Anthropic (interpretability)
Model introspection, emotion and workspace representations, interpretability of the Assistant character.
Placed by the editor, v1 — corrections welcome.
In the Atlas
- Sofroniew et al 2026 - Emotion concepts
- Lu et al 2026 - The Assistant Axis
- Chen et al 2025 - Persona vectors
- Lindsey 2025 - Emergent introspective awareness
- Marks Lindsey Olah 2026 - The persona selection model
- Gurnee Sofroniew et al 2026 - A global workspace in language models
- Marks et al 2025 - Auditing for hidden objectives