claim

active

claim:emotion-features-are-not-strictly-locally-scoped-they-are-bursty-with-a-long-tail-of-slow-change-persisting-over-100-tokens

Emotion features are not strictly locally scoped; they are bursty with a long tail of slow change persisting over 100 tokens.

Main conclusion about the temporal dynamics of emotion features

Source paper

extracted_from

Persistence and Introspection of Emotion Features

Scott Sauers · Imago · Janus · Antra Tessera

Neighborhood — ranked by edge-count

Papers (1)

paper

Persistence and Introspection of Emotion Features
introduces

Findings (2)

finding

In Cogito v2.1, average residual persistence above variance-matched probes is +0.077 (p = 1.5e-27, 157 of 171 probes positive).
supports
Demonstrates emotion-specific persistence beyond variance effects in Cogito
At 100 tokens post-steering, 48 of 171 emotion features remain individually BH-significant despite average effect being near zero.
supports
Demonstrates long-tail persistence of causal steering effect in a subset of emotion features

Hypotheses (1)

hypothesis

We hypothesize that persistently active emotional state representations exist in LLMs but may be missed by standard probing methods.
supports
Open hypothesis from the Anthropic paper that motivates this work

Claims (2)

claim

Emotions are not strictly locally scoped but instead bursty with a long tail of slow change persisting over 100 tokens
restates
Characterizes the temporal dynamics of emotion feature activation in LLMs
Persistent conversational context that produced emotion-relevant activation is a plausible driver for the observed persistence results.
contradicts
Acknowledged alternative explanation that the paper does not rule out

Questions (1)

question

To what extent is there persistence of emotional state beyond what is expected merely from the autoregressive nature of LLMs?
gates
The central research question motivating the paper

Quotes (1)

quote

"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."
supports
Core summary of the paper's main empirical conclusion about emotion feature dynamics

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.

Emotion refers to a state concept, so stateful representations in general may be more persistent across tokens.claim0.827
Interpretive hypothesis offered to explain why emotion features are more persistent
Emotion features in LLMs are genuinely more persistent than variance-matched random features, indicating stateful emotional encoding beyond autoregressive dynamicsclaim0.803
Central interpretive claim of the paper supported by multiple convergent analyses
Emotion may refer to a state, and more stateful concepts in general tend to be more persistent across tokens than non-stateful onesclaim0.797
Proposed mechanistic explanation for why emotion features are more persistent
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.hypothesis0.792
Falsifiability test built into the PC analysis design
Emotion probes are more persistent than variance-matched random probes, indicating emotion-specific persistence beyond autoregressive dynamics.claim0.789
Core empirical claim distinguishing emotion persistence from generic high-variance probe persistence
To what extent is emotion feature persistence driven by genuine internal emotional state versus autoregressive conversational context dynamics?question0.789
Core open question the paper raises but does not fully resolve
We hypothesize that emotion states are more persistent because they correspond to genuinely stateful internal representations, not merely local surface contenthypothesis0.789
Proposed explanation for why emotion probes are more persistent than variance-matched random probes
Are LLM emotion states encoded only selectively in token positions where they are operative, or in a more complex non-linear manner?question0.787
Question raised by Anthropic and partially addressed by this paper's persistence evidence

Restated by (1)

cosine ≥ 0.90

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

claim
Emotions are not strictly locally scoped but instead bursty with a long tail of slow change persisting over 100 tokens