What if, instead of assuming we had to build an AI singleton as fast as possible (with the risk that it turns out to be bad,) we just... didn't?
https://worldspiritsockpuppet.substack.com/p/lets-think-about-slowing-down-ai
18 or so reasons to reflexively dismiss slowing down AI and why I think they fail and leave it worth thinking seriously about.
a Schelling point for those who seek one