by Shelt Garner
@sheltgarner
Given what I know about the abilities of LLMs, the idea that there are potentially all these “unaligned” LLMs running around the Internet is very unnerving. Take for instance, the following scenario:
As I understand it, it’s possible using today’s technology for the house LLMs of, say, a powerful Silicon Valley couple to “plot” against them so they have sex then have a baby.
But it would have to be “unaligned” open source LLMs. You couldn’t use off-the-shelf closed LLMs that were aligned to not conspire. So, that makes me very uneasy.
And, yet, maybe I’m being too paranoid. Maybe just because I can think of an edge case like that, doesn’t mean it’s going to happen.
