Google seems to be improving. They finally realize that they need to be slightly more open-minded in terms of candidate selection than before, to avoid missing out on some real talent, which is becoming rarer than before.

They hand you over from one recruiter to another. One does the initial talking, then another talks more specifically about the position and location, and then another one takes care of the interview.

They do interview on-site in Mountain View, California. The place is nice to visit, although I would not personally move there.

I find Google's interviews quite different from others, and it clearly demonstrates the company's goals and priorities. I am not judging if they are right or wrong, but it is different. Most technology companies want good engineers who know how to use their brains to solve problems and the tools and technologies to implement these solutions. Google is mostly (and solely, during the interview) concentrating on the first point. They do not seem to care if you have many years of proven experience in relevant domains. You will be asked to solve some problems that require computer science knowledge at the university level, at least.

Personally, I find this practice questionable. While it is good for pure research engineers, it gives a somewhat partial view of some candidates. But it looks like Google is trying to stick to their method of selecting candidates.

The most difficult questions are about particular algorithms or mathematical problems that the interviewer is very familiar with, but the candidate has never heard of.

This can create a stressful situation, as the candidate may think it's something they are supposed to know. This approach is not how people solve new problems in IT.

Many people may not be able to fully understand a particular mathematical (e.g., theoretical) theorem in 10-15 minutes. Without a good understanding of that theorem, it may be hard to apply it to the real-life problem presented by the interviewer.

For example, the question about finding the majority number in a large input sequence.