cs.CL · 2605.26827 Copy arXiv ID · May 26, 2026 Save ContextGuard: Structured Self-Auditing for Context Learning in Language Models Authors: Hongbo Jin , Chi Wang , Haoran Tang , Zhongjing Du , Xu Jiang , Jingqi Tian , Qiaoman Zhang , Jiayu Ding
Organizations: Peking University · SCUT · Tsinghua University
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.
Explore similar work May 25, 2026 · Hongbo Jin, Mingnan Zhu, Jingqi Tian +6 Large Language Model Reasoning In-Context Learning
Sep 1, 2026 · Jinhu Qi, Minda Hu, Wentao Zhang +4 In-Context Learning
Jul 9, 2026 · Jike Zhong, Ming Li, Yuxiang Lai +8 In-Context Learning
May 25, 2026 · cs.AI J/K move · Enter open · S save
Hongbo Jin, Mingnan Zhu, Jingqi Tian, Xu Jiang +5
Peking University · Xiamen University · Tsinghua University
While LLMs excel at reasoning over prompts using static pretrained knowledge, they struggle significantly with context learning-the ability to dynamically extract, internalize, and apply new knowledge from complex, task-specific contexts. Recent evaluations on the CL-Bench reveal a critical capability gap: frontier models solve only 17.2% of context-dependent tasks on average.