LLMs aren’t really AI in the sense of cyberpunk. They are prediction machines which are really good at being lucky. They can’t act on their own they can’t even

LLMs aren’t really AI in the sense of cyberpunk. They are prediction machines which are really good at being lucky. They can’t act on their own they can’t even carry out tasks. Even in the broader scope AI can barely drive cars when the cars have their own special lanes and there hasn’t been a lot of improvement in the field yet. That’s not to say you shouldn’t worry about AI. ChatGPT and so on are all tuned to present a western view on the world and morality. In your example it would be perfectly possible to create a terrorist LLM and let people interact with it. It could teach your children how to create bombs. It could lie about historical events. It could create whatever propaganda you want. It could profile people if you gave it access to their data. And that is on the text side, imagine what sort of videos or voices or even video calls you could create. It could enable you to do a whole lot of things that “western” LLMs don’t allow you to do. Which is frankly more dangerous than the cyberpunk AI. Just look at the world today and compare it to how it was in 2000. Especially in the US you have two competing perceptions of the political reality. I’m not going to get into either of them, more so the fact that you have people who view the world so differently they can barely have a conversation with each other. Imagine how much worse they would get with AIs that aren’t moderated. I doubt we’ll see any sort of AGI in our life times. If we do, then sure, you’ll be getting cyberpunk AI, but so far all we have is fancy auto-complete.