they did it themselves. Or if it took them 12h to do it.

When I was giving interviews, I'd have the candidate bring up the code sample that they submitted on their own laptop, and start making changes. If I found any bugs, I'd ask them to fix them; otherwise, I'd extend the problem in some way that their existing code didn't anticipate and ask them to change their code to meet the new spec.

Not only did I find this very helpful in evaluating prospective employees, several people that we hired told me later that it was the most fun they'd ever had in an interview. That is, of course, a biased sample; some of the people that I interviewed obviously did not find my interview style entertaining.

The litmus test that I generally used was whether they were able to, given an erroneous output of their own creation, they could reason about why it happened and fix it before the hour-long interview slot was over. Also, if I (watching over their shoulder) couldn't spot the cause of the bug, I considered that it was probably unreasonable to expect them to be able to find and correct it during the interview.

While I encountered a lot of candidates that froze in the face of compiler errors or started making changes without stopping to think about what the cause of the error might be, I never once got the feeling that anyone was less familiar with the code that they submitted than someone who had written it themselves.