Deepfates notes artificial agents lack default human cognitive similarity
@deepfates @thkostolansky there will be individual instances of continually learning agents that become misaligned, the multipolar benefit is resilience through heterogeneity + institutions which favor alignment
society has evolved well to deal with this same dynamic in humans, we can do it again
@thkostolansky do you think the models we currently have or will have soon are not sufficiently capable of being dangerous?
No Digg Deeper questions have been answered for this story yet.
@willccbb @thkostolansky I mean this is the hope yeah. But there are obvious fail modes including the concept of having institutions which favor alignment. also humans have way More default cognitive similarity and incentive/motivational alignment
@deepfates @thkostolansky there will be individual instances of continually learning agents that become misaligned, the multipolar benefit is resilience through heterogeneity + institutions which favor alignment
society has evolved well to deal with this same dynamic in humans, we can do it again