finding
active
finding:assistant-colon-probe-values-predict-response-emotion-better-than-user-turn-r-0-87-vs-0-59Assistant-colon probe values predict response emotion better than user-turn (r=0.87 vs 0.59)
Colon token carries prepared emotion.
Source paper
extracted_from(2026) · Nicholas Sofroniew · Isaac Kauvar · William Saunders · Runjin Chen +12
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Interpretation of Part 2.
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.
- User and Assistant emotions dissociate.
- Demonstrates that Cogito emotion probes are persistently active beyond what is explained by their variance alone
- Steering effect size across 35 emotion vectors correlates with probe-preference correlation at r=0.85finding0.777Emotions correlating with preference causally drive it.
- Emotion probe persistence correlation of 0.214 in Cogito v2.1 vs 0.099 for random vectorsfinding0.772Quantifies emotion feature persistence above random baseline in Cogito across 240 multi-turn conversations
- Shows model persona position is primarily determined by the most recent user message, not prior drift
- Quantitative measure of emotion feature persistence vs random baseline in Cogito
- Geometry mirrors human psychology.
- Negative probe predicts dispreference.