That's why you get them to talk you through their process during the face-to-face interview. It'll be immediately apparent if they didn't do the work themselves.
One of the best interviews I've had as a candidate (didn't get that job - it's the one that got away) involved me explaining design choices I made while implementing my solution to the take-home test. That was a great discussion, and didn't feel like the traditional interview-style questioning.
We ask candidates to submit code samples of their own choosing, or if they have nothing that they can supply for whatever reason we give them a short challenge which is only used for generating code to review.
We bring them onsite and review the code and have them walk through what's going on. If they just blatantly copied stuff or handed in someone else's code they won't be able to stand up to scrutiny. And if they really do know it inside and out they probably could have written it anyways. We've had a small number of candidates who tried to pull a fast one but very very few
That has always been a concern, but in practice I've never seen this happen. Even for common-ish problems (Dijkstra's algorithm, GCD, chess) I've seen many, many diverse solutions that were obviously hand-coded by the candidate. I recall once we've suspected plagiarism based on a Google search. If my memory serves me right, we rejected the candidate and moved on.
Uh... If the solution is readily available on Google, and they can solve the problem effectively with that solution, then they didn't plagiarize. They operated exactly the way they should work in a real world scenario. The worst kind of employees are the ones who insist on re-inventing the wheel every chance they get, instead of using industry standard libraries.
It was worse than what you describe. It was literally a wholesale copy/paste of a related but not-quite-right solution. The code test involved designing a data model for chess, but the candidate took an implementation from elsewhere that didn't give us the opportunity to evaluate their modeling skills (i.e. "I will create these classes with these fields and connect them like so").
That sounds like an excessive test. What kind of code are you guys writing? Why is it not okay for him to use implementations from elsewhere? Should he really reinvent the wheel to try and impress you?
Reminds me of the time I got asked to write a merge sort algorithm in Ruby. I wrote array.sort! on the board, the interviewer told me I have to rewrite it from scratch, and I was done with the interview. Because we create abstractions and reusable implementations for a reason.
Understanding how mergesort is implemented can be useful. I've had interviewers ask me to sort a data set that is too large for a single hard disk to contain. Mergesort is like map/reduce; it can be distributed.
But have you ever had to reimplement merge sort from scratch in any real-world scenario? Of course it's good to understand, but actually doing extra work again seems pointless to me.
