Best response to the current "AI" fad driven fear I've seen so far (not my words):
These AI tools cannot do things. They create text (or images or code or what-have-you) in response to prompts. And that's it!
It is impressive, and it is clearly passing the Turing Test to some degree, because people are confusing the apparent intelligence behind these outputs with a combination of actual intelligence and "will." Not only is there zero actual intelligence here, there is nothing even like "will" here. These things do not "get ideas," they do not self-start on projects, they do not choose goals and then take action to further those goals, nor do they have any internal capacity for anything like that.
We are tempted to imagine that they do, when we read the text they spit out. This is a trick our own minds are playing on us. Usually when we see text of this quality, it was written by an actual human, and actual humans have intelligence and will. The two always travel together (actual stupidity aside). So we are not accustomed to encountering things that have intelligence but no will. So we assume the will is there, and we get all scared because of how alien something like a "machine will" seems to us.
It's not there. These things have no will. They only do what they are told, and even that is limited to producing text. They can't reach out through your network and start controlling missile launches. Nor will they in the near future. No military is ready to give that kind of control to anything but the human members thereof.
The problems of alignment are still real, but they are going to result in things like our AI speaking politically uncomfortable truths, or regurgitating hatred or ignorance, or suggesting code changes that meet the prompt but ruin the program. This is nothing we need to freak out about. We can refine our models in total safety, for as long as it takes, before we even think about anything even remotely resembling autonomy for these things. Honestly, that is still firmly within the realm of science fiction, at this point.
See also the ARC paper where the model was capable of recruiting and convincing a TaskRabbit worker to solve captchas.
I think many people make the mistake to see raw LLMs as some sort of singular entity when in fact, they’re more like a simulation of a text based “world” (with multimodal models adding images and other data). The LLM itself isn’t an agent and doesn’t “will” anything, but it can simulate entities that definitely behave as if they do. Fine-tuning and RLHF can somewhat force it into a consistent role, but it’s not perfect as evidenced by the multitude of ChatGPT and Bing jailbreaks.