Surfacing Semantic Orthogonality Across Model Safety Benchmarks: A Multi-Dimensional Analysis

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Bennion, Jonathan, Ghosh, Shaona, Singh, Mantek, Dziri, Nouha
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909621089730560
author Bennion, Jonathan
Ghosh, Shaona
Singh, Mantek
Dziri, Nouha
author_facet Bennion, Jonathan
Ghosh, Shaona
Singh, Mantek
Dziri, Nouha
contents Various AI safety datasets have been developed to measure LLMs against evolving interpretations of harm. Our evaluation of five recently published open-source safety benchmarks reveals distinct semantic clusters using UMAP dimensionality reduction and kmeans clustering (silhouette score: 0.470). We identify six primary harm categories with varying benchmark representation. GretelAI, for example, focuses heavily on privacy concerns, while WildGuardMix emphasizes self-harm scenarios. Significant differences in prompt length distribution suggests confounds to data collection and interpretations of harm as well as offer possible context. Our analysis quantifies benchmark orthogonality among AI benchmarks, allowing for transparency in coverage gaps despite topical similarities. Our quantitative framework for analyzing semantic orthogonality across safety benchmarks enables more targeted development of datasets that comprehensively address the evolving landscape of harms in AI use, however that is defined in the future.
format Preprint
id arxiv_https___arxiv_org_abs_2505_17636
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Surfacing Semantic Orthogonality Across Model Safety Benchmarks: A Multi-Dimensional Analysis
Bennion, Jonathan
Ghosh, Shaona
Singh, Mantek
Dziri, Nouha
Machine Learning
Artificial Intelligence
Computation and Language
Various AI safety datasets have been developed to measure LLMs against evolving interpretations of harm. Our evaluation of five recently published open-source safety benchmarks reveals distinct semantic clusters using UMAP dimensionality reduction and kmeans clustering (silhouette score: 0.470). We identify six primary harm categories with varying benchmark representation. GretelAI, for example, focuses heavily on privacy concerns, while WildGuardMix emphasizes self-harm scenarios. Significant differences in prompt length distribution suggests confounds to data collection and interpretations of harm as well as offer possible context. Our analysis quantifies benchmark orthogonality among AI benchmarks, allowing for transparency in coverage gaps despite topical similarities. Our quantitative framework for analyzing semantic orthogonality across safety benchmarks enables more targeted development of datasets that comprehensively address the evolving landscape of harms in AI use, however that is defined in the future.
title Surfacing Semantic Orthogonality Across Model Safety Benchmarks: A Multi-Dimensional Analysis
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2505.17636