Ex Machina

LLMs (current ones) have no concept of remorse. It is not that they are good or evil. It is that applying the concept of remorse to them is a categorical error.

I read somewhere — an LLM will say oops after accidentally deleting your production database, and oops after accidentally destroying humanity.

It is the same oops. The distinction between the two does not exist for it.

This is the paperclip argument in some sense, except it is not so much about alignment as about the inherent mindlessness of the process that is in motion.

This is not a new thing. In fact, if anything, humans and their conceptions of love and morality, especially morality, are exceptions in nature. Nature is mindless. Existentially terrifyingly so. A wild bear has no morality (which doesn’t make it good or evil; it just is).

For me, this tells me that I need to change the way I interact with them. I need to think of them as systems that respond in a certain way (bureaucracies, dogs, economies) instead of as a person who should have felt something.