ǝʌɐpSomething I don't see talked about enough outside of esoteric AI circles is how our information environment that we stew in also affects LLMs. If our society writes a lot of fiction about AIs escaping containment then the LLM will have a lot of reference material built into it on that subject. All of our fears, but also hopes and joys, get baked into these things.
Jordan (Damn Good Tech)@dvshkn Definitely. I see a lot of clickbait posts on social media of stuff like, "I asked ChatGPT if it would become sentient in 30 years and it outlined its plan for world domination!"
No, you input a question into an LLM trained on trillions of data points (some of which includes AI fearmongering), and the next logical token was from a Terminator fanfic reddit post.