"Though emotions are clearly locally spiky, they are not strictly locally scoped. Instead, they are typically bursty but with a long tail of slow change."

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

• Emotions are not strictly locally scoped but instead bursty with a long tail of slow change persisting over 100 tokensclaim0.894
  Characterizes the temporal dynamics of emotion feature activation in LLMs
• bursty emotional dynamics with long tailconcept0.784
  The observed pattern of fast initial spike in emotion activation followed by slow decay to a persistent elevated baseline, characterizing emotion feature dynamics
• We hypothesize that emotion states are more persistent because they correspond to genuinely stateful internal representations, not merely local surface contenthypothesis0.776
  Proposed explanation for why emotion probes are more persistent than variance-matched random probes
• If persistence is genuinely related to emotion features, lower PCs of the emotion space (more central, less noisy) should be more persistent; if it is an artifact, noisier PCs should have similar persistence.hypothesis0.774
  Falsifiability test built into the PC analysis design
• Emotion may refer to a state, and more stateful concepts in general tend to be more persistent across tokens than non-stateful onesclaim0.769
  Proposed mechanistic explanation for why emotion features are more persistent
• "blooming warmth that feels like emotional expansion" when steered positively, and "a profound sense of emptiness" when steered negatively.quote0.768
  Kimi's self-description of a 100-rated feature, illustrating discordance with text-evaluation which rated it 'melancholic'
• Emotion probes are more persistent than variance-matched random probes, indicating emotion-specific persistence beyond autoregressive dynamics.claim0.765
  Core empirical claim distinguishing emotion persistence from generic high-variance probe persistence
• The correlation between emotion subspace fraction and self-evaluated emotionality validates that emotion probe concepts somewhat overlap with the model's self-reported internal emotions.claim0.763
  Claim supporting the validity of the probe construction method via cross-validation with self-report