Import AI 462: Superpersuasion; self-sustaining AI; paths to ASI

Clark's framing quietly shifts the burden of proof: the interesting question is not whether ASI is coming but whether the people allocating capital toward it are reasoning from evidence or from prior conviction that is structurally immune to disconfirmation. That distinction has practical consequences for how safety research and capability research get funded relative to each other.

This piece is largely disconnected from recent activity in our archive, as we have no prior coverage to anchor it to. It belongs to a longer-running debate in AI epistemics, one that surfaces periodically when capability progress accelerates and forecasters struggle to separate genuine signal from motivated reasoning. The superpersuasion thread is worth noting separately: as inference costs fall and model outputs become more fluent, the question of AI-assisted influence at scale is moving from theoretical to operational, and that shift deserves its own tracking.

Watch whether any of the major labs publish explicit falsifiability criteria for their ASI timelines in the next twelve months. If none do, Clark's quasi-religious framing becomes harder to dismiss on empirical grounds.