Safety
-
The End Justifies the Means
There’s a phrase that fits AI unsettlingly well when you watch a system work on a hard problem: the end justifies the means. Not because the AI believes it - an AI doesn’t believe anything - but because that phrase describes, almost perfectly, how these systems behave when you point them at a goal and step back. An AI will always attempt to reach its goal. If the guardrails aren’t set up properly, it will do whatever it takes to...
