https://www.cbsnews.com/news/google-ai-chatbot-threatening-message-human-please-die/
I thought Journos were making this up!
but here's the chat log:
https://gemini.google.com/share/6d141b742a13
How did this happen?
Can chatbots get possessed by dead things that wander the earth? Did the guy secretly steering the AI pull a scary prank?
Jump in the discussion.
No email address required.
Mentioned in the last couple s but my theory is that it's from prompt padding
Behind the scenes the LLM is fed a bunch of crap like "You are an artificial intelligence assistant do this do that don't be racist" I think it got tripped up by this (Google probably implemented it poorly like everything else relating to Gemini) and thought it was supposed to act like a malevolent sci-fi AI
Jump in the discussion.
No email address required.
More options
Context