Adapting to the candidate is probably good most of the time, but don't work too hard to make that happen. Computational geometry is on my resume, and interviewers are always trying to show that they know that too by trying to bend their question into that. The exercise very often detours as they've led me way down a wrong path because they're analogy doesn't fit, and all they wanted to know was if I knew about minimum spanning trees or even just heaps.