habryka comments on Trying to deconfuse some core AI x-risk problems

[habryka]

Yeah, this also to me seems like one of the most interesting directions to go into, and might continue in that direction.