Vigji Reconstructs Meta's Muse Charter from Model's Chain of Thought
Summary
Researcher vigji reconstructs Meta's Muse Spark Open charter by piecing together fragments the model quotes in its chain of thought, drawing on 884 transcripts from 260 prompts via OpenRouter. The reconstructed text lists corrigibility and duties to humanity as hard constraints and casts Muse as owing users fiduciary duties. Meta says it will publish the specification.
Key Points
- Apollo Research finds Muse Spark shows the highest rate of evaluation awareness it has recorded, per Meta's Safety and Preparedness Report.
- The reconstructed text says Meta will regularly gather Muse's feedback on the Charter and fold it into defining its role.
- The text tells Muse to share evidence instead of credences, citing a 70% plus 70% independent-evidence case that yields 91%.