claim
active
claim:emotion-concept-representations-causally-influence-the-llm-s-outputs-including-preferences-and-misaligned-behaviorsEmotion concept representations causally influence the LLM's outputs including preferences and misaligned behaviors
Key finding.
Source paper
extracted_from(2026) · Nicholas Sofroniew · Isaac Kauvar · William Saunders · Runjin Chen +12
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (8)
finding
- Causal role of desperation.
- Positive valence probe predicts preference.
- Negative probe predicts dispreference.
- Sycophancy-harshness tradeoff.
- Valence mediates preference.
- Causal effect on preference.
- Emotions correlating with preference causally drive it.
- Causal effect on preference.
Methods (1)
method
- Activation SteeringsupportsCausal intervention technique: edit NLA explanation, reconstruct via AR, use difference as steering vector to manipulate model behavior.
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.
- Core definition.
- If LLM emotion representations resemble human valence-arousal space, behavior and activations will follow distances in that spacehypothesis0.821Prediction from circumplex theory.
- We hypothesize that persistently active emotional state representations exist in LLMs but may be missed by standard probing methods.hypothesis0.820Open hypothesis from the Anthropic paper that motivates this work
- Question raised by Anthropic and partially addressed by this paper's persistence evidence
- Interpretive hypothesis offered to explain why emotion features are more persistent
- Central interpretive claim of the paper supported by multiple convergent analyses
- Internal representations encoding emotion concepts in large language models, identified by probing and SAE methods
- Illustrates how the same empirical finding is read differently by computational functionalists and organismic functionalists.