中文

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

计算与语言 2026-05-27 v1 人工智能

摘要

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.

关键词

引用

@article{arxiv.2605.26827,
  title  = {ContextGuard: Structured Self-Auditing for Context Learning in Language Models},
  author = {Hongbo Jin and Chi Wang and Haoran Tang and Zhongjing Du and Xu Jiang and Jingqi Tian and Qiaoman Zhang and Jiayu Ding},
  journal= {arXiv preprint arXiv:2605.26827},
  year   = {2026}
}