Metacognition in LLMs: Foundations, Progress, and Opportunities — Yale Survey

Yale NLP Lab

Research official 2 src. ~1 min

The first comprehensive survey of metacognitive capabilities in large language models, covering measurement methods, benchmarks for self-monitoring and calibration, enhancement techniques, and open research questions. Reviews how models assess their own uncertainty, detect knowledge gaps, and regulate problem-solving strategies, mapping out where current approaches fall short compared to human metacognition. Resources released on GitHub.

Why it matters

Metacognition—knowing what you don't know—is foundational for reliable autonomous agents; this survey consolidates a scattered literature and identifies specific deficits (poor self-correction under distribution shift) relevant for deployment safety.

Importance: 2/5

First comprehensive LLM metacognition survey from Yale NLP; consolidates calibration and self-monitoring literature with implications for agent reliability

Sources