Control Vectors (CVs)

Attribute-aligned shifts added to the residual stream to steer LLM generation toward desired personality traits

Neighborhood — ranked by edge-count

paper

framework

Contrastive Sparse AutoEncoder (CV-SAE) Framework
implements
The primary novel framework introduced in the paper for learning facet-level personality control vectors

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.

steering vectorsconcept0.785
A method for modifying model behavior by adding perturbation vectors to activations, used here to try to reduce eval awareness.
Sense Vectorsconcept0.785
Vectors acquired during pretraining in Backpack LMs that have a multiplication effect on model generation
Persona Vectors (Chen et al.)framework0.763
Prior framework for monitoring and controlling character traits in LLMs via activation directions; this paper extends it to 275 roles
Control And Perception Loopframework0.732
Composite Steering Vector (v_role-truth)concept0.727
Steering vector extracted in Experiment 2 capturing latent representation of desired role behavior and honesty semantics
concept vector computationmethod0.727
Procedure extracting concept vectors as difference of mean activations between concept-exemplifying and baseline/negative sentences
concept vectorconcept0.724
Computed directional vector in activation space representing a specific concept, used for injection experiments
Contrastive Steering Vector Constructionmethod0.723
Method for computing steering vectors as mean activation differences between reflection levels at a given layer.