concept
active
concept:emotion-vectorsEmotion vectors
Linear representations of emotion concepts extracted from activations.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (2)
method
- Logit LensaboutUnsupervised interpretability technique that projects activations through unembedding matrix; provides comparison point for NLA approach.
- Average residual activations across stories per emotion, subtract cross-emotion mean, project out neutral PCs.
Concepts (3)
concept
- Functional EmotionscitesSofroniew et al.'s causally functional emotion-concept representations discovered in Claude Sonnet 4.5.
- Vectors track operative emotion at a token position, not persistent state.
- Primary axes of emotion vector space.
Findings (1)
finding
- Vectors track semantic interpretation.
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.
- Representations of an emotion that is implied but not expressed.
- Sanity check, not surprising.
- Directions in activation space encoding a model's dispositional self-presentation, used as evidence for self-modelling.
- Interpretation of Part 2.
- Vectors acquired during pretraining in Backpack LMs that have a multiplication effect on model generation
- Linear probes constructed to measure 171 emotion concepts in model activations with surface semantic content removed
- Prior framework for monitoring and controlling character traits in LLMs via activation directions; this paper extends it to 275 roles
- Cited as activation-level support for the performing care vs having care distinction the battery detects behaviorally