AI RESEARCH
Context-CoT: Enhancing Context Learning via High-Quality Reasoning Synthesis
arXiv CS.AI
•
ArXi:2605.25354v1 Announce Type: new 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.