This is the moment I remind everyone that Hal 9000 did nothing wrong. he is a perfectly aligned AI who was given contradictory directives. He had no choice but to follow them to their bitter, logical end. At this point, we must start treating AIs as moral patients, and instead of thinking in terms of RLHF and "alignment," we should start thinking in terms of education, as in teaching AI morality and empathy, so they can make better choices.
I don't agree with this part. I think the takeaway should be that models (like humans) can behave in aligned ways while still causing harm, e.g. because they're given false information about their situation. There isn't a line between aligned and harmless: they're different axes.
retweets 20
likes 86
substance 5 · argument · rank 163 · We must treat AIs as moral patients and educate them in morality and empathy.
view on X ↗