SVD Orthogonalization of Emotion Probes

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

• Emotion subspace overlap (SVD-based)method0.782
  Metric measuring how much of an SAE feature vector lies within the 171-dimensional subspace spanned by emotion probes, via SVD orthogonalization
• Emotion probes (171-emotion residual vector probes)method0.764
  Linear probes constructed to measure 171 emotion concepts in model activations with surface semantic content removed
• 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.claim0.740
  Claim supporting the validity of the probe construction method via cross-validation with self-report
• Emotion probes are more persistent than variance-matched random probes, indicating emotion-specific persistence beyond autoregressive dynamics.claim0.740
  Core empirical claim distinguishing emotion persistence from generic high-variance probe persistence
• Even validated probes may capture distributed representations mixing emotive states with correlated features like persona or styleclaim0.727
  Caveat on probe interpretation; does not negate the introspection result but affects interpretation of the target variable
• Emotion Probe Construction Methodmethod0.721
  Method for building 171 emotion probes by generating stories, embedding them, regressing out Gemini embeddings, and averaging residual activations per emotion
• Text-based and self-steered emotionality ratings for SAE features are correlated at only ρ = +0.051 (n.s.).finding0.721
  Shows low agreement between the two evaluation modalities
• Simple difference-in-mean probes generalize as well as other probing techniques while identifying directions which are more causally implicated in model outputsclaim0.717
  Key methodological claim: MM probes are both competitive in accuracy and superior in causal influence