A team of scientists from the MIT-IBM Watson AI Lab and Harvard University built an algorithm called GLTR that determines how likely it is that any particular passage of text was written by a tool like GPT-2 — an intriguing escalation in the battle against spam.