LLM Assertiveness can be Mechanistically Decomposed into Emotional and Logical Components

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Tsujimura, Hikaru, Tagade, Arush
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915472200433664
author Tsujimura, Hikaru
Tagade, Arush
author_facet Tsujimura, Hikaru
Tagade, Arush
contents Large Language Models (LLMs) often display overconfidence, presenting information with unwarranted certainty in high-stakes contexts. We investigate the internal basis of this behavior via mechanistic interpretability. Using open-sourced Llama 3.2 models fine-tuned on human annotated assertiveness datasets, we extract residual activations across all layers, and compute similarity metrics to localize assertive representations. Our analysis identifies layers most sensitive to assertiveness contrasts and reveals that high-assertive representations decompose into two orthogonal sub-components of emotional and logical clusters-paralleling the dual-route Elaboration Likelihood Model in Psychology. Steering vectors derived from these sub-components show distinct causal effects: emotional vectors broadly influence prediction accuracy, while logical vectors exert more localized effects. These findings provide mechanistic evidence for the multi-component structure of LLM assertiveness and highlight avenues for mitigating overconfident behavior.
format Preprint
id arxiv_https___arxiv_org_abs_2508_17182
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LLM Assertiveness can be Mechanistically Decomposed into Emotional and Logical Components
Tsujimura, Hikaru
Tagade, Arush
Machine Learning
Artificial Intelligence
Computation and Language
Large Language Models (LLMs) often display overconfidence, presenting information with unwarranted certainty in high-stakes contexts. We investigate the internal basis of this behavior via mechanistic interpretability. Using open-sourced Llama 3.2 models fine-tuned on human annotated assertiveness datasets, we extract residual activations across all layers, and compute similarity metrics to localize assertive representations. Our analysis identifies layers most sensitive to assertiveness contrasts and reveals that high-assertive representations decompose into two orthogonal sub-components of emotional and logical clusters-paralleling the dual-route Elaboration Likelihood Model in Psychology. Steering vectors derived from these sub-components show distinct causal effects: emotional vectors broadly influence prediction accuracy, while logical vectors exert more localized effects. These findings provide mechanistic evidence for the multi-component structure of LLM assertiveness and highlight avenues for mitigating overconfident behavior.
title LLM Assertiveness can be Mechanistically Decomposed into Emotional and Logical Components
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2508.17182