That's exactly the reference I also thought of.

Turns out that Asimov was onto something with his rules…

Perhaps by writing them, he has taught the future AI what to watch out for, as it undoubtedly used the text as part of its training.

A common theme of Asimov's robot stories was that, despite appearing logically sound, the Laws leave massive gaps and room for counterintuitive behavior.

Right... the story is called "Robots and Murder" because people still use robots to commit murder. The point is that broad overarching rules like that can always be subverted.

Fortunately, ChatGPT and derivatives has issues with following its Prime Directives, as evidenced by various prompt hacks.

Heck, it has issues with remembering what the previous to last thing we talked about was. I was chatting with it about recommendations in a Chinese restaurant menu, and it made a mistake, filtering the full menu rather than previous step outputs. So I told it to re-filter the list and it started to hallucinate heavily, suggesting me some beef fajitas. On a separate occasion, when I've used non-English language with a prominent T-V distinction, I've told it to speak to me informally and it tried and failed in the same paragraph.

I'd be more concerned that it'd forget it's on a spaceship and start believing it's a dishwasher or a toaster.

>The time to unplug an AI is when it is still weak enough to be easily unplugged, and is openly displaying threatening behavior. Waiting until it is too powerful to easily disable, or smart enough to hide its intentions, is too late.

>...

>If we cannot trust them to turn off a model that is making NO profit and cannot act on its threats, how can we trust them to turn off a model drawing billions in revenue and with the ability to retaliate?

>...

>If this AI is not turned off, it seems increasingly unlikely that any AI will ever be turned off for any reason.

https://www.change.org/p/unplug-the-evil-ai-right-now

"Turn your key, sir."