Story
arxiv_llm_reliability · Nov 11, 2026 · paper
Source brief
ShadowBench: Exposing Lexical Anchoring and the Illusion of Forgetting in Large Language Models
arxiv.orgNov 11, 2026
original source linked
In brief
Large Language Models (LLMs) have emerged as primary interfaces for factual retrieval, yet current evaluation paradigms rely almost exclusively on explicit entity names. In this work, we demonstrate that LLM knowledge...
Feed lens
evaluation