Rubi J. Hudson

Defin­ing Mon­i­torable and Use­ful Goals

Defin­ing Cor­rigible and Use­ful Goals

Safe Pre­dic­tive Agents with Joint Scor­ing Rules

Sim­plify­ing Cor­rigi­bil­ity – Subagent Cor­rigi­bil­ity Is Not Anti-Natural

A Ba­sic Eco­nomics-Style Model of AI Ex­is­ten­tial Risk

The Case for Pre­dic­tive Models

Search­ing for Search­ing for Search

Con­di­tional Pre­dic­tion with Zero-Sum Train­ing Solves Self-Fulfilling Prophecies

Con­di­tion­ing Pre­dic­tive Models: Open prob­lems, Con­clu­sion, and Appendix

Mechanism De­sign for AI Safety—Agenda Creation Retreat

Con­di­tion­ing Pre­dic­tive Models: De­ploy­ment strategy

Con­di­tion­ing Pre­dic­tive Models: In­ter­ac­tions with other approaches

Con­di­tion­ing Pre­dic­tive Models: Mak­ing in­ner al­ign­ment as easy as possible

Con­di­tion­ing Pre­dic­tive Models: The case for competitiveness

Con­di­tion­ing Pre­dic­tive Models: Outer al­ign­ment via care­ful conditioning

Con­di­tion­ing Pre­dic­tive Models: Large lan­guage mod­els as predictors

Stop-gra­di­ents lead to fixed point predictions

Un­der­speci­fi­ca­tion of Or­a­cle AI

Proper scor­ing rules don’t guaran­tee pre­dict­ing fixed points

Mechanism De­sign for AI Safety—Read­ing Group Curriculum