Hypothesis Subspace

A living collection of alignment proposals I’m exploring at Refine, a program hosted by Conjecture.

Over­sight Leagues: The Train­ing Game as a Feature

Ide­olog­i­cal In­fer­ence Eng­ines: Mak­ing Deon­tol­ogy Differ­en­tiable*

Rep­re­sen­ta­tional Tethers: Ty­ing AI La­tents To Hu­man Ones

In­ter­lude: But Who Op­ti­mizes The Op­ti­mizer?

(Struc­tural) Sta­bil­ity of Cou­pled Optimizers

Boolean Prim­i­tives for Cou­pled Optimizers

Cat­a­logu­ing Pri­ors in The­ory and Practice