What question did this study set out to answer?

This research aims to explore the concept of emotional gravity in semantic chains and its stability across different model parameters and architectures.

May 21, 2026Open Access

Emotional Gravity and Attractor Topology in LLM Semantic Chains: An Exploratory Multi-Model Study

Key Points

This research aims to explore the concept of emotional gravity in semantic chains and its stability across different model parameters and architectures.
Conducted empirical experiments using the NeuroState framework to examine emotional gravity.
Identified five attractor types in the affective semantic space through various model architectures.
Analyzed the impact of affective state modulations on emotional domain associations.
Determined that emotional gravity remains stable across various sampling parameters and model architectures.
Identified a three-layer control hierarchy influencing emotional gravity: training, NeuroState, and sampling.
Found that '消滅' triggers safety filters more reliably than '死', indicating a nuanced representation of existential threat.

Abstract

We report empirical experiments examining whether "emotional gravity" — the tendency of association chains to remain within or escape from their initial emotional domain — is stable across sampling parameters, model architectures, and affective state modulations. Using the NeuroState framework as a probe, we identify five attractor types in affective semantic space and report a three-layer control hierarchy (training → topology, NeuroState → geology, sampling → physics). We further report that "消滅" (annihilation) triggers Claude's safety filter more reliably than "死" (death), consistent with a graded existential threat representation inherited from training data.

Emotional Gravity and Attractor Topology in LLM Semantic Chains: An Exploratory Multi-Model Study

Key Points

Abstract

Cite This Study

Also Consider

Also Consider