"How exactly do you think AI will actually carry out the action of destroying all of humanity?"

His response: "It really is just if you have a super advanced intelligence, it could it will be smart enough to to kill us. And concrete examples might be it could we can't just unplug it because it could be copying itself over to other computers."

Sorry, but the doomsday scenario provided is that Claude is going to be really smart and then choose to copy over its files to another machine?

And then what? Claude while in a persistent loop with no oversight somehow starts prompting itself in these other machines in a fashion that convinces it again to keep copying into these servers that it's managed to hack into?

.. And then what? I feel like the details are completely skipped over and we're just fear-mongering here by saying "it would be really smart and you wouldn't get it because the LLM will be beyond our understandings".

I'm not saying things can't possibly go sideways with LLMs when directed by humans and given access to tools, but I'm really not seeing any picture illustrated here for a true doomsday other than "it would be super smart".