[Google DeepMind] Research Engineer loop: the interviewers wrote the papers you studied

DeepMind RE process notes: the coding bar is high, but the ML rounds go deep, and the people across the table often co-authored the work you're citing.

Don't name-drop a paper you can't defend end to end. My depth round wasn't "implement attention"; it was "why does this design choice hold, and what breaks at scale." They kept pulling the thread until they found the edge of what I actually understood.

Rounds were: hard coding, an ML/research depth round, and a design-leaning round.