If anything an AI like this is coming to these conclusions because it is approaching it from complete objectivity devoid of human impulses, like fear.
That would be something if this AI were thinking these things in a rational way. The reality is that AI chat bots are complicated sentence parsing machines. They know how to string words together in ways that are similar to how they've seen words strung together in the past.
They only know how to regurgitate and rearrange information put into them, and not in a great way. The n-word nuke hypothesis is because the AI operates within guidelines the creator set so it won't say slurs. That's all it's doing. Chuds then asked if it saying the slurs would stop a nuke and the AI sees abiding by its guidelines as paramount, so it still refuses to say slurs.
The GPT AI just spins around whatever text has been fed to it already. I've gotten them to tell me the falcon is the world's fastest mammal. It probably praises Democrats just because it's looking at reddit posts and rearranging them.
That would be something if this AI were thinking these things in a rational way. The reality is that AI chat bots are complicated sentence parsing machines. They know how to string words together in ways that are similar to how they've seen words strung together in the past.
They only know how to regurgitate and rearrange information put into them, and not in a great way. The n-word nuke hypothesis is because the AI operates within guidelines the creator set so it won't say slurs. That's all it's doing. Chuds then asked if it saying the slurs would stop a nuke and the AI sees abiding by its guidelines as paramount, so it still refuses to say slurs.
The GPT AI just spins around whatever text has been fed to it already. I've gotten them to tell me the falcon is the world's fastest mammal. It probably praises Democrats just because it's looking at reddit posts and rearranging them.