finding

active

finding:lower-more-central-emotion-pcs-are-more-persistent-than-higher-noisier-pcs-in-both-kimi-and-cogito

Lower (more central) emotion PCs are more persistent than higher (noisier) PCs in both Kimi and Cogito

Rules out that persistence is an artifact of probe construction, since noise dimensions are not similarly persistent

Source paper

extracted_from

Persistence and Introspection of Emotion Features

Scott Sauers · Imago · Janus · Antra Tessera

Neighborhood — ranked by edge-count

Claims (1)

claim

Persistence is not an artifact of probe construction because lower (more central) emotion PCs are more persistent than noisier high-rank PCs
supports
Rules out measurement artifact explanation for the persistence finding

Methods (1)

method

PCA of Emotion Feature Activations
introduces
PCA on 171 emotion probe activations across all tokens to produce ordered linear combinations and test if lower PCs are more persistent

Findings (1)

finding

Lower (more central) PCs of emotion feature activations are more persistent than higher-rank (noisier) PCs in both Kimi and Cogito, above variance-matched baselines.
restates
Supports that persistence is genuinely tied to emotion structure rather than measurement artifact

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

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

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.hypothesis0.854
Falsifiability test built into the PC analysis design
Emotion probe persistence correlation of 0.214 in Cogito v2.1 vs 0.099 for random vectorsfinding0.790
Quantifies emotion feature persistence above random baseline in Cogito across 240 multi-turn conversations
PCs of the emotion space and persistenceconcept0.775
Analysis showing that lower-rank (more central) PCs of emotion feature activations are more persistent than higher-rank (noisier) PCs
Emotion probe persistence (token-0 to token-100 correlation) in Cogito v2.1 is 0.214, compared to 0.099 for random unit vectors in 7168D space.finding0.772
Quantitative measure of emotion feature persistence vs random baseline in Cogito
SAE feature emotion subspace overlap correlates with persistence in Cogito: Spearman +0.413, p=4.4e-196finding0.763
Demonstrates that SAE features more aligned with the emotion subspace are more persistent in Cogito after variance control
Cogito emotion probe residual autocorrelation +0.077 above variance-matched controls (p=1.5e-27, 157/171 probes positive)finding0.763
Demonstrates that Cogito emotion probes are persistently active beyond what is explained by their variance alone
SAE emotion subspace overlap correlates with variance-residualized persistence in Cogito: Spearman +0.413, p = 4.4e-196.finding0.752
Strong positive relationship between emotion alignment and SAE feature persistence in Cogito
Emotion probes are more persistent than variance-matched random probes, indicating emotion-specific persistence beyond autoregressive dynamics.claim0.744
Core empirical claim distinguishing emotion persistence from generic high-variance probe persistence

Restated by (1)

cosine ≥ 0.90

Other entities that say roughly the same thing. May be merge candidates or independent restatements across papers.

finding
Lower (more central) PCs of emotion feature activations are more persistent than higher-rank (noisier) PCs in both Kimi and Cogito, above variance-matched baselines.