arXiv 2509.04504
Behavioral Fingerprinting of Large Language Models
By Zehua Pei, Hui-Ling Zhen, et al.
Published 2025-09-02
Discussion
Read the public discussion and references gathered around this paper.
Current benchmarks for Large Language Models (LLMs) primarily focus on performance metrics, often failing to capture the nuanced behavioral characteristics that differentiate them. This paper introduces a novel ``Behavioral Fingerprinting'' framework designed to move beyond traditional evaluation by creating a multi-faceted profile of a model's intrinsic cognitive and interactive styles. Using a curated Diagnostic P…