They’ve already been used to beat other kinds of algorithms, like spam filters. Those adversarial examples are also much easier to create than was previously understood, according to research ...