Recent research investigates how large language models (LLMs) learn in-context, revealing that they utilize both local transition copying and global structure inference. The findings suggest that LLMs operate using a dual-mechanism approach, which could influence future AI model development and understanding.