Knowledge editing, a field focused on modifying large language models (LLMs) without extensive retraining, is vital for correcting inaccuracies and biases. While early efforts centered on isolated facts, real-world information is often unstructured. We classify advancements beyond structured facts to include unstructured, common-sense, and event-level knowledge. We explore the expansion of knowledge editing applications into complex reasoning, medical domains, and code generation, highlighting both utility and limitations. We also point out future research directions, emphasizing the need for improved robustness, broader task generalization, and deeper integration of edited knowledge for complex reasoning. We specifically address the application of knowledge editing within intricate real-world settings, a contrast to existing surveys.