[digging]

> You (and pretty much the entire debate around AI safety) have smuggled in the notion that these AIs even have the capacity to "love" and "hate," and have the agency to perhaps act vindictively.

It's a metaphor, actually. I guess I needed heavier use of quotes.

The one thing we can be sure of is that if goal-oriented AI exists, it will follows goals.

> How we get from predicting the most plausible next token to a god-like entity with agency is... not clear at all.

It doesn't have to be clear. If it was clear, we'd be working on it. That's... how invention works.