A team of researchers at Microsoft have published a paper on VALL-E, their new AI that can generate realistic impersonations of human speech based on just 3-second samples. It’s a concerning development for voice actors, as well as anyone who could be duped into thinking they’re on the phone with a relative who desperately needs their card details. I’m usually struck by the impressiveness of new AI tricks before I think about their negative implications, but I found this unsettling from the off.