Context-CoT: Enhancing Context Learning via High-Quality Reasoning Synthesis
2026-05-25 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors explain that large language models (LLMs) are good at using what they learned before to answer questions but have a hard time learning new information from the specific context of a task. They tested models on a set of tasks called CL-Bench, which require understanding and using new context. The results show that even the best models can only complete about 17% of these tasks correctly. This reveals a big gap in the ability of LLMs to learn from and use new information dynamically.
Large Language ModelsContext LearningPretrained KnowledgeTask-Specific ContextCL-BenchDynamic Knowledge ExtractionReasoningModel Evaluation
Authors
Hongbo Jin, Mingnan Zhu, Jingqi Tian, Xu Jiang, Zhongjing Du, Haoran Tang, Siyi Xie, Qiaoman Zhang, Jiayu Ding
Abstract
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.