Tag
This paper introduces the FORGE benchmark to evaluate how web content polluted by generative engine optimization can mislead search-augmented LLM recommenders into promoting fake products, revealing significant vulnerabilities and ineffective defenses.
This paper introduces Sampled-BPE, a lightweight token-level auditing pipeline for web-scale Chinese corpora, and applies it to reveal widespread but uneven pollution across open Chinese datasets and Common Crawl snapshots. It also releases a hierarchical dataset of over 630k polluted token records with context and explanations.