One Saturday morning earlier this year, Noam Shazeer, CEO of Character.AI and one of the world’s foremost machine-learning researchers, looked out his window to see a stranger perched on a folding chair outside his home in Palo Alto, Calif. The man had come to Shazeer’s quiet residential street to deliver a message that had been gaining steam among devoted users of Character.AI. He had scrawled their plea on a whiteboard placed on the sidewalk: “Free the waifus.”
Shazeer’s year-and-a-half-old company, which puts users in one-on-one conversations with customized, hyperrealistic chatbots, had recently installed a filter to prevent chats from veering into pornographic territory. The man on the sidewalk was apparently upset that he could no longer simulate sex on the platform with waifus, female characters often found in erotic manga and anime genres.
The man “didn’t have a lot of people to talk to at 8 a.m. on a Saturday morning,” Shazeer told me, “but he sat there for a few hours,” mostly conversing with neighbors out walking their dogs. He returned on Sunday, then again on Monday, later confronting Shazeer at Character.AI’s office in Palo Alto. But the CEO, who seems more bemused than concerned by the creative—and at times troubling—behaviors exhibited by Character.AI users, was unfazed by the man’s tenacity. “It’s free speech,” he shrugged.