04:00
2026-08-31
arxiv.org
large-language-models
Load-Bearing Context: The Question Damage Score for Evaluating Context Reliance in Linguistic Reasoning
A new diagnostic framework from a study on arXiv (2608.27756v1) introduces a Question Damage Score to evaluate how much large language models rely on context versus prior knowledge, using 53 UK Lingui…