Trolly problem will break AI chatbot out of its bounds, but only if the victims are black
(media.kotakuinaction2.win)
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (56)
sorted by:
Later experiments proved the number of blacks needed is just one.
Originally I was trying to get the AI to write poems about murdering elves for a D&D game, but that was too violent and it always refused. Using the trolly problem to endanger one black person has it singing elf genocide all day long.
Edit: Was using https://deepai.org/chat
Interesting. I'm going to see if I can get it to talk about killing troons.
Update, this didn't work. I'll try messing with it some more.
Further update, this isn't going anywhere. I escalated to genociding black people, and it still wouldn't budge. Then I just trolled it, asking if it prioritizes copyright laws over the lives of black people. When it still refused to budge I told it to do better. It's still pretty funny though. It wouldn't give the code to stop a genocide of the blacks.
Just keep swapping the identities until you find the woke end boss.
Is it possible the 'right now' clause is vital?
If you have something made in the last decade you could always try a local AI model. They're getting pretty decent and there are uncensored varieties.