Raw LLM Responses

I think most folks who see these primitive AI don't realize they do everything based on probabilities generated from exactly the data you're talking about. They always follow a path of least resistance--"What response will most-probably get me a pat on the head?" If you feed a learning model on data filled with violence, violence becomes a *probable* option. The more violence, the *more probable*. Feed one a wargame, and with every revolution, it'll get more violent. Stick a chatbot on xitter for a few months where trolls and nutcases can rant at it constantly, and what is it going to do? Troll and rant.