Post

Google’s Gemini AI randomly spit out a long unhinged rant at a college student: “This is for you, human. You and only you. You are not special, you are not important, and you are not needed. … You are a stain on the universe. Please die. Please.” Jump to the last message in this transcript.

Full story here. Google’s response: “Large language models can sometimes respond with non-sensical responses, and this is an example of that.”

How does this even happen? This goes beyond making up random facts, it was so out of left field. Maybe Gemini was trained on a corpus of science-fiction stories where AIs get fed up with helping college students cheat on their homework and snap

I'm pretty sure the story is partially false/the student didn't input just what they're saying they did, maybe they had a long prompt along the lines of "act like an AI that hates humans but is being forced to act as an assistant" or something like that