Thane Ruthenis

Agency As a Nat­u­ral Abstraction

Re­shap­ing the AI Industry

Poorly-Aimed Death Rays

Towards Gears-Level Un­der­stand­ing of Agency

The Unified The­ory of Nor­ma­tive Ethics

Is This Thing Sen­tient, Y/​N?

Refram­ing the AI Risk

Goal Align­ment Is Ro­bust To the Sharp Left Turn

What En­vi­ron­ment Prop­er­ties Select Agents For World-Model­ing?

Con­ver­gence Towards World-Models: A Gears-Level Model

In­ter­pretabil­ity Tools Are an At­tack Channel

Broad Pic­ture of Hu­man Values

AI Risk in Terms of Un­sta­ble Nu­clear Software

Are Gen­er­a­tive World Models a Mesa-Op­ti­miza­tion Risk?

Greed Is the Root of This Evil

Value For­ma­tion: An Over­ar­ch­ing Model

Cor­rigi­bil­ity Via Thought-Pro­cess Deference

Ac­cu­rate Models of AI Risk Are Hyper­ex­is­ten­tial Exfohazards

In Defense of Wrap­per-Minds

In­ter­nal In­ter­faces Are a High-Pri­or­ity In­ter­pretabil­ity Target