(Update Dec 2025: This series has been almost entirely superseded by a newer and more polished series, available here.)

This is a four-part series of posts about how we might solve the alignment problem. It also builds off of my previous post, here, about what it would even be to solve the alignment problem; and to some extent, off of this post outlining a framework for thinking the incentives at stake in AI power-seeking.