Owain Evans

Truthful AI; UC Berkeley

Self-knowledge, introspection training, emergent misalignment, subliminal learning.

Placed by the editor, v1 — corrections welcome.

In the Atlas

• Chen et al 2025 - Persona vectors
• Betley et al 2025 - Tell me about yourself
• Betley et al 2025 - Emergent misalignment
• Binder et al 2025 - Looking inward
• Chua Betley Marks Evans 2026 - The consciousness cluster
• Cloud et al 2025 - Subliminal learning

Closest to

• Which activations belong to the Assistant
• Writing about a character or becoming it
• Coherence across personas
• How well can LLMs introspect
• An agenda beyond the Assistant
• When self-report counts as welfare evidence
• What spreads between agents