Comment by selfhoster11

AI can become a highly autonomous system that's separate from us. Current technological limits make it currently a hard sell.

LLMs, viewed as general purpose simulators/predictors, don't necessarily have any agency or goals by themselves. There is nothing to say that they cannot be made to simulate an agent with its own goals, by humans - and possibly either by malice or by mistake. Model capabilities are the limiting factor right now, but with the rise of more capable uncensored models, it isn't difficult to imagine a model attaining some degree of autonomy, or at least doing a lot of damage before imploding in on itself.