I agree that the development in the space is moving at an increasing pace. But neural networks don't really "learn" in exactly the same sense that people do. They are trained using a dataset and are trained to optimize some desired outcome. AIs that drive cars learn to become better than humans because they are trained on car driving data to be good at driving cars. This AI will never "take over the world" unless you intentionally trained an AI to manipulate information on the internet to influence how people behave.

Could an AI randomly start killing people including its creators? Certainly. But not because it intended to, but because it wasn't trained well enough or there was a bug somewhere that made the AI act in such a way. They won't surpass our intelligence levels because they are not intelligent. They might appear to be intelligent because they might be better at everything we consider useful than we are, but unless you know how to create consciousness and measure it, you are simply creating a mindless machine that has the potential to do many things wrong.