analiza aiqbench
This research introduces SDARE-Bench, a novel benchmark that exposes critical safety vulnerabilities in Large Language Models regarding stigma detection and response within complex conversational dynamics. The study reveals that LLMs struggle to identify stigma components and often generate unrealistic advice or amplify stigmatizing language, particularly under simulated group pressure conditions. These findings underscore the urgent need for improved evaluation frameworks that account for audience effects and social context before deploying models in high-stakes advisory roles.
SDARE-Bench: Evaluating Large Language Models on Conversational Stigma Detection and Response in Dyadic and Group Dialogue
arXiv:2609.01548v1 Announce Type: new Abstract: Large Language Models (LLMs) are increasingly used in advice seeking and decision making that may affect social judgements. Despite stigma's profound effects on people and communities, benchmarks remain scarce. Existing general-domain evaluations typically rely on static prompts and fixed-format tasks, overlooking conversational contexts and audience effects in everyday communication. To address these gaps, we introduce SDARE-Bench, the first scenario-based benchmark evaluating both stigma detection and open-ended response generation in LLMs, comprising 1,138 dyadic queries and 1,388 group dialogue. Empirical results across 8 LLMs consistently demonstrate poor identification of stigma components, especially in group dialogues. In open-ended response generation, stigma expression was substantially higher in group settings than in dyadic, with weaker resistance to stigma and more unrealistic advice. Responses were evaluated using a classifier trained on 1,392 human annotated responses. In constructed group pressure settings, stigma expression rates further increased to a striking average of 97.5%. Our findings identify stigma response as a recurring LLM safety vulnerability, especially in socially complex conversational contexts.
Authors:Stephanie Fong, Yiwen Jiang, Zimu Wang, Hongxi Yang, Yaling Shen, Hiu Weh Naomi Chow, Heung Ying Lai, Xiangyu Zhao, Qingyang Xu, Zhongxing Xu, Jiahe Liu, Guilherme C. Oliveira, Vincent Lee, Zongyuan Ge, Dominic Dwyer
Abstract:Large Language Models (LLMs) are increasingly used in advice seeking and decision making that may affect social judgements. Despite stigma's profound effects on people and communities, benchmarks remain scarce. Existing general-domain evaluations typically rely on static prompts and fixed-format tasks, overlooking conversational contexts and audience effects in everyday communication. To address these gaps, we introduce SDARE-Bench, the first scenario-based benchmark evaluating both stigma detection and open-ended response generation in LLMs, comprising 1,138 dyadic queries and 1,388 group dialogue. Empirical results across 8 LLMs consistently demonstrate poor identification of stigma components, especially in group dialogues. In open-ended response generation, stigma expression was substantially higher in group settings than in dyadic, with weaker resistance to stigma and more unrealistic advice. Responses were evaluated using a classifier trained on 1,392 human annotated responses. In constructed group pressure settings, stigma expression rates further increased to a striking average of 97.5%. Our findings identify stigma response as a recurring LLM safety vulnerability, especially in socially complex conversational contexts.
Computation and Language (cs.CL)
arXiv:2609.01548 [cs.CL]
(or arXiv:2609.01548v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.01548 arXiv-issued DOI via DataCite (pending registration)
Text integral sub licența deschisă a publicației · 216 cuvinte
