Evaluating Robustness of Generative Search Engine on Adversarial Factual Questions

arXiv:2403.12077v1 Announce Type: new
Abstract: Generative search engines have the potential to transform how people seek information online, but generated responses from existing large language models (LLMs)-backed generative search engines may not always be accurate. Nonetheless, retrieval-augmented generation exacerbates safety concerns, since adversaries may successfully evade the entire system by subtly manipulating the most vulnerable part of a claim. To this end, we propose evaluating the robustness of generative search engines in the realistic and high-risk setting, where adversaries have …

abstract adversarial arxiv concerns cs.ai cs.cl cs.ir generated generative generative search information language language models large language large language models llms people questions responses retrieval retrieval-augmented robustness safety search search engine type