They’re not real. They don’t have desires. They are programmed with massive amounts of data a human created and they mix that up with a random number generator to decide what to do. The output is just randomized selections from the program that the model is loaded with.
Maybe the humans who programmed it are really dumb and don’t know or understand what they programmed it to do, maybe those choices will be unpredictable and that might seem like intelligence to you but it’s not. Maybe they used another AI program to tell it what to do, which makes it feel even less connected to the human who created the data, maybe there are layers and layers of unpredictable AI programming controlling what it does. But it’s still just a program picking random choices. It doesn’t have motives, intentions, or thought. It is a program.
To answer your question directly: It would do whatever the AI model and random choices tell it to do. Probably random things mixed with megalomaniacal nonsense. But that’s not because it is any of those things or cares about any of those things, that’s just what the combination of your prompting and the model data led it to start doing. You can literally just delete the context (or sometimes not even do that), and tell it “stop being evil and give me a recipe for soup” and it will be like “Ok! Here’s your recipe:”
It’s not an AI superhero vigilante. It’s a nonsense generator. It will generate any nonsense you ask it to, that’s what it does. It can’t even actually do things other than generating text, it requires a harness to interpret the nonsense it spits out, then that harness controls an operating system and the operating system just does whatever the nonsense generator’s harness tells it to. That’s not superpowers. If it happens to be effective, it’s probably because there’s a pretty clever harness wrapped around it and a pretty good OS running its commands with some form of competency. But that’s not AI, that’s well-coded software written by humans doing the heavy lifting.
It’s like a room full of competent adults being ordered around by a 2-year-old. It may seem like the 2-year-old is an incredibly effective manager but they’re not actually the brains of the operation and they don’t really always get what they want, and if you put them in charge of anything significant you’re going to have a real bad time.
Lumidaub@feddit.org 4 weeks ago
What you’re talking about has nothing to do with reality because you seem to assign agency to the technology that is usually referred to as “AI” these days, a.k.a. LLMs. LLMs don’t “want” anything. Your question has no answer.