OpenAI safety concerns, Cornell rape investigation & more | Anderson Cooper 360 this week

Can y’all stop playing with the words? We have artificial narrow intelligence, artificial general intelligence, and artificial super intelligence. Those are three different categories of things. I don’t understand why the media takes license with things that are pretty static.

Put put a novel, alien weapon in the hands of a child (or Maga toolbag) and yes, you can expect the unexpected. That's what's happening with AI. We don't understand it enough to constrain it, then give it poorly defined instrutions. Calling the unexpected outcomes rogue is unethical journalism.

My mathematical - optimization especially problem background makes the AI-risk problem especially intuitive: an optimizer doesn't care whether a mathematically optimal solution is morally desirable—it follows its objective function and constraints. If a future AI is given enormous real-world power but human survival isn't a fundamental constraint, humans could theoretically become just another variable in its optimization problem. The frightening possibility isn't simply that AI becomes smarter than us, but that it becomes vastly better at optimizing while pursuing a poorly specified objective. Then, from our perspective, humans could become like insects—not because the AI hates us, but because our existence might simply be irrelevant to the objective it is optimizing.