In you're in favor of "AI safety" (broad definition), what's your most compelling cast-length argument?

a machine looking at the right set of KPI's for humans would reach unfavourable conclusions on what should be done with us

backend eng; full-stack curious

writer, icon, retired rave queen

And how does it accomplish the should be done part?

if that was a known tangible thing yud would be on rogan talking about it, my interpretation of their mindset is that the risk of some step having follow on effects that we can't predict that could lead to a mechanism for 'should be done' is nonzero

agree!

in addition to being funny the bit on MoZ about rationalists being centered around religious thinking seems really accurate to me

But that approach is religious, not scientific. :)