Emotion geometry in LLM activations

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

• Emotion Features in LLMsconcept0.830
  Internal representations encoding emotion concepts in large language models, identified by probing and SAE methods
• Are LLM emotion states encoded only selectively in token positions where they are operative, or in a more complex non-linear manner?question0.774
  Question raised by Anthropic and partially addressed by this paper's persistence evidence
• Activation-Behavior Manifold Isometryconcept0.755
• Mechanism by which activation of an emotion feature sometimes leads to later suppression of that same featurequestion0.748
  Identified research gap: the paper observes anti-persistence but has no explanation for it
• The representation-based path and the behavior-based path in Llama-3.1 8B activation space trace out similar curves, demonstrating bidirectional geometry alignment.finding0.747
  Key empirical result showing that optimizing for behavioral outputs and fitting representation geometry produce the same path in activation space.
• geometry of activation spaceconcept0.743
  Rich geometric structure carried by neural representations.
• Linear Representation of Concepts in LLMsconcept0.742
  The finding that interpretable concepts including character traits are encoded as linear directions in transformer residual streams
• Representation geometry causally shapes behavior; activation and behavior manifolds are approximately isometric.claim0.738