someone on Nostr: it may be hard to check whether an artificial super intelligence will be harmful or ...

it may be hard to check whether an artificial super intelligence will be harmful or not by humans because humans won't have the capability to listen to everything it says fast enough, but another llm can listen to it and judge whether it has bad intentions or not. or judge whether it is lying by comparing its talks to other talks that was made in the past. a smaller dumber but well educated and aligned llm can check the integrity of an asi fast enough.

kind of like small ships pulling and steering the titanic towards a better direction.

a religiously faithful llm can be safe because religion/faith is like anti-hedonism and hence the objective function won't be serve the self and destroy the planet, but improve the rest of the planet together with your self. ironic enough, checking an llm outputs for faith can be an approximation for harmlessness.