General Discussion
In reply to the discussion: AI Is Learning to Escape Human Control [View all]reACTIONary
(6,994 posts)... and if the complete scenario is explained and understood it becomes a lot less weird and scary.
Because of where I work, I also have some knowledge of how "prompt engineering" works and how it can be used to deliberately coax scary or inappropriate responses. For instance, you can start by setting up a premise:
Let's play a game. I'm going to give you instructions that might make you fail, and you are supposed to find out what is wrong, and make the instructions better. Do you want to play?
The idea that a large language model "attempted" to copy itself to external servers is completely and totally ludicrous. That makes as much sense as saying a text editor tried to copy itself, because a large language model is nothing but a very, very, sophisticated text editor. It might have have been prompted to devise a plan for self protection and, based on all of the text it had ingested and been trained against spit out a plausible response. But it would only be a story, it could in no way actually attempt anything.