This paper presents a comprehensive review of metacognition in large language models. It surveys how metacognitive abilities are defined, measured, and benchmarked, as well as methods for eliciting, improving, and applying them. The review connects metacognition with learning, problem solving, decision-making, communication, reliability, transparency, and broader AI capability. It also summarizes current findings, applications, open questions, and research opportunities. The authors provide an organized paper list through the Yale-NLP GitHub repository, making the work useful as an entry point into this emerging research area.
No heat snapshots are available in the last 24 hours.