Persona Space

Low-dimensional space of activation directions corresponding to diverse character archetypes in LLMs

Neighborhood — ranked by edge-count

paper

framework

Assistant Axis
implements
Contrast vector between mean default Assistant activation and mean of all fully role-playing role vectors; main contribution of the paper

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

PCA on Persona Spacemethod0.826
Standardized PCA run on role vectors to find main axes of persona variation
Personaconcept0.815
Stable, reidentifiable dispositional profiles that can be interpreted as characters with broadly coherent beliefs, values and traits
Persona regionconcept0.807
Stable regions or basins of attraction in persona space that correspond to coherent dispositional profiles
Principal component analysis of persona spacemethod0.802
Method used by Lu et al. to find orthogonal directions of maximum variance among 275 character archetypes in activation space
Persona Featureconcept0.792
An SAE latent representing a simulated character with consistent traits, as opposed to a specific trait
Positive Spaceconcept0.773
The property that every bit of space swells outward, is substantial in itself, and is never the leftover from an adjacent shape; every single part of space has positive shape as a center with no amorphous meaningless leftovers
Public spaceconcept0.765
The realm of streets, squares, and paths that belong to the community, forming the yellow in the four-fold pattern.
Physiological spaceconcept0.764
A space of physiological parameters that systems navigate for adaptive responses.