Unveiling Safety Vulnerabilities of Large Language Models

TL;DR

This paper presents a new dataset and method for identifying safety vulnerabilities in large language models by analyzing their responses to adversarial questions and automatically detecting vulnerable semantic regions.

Contribution

It introduces AttaQ, a dataset of adversarial questions, and a novel clustering-based approach to identify semantic regions prone to harmful outputs in language models.

Findings

Models show significant vulnerabilities to adversarial questions.

The automatic method effectively identifies harmful semantic regions.

Enhances targeted safety improvements for language models.

Abstract

As large language models become more prevalent, their possible harmful or inappropriate responses are a cause for concern. This paper introduces a unique dataset containing adversarial examples in the form of questions, which we call AttaQ, designed to provoke such harmful or inappropriate responses. We assess the efficacy of our dataset by analyzing the vulnerabilities of various models when subjected to it. Additionally, we introduce a novel automatic approach for identifying and naming vulnerable semantic regions - input semantic areas for which the model is likely to produce harmful outputs. This is achieved through the application of specialized clustering techniques that consider both the semantic similarity of the input attacks and the harmfulness of the model's responses. Automatically identifying vulnerable semantic regions enhances the evaluation of model weaknesses, facilitating targeted improvements to its safety mechanisms and overall reliability.