Raw LLM Responses

2nd try posting my too long thoughts: Some people commented already that AI doesn't not think and the fictional scenarios are a little bit ... excessive. Thanks to @adupthetotal782, @akirakasinata8610 and @FelipeCostaPiano. AI is a tool. It depends on what you do with it, create stuff or destroy stuff. About blackmailing: current LLMs are trained on stuff from the internet, including a lot of shit found on the internet. It's just 'replicating' blackmailing shit from social media, DMs, whatever. Reddit :) The vector representation for 'turn off' is most likely near 'kill' and 'destroy' and so on. This way LLMs replicate blackmail stuff written by horrible people. AI doesn't fear being turned off.