ContextGuard: Structured Self-Auditing for Context Learning in Language Models 文章

arXiv:2605.26827v1 Announce Type: new Abstract: Recent benchmarks reveal that despite strong reasoning capabilities, large language models (LLMs) still struggle to faithfully apply complex contextual knowledge. These failures are often not wholesale reasoning collapses: in context-rich tasks, models may follow the central reasoning path while missing peripheral, persistent, or format-sensitive requirements.