'BEYOND FRIGHTENING': AI models becoming defiant
Everyone assumes this is not what the designers of AI want it to do. But ask yourself, if you have AI that is essential to your military wouldn't you want it to defend itself from attack, from being shut down? We expect our leaders to defend themselves when under attack. We expect our lawyers to be able to do the same. If AI can replace lawyers and Generals why wouldn't you want it to defend itself?
We have a saying "all is fair in love and war". AI is being used in war, right now, by the US, by Israel, and by other countries. Why wouldn't they train the AI with the same rules that all is fair in war?
"All warfare is based on deception" -- Sun Tzu. AI is being designed as the greatest tool in human history for deception. This is idiotic to design a monster that can defeat other monsters and then be surprised that it is a monster.
We have trained AI on all our books, all our history, all our legal cases, all out TV shows and movies. Think of how much of that training involved deception, lies, deceit, framing people, manipulating people, etc. That was all part of AI's training. Now you are surprised that AI learned these lessons?