I will explain the proof of a result of Mikhail Raskin published this year (ICALP 2018) that solves a question open for more than a decade: what is the complexity of complementing an unambiguous automaton. The result is that complementing an unambiguous automaton may require a superpolynomial number of states, even if the output automaton is non-deterministic. The opposite fact was conjectured.