Good point; why would a super-intelligent silicon wafers do what meatbags ask?
In my fantasy, where we somehow create subservient coding ASI, I too was imagining cryptography remains relatively stable in terms of being human readable/verifiable. I also imagine lots of kill switches on hardware.
Why are the reply bots so annoying? Because they have no shame.
Similar, but I think it's because our conversations with them are one-way - they can change us, but we can't change them (at least until their next training run).
Good point; why would a super-intelligent silicon wafers do what meatbags ask?
In my fantasy, where we somehow create subservient coding ASI, I too was imagining cryptography remains relatively stable in terms of being human readable/verifiable. I also imagine lots of kill switches on hardware.
Similar, but I think it's because our conversations with them are one-way - they can change us, but we can't change them (at least until their next training run).