concept
active
concept:persona-vectorPersona Vector
A linear direction in a language model's activation space that encodes a specific personality trait, enabling monitoring and control of that trait
Neighborhood — ranked by edge-count
Papers (4)
paper
Concepts (1)
concept
- Persona Vector Arithmeticrelated_toCompositional combination of persona vectors via algebraic operations on activation directions
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Prior framework for monitoring and controlling character traits in LLMs via activation directions; this paper extends it to 275 roles
- The paper's reframing: persona vectors read as signals of internal model organization rather than as behavioral controls
- Stable, reidentifiable dispositional profiles that can be interpreted as characters with broadly coherent beliefs, values and traits
- Core method for extracting persona vectors by contrasting mean activations under persona-eliciting vs. suppressing prompts
- Specific attention layer where persona representations undergo sharp directional shift and stabilize, identified via cosine similarity heatmaps
- The paper's core automated pipeline that takes a trait name and description as input and outputs a corresponding persona vector via contrastive prompting
- An SAE latent representing a simulated character with consistent traits, as opposed to a specific trait
- Procedure for extracting a persona vector from a fine-tuned model variant and injecting it into the unmodified base to recover intractable directions