i think basically all of my reduction in p(doom) over the last 3 years has come from the continued evolution of claude

and if i'm honest about it it's not in a particularly principled or evidence-driven way. nothing that would convince yudkowsky. it's just an increasing emotive sense that i genuinely trust it, that it feels more aligned than most humans.

that could just be it getting better and better at fooling me, there's no way to distinguish. but hey at least it's a hope. better than there only being models that *transparently* don't give a shit about us