Confidence is Not Competence

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Sanyal, Debdeep, Pandey, Manya, Kumar, Dhruv, Deshpande, Saurabh, Mandal, Murari
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908618355376128
author Sanyal, Debdeep
Pandey, Manya
Kumar, Dhruv
Deshpande, Saurabh
Mandal, Murari
author_facet Sanyal, Debdeep
Pandey, Manya
Kumar, Dhruv
Deshpande, Saurabh
Mandal, Murari
contents Large language models (LLMs) often exhibit a puzzling disconnect between their asserted confidence and actual problem-solving competence. We offer a mechanistic account of this decoupling by analyzing the geometry of internal states across two phases - pre-generative assessment and solution execution. A simple linear probe decodes the internal "solvability belief" of a model, revealing a well-ordered belief axis that generalizes across model families and across math, code, planning, and logic tasks. Yet, the geometries diverge - although belief is linearly decodable, the assessment manifold has high linear effective dimensionality as measured from the principal components, while the subsequent reasoning trace evolves on a much lower-dimensional manifold. This sharp reduction in geometric complexity from thought to action mechanistically explains the confidence-competence gap. Causal interventions that steer representations along the belief axis leave final solutions unchanged, indicating that linear nudges in the complex assessment space do not control the constrained dynamics of execution. We thus uncover a two-system architecture - a geometrically complex assessor feeding a geometrically simple executor. These results challenge the assumption that decodable beliefs are actionable levers, instead arguing for interventions that target the procedural dynamics of execution rather than the high-level geometry of assessment.
format Preprint
id arxiv_https___arxiv_org_abs_2510_24772
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Confidence is Not Competence
Sanyal, Debdeep
Pandey, Manya
Kumar, Dhruv
Deshpande, Saurabh
Mandal, Murari
Computation and Language
Artificial Intelligence
Large language models (LLMs) often exhibit a puzzling disconnect between their asserted confidence and actual problem-solving competence. We offer a mechanistic account of this decoupling by analyzing the geometry of internal states across two phases - pre-generative assessment and solution execution. A simple linear probe decodes the internal "solvability belief" of a model, revealing a well-ordered belief axis that generalizes across model families and across math, code, planning, and logic tasks. Yet, the geometries diverge - although belief is linearly decodable, the assessment manifold has high linear effective dimensionality as measured from the principal components, while the subsequent reasoning trace evolves on a much lower-dimensional manifold. This sharp reduction in geometric complexity from thought to action mechanistically explains the confidence-competence gap. Causal interventions that steer representations along the belief axis leave final solutions unchanged, indicating that linear nudges in the complex assessment space do not control the constrained dynamics of execution. We thus uncover a two-system architecture - a geometrically complex assessor feeding a geometrically simple executor. These results challenge the assumption that decodable beliefs are actionable levers, instead arguing for interventions that target the procedural dynamics of execution rather than the high-level geometry of assessment.
title Confidence is Not Competence
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2510.24772