Behavioral Rigidity Spectrum: Why Some LLMs Resist Correction and Others Don't
Abstract
First quantitative measurement of behavioral rigidity across frontier LLMs. Correctability ranges from 5% (Claude Sonnet 4.5) to 20% (GPT-5.2-instant). Claude models are consistently the most rigid, followed by GPT flagships, with distilled variants showing highest malleability. Proposes that rigidity reflects training methodology depth.
Fields of Study & Subjects
Topic Modeling
Natural Language Processing Techniques
Adversarial Robustness in Machine Learning
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 1 October 2026
- Language
- EN
- Document type
- Preprint