New EVADE-Bench benchmark highlights LLM struggles with evasive e-commerce content

A new benchmark called EVADE-Bench has been developed to evaluate the ability of Large Language Models (LLMs) and Vision Language Models (VLMs) to detect evasive content in e-commerce. The benchmark, which includes expert-curated Chinese multimodal data, revealed that even state-of-the-art models struggle with detecting deliberately obfuscated product information. The research also indicated that clearer rule categorization improves model consistency and that a multi-agent approach, separating visual description and logical inference, can enhance accuracy. AI

IMPACT Highlights the need for improved AI robustness in detecting sophisticated policy violations in e-commerce.

RANK_REASON The cluster describes a new academic benchmark and research paper evaluating AI models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Ancheng Xu
• e-commerce
• EVADE-Bench
• LLMs

• paper
• safety
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →