Large Language Models (LLMs) have demonstrated remarkable proficiency in natural language processing applications, encompassing question answering, text generation, and reasoning capabilities. However, their metacognitive abilities, which involve self-assessment, uncertainty awareness, and cognitive control, remain insufficiently explored. This investigation examines over 97 publications released between 2021 and 2025. These studies encompass prompt-based methodologies, fine-tuning techniques, retrieval-augmented generation, and agentic AI frameworks that implement metacognition within lifelong learning models (LLMs). Metacognitive interventions produce measurable, task-dependent performance gains. Reported improvements range from approximately 3% to over 20% across evaluated benchmarks. These gains are observed in medical reasoning, multi-hop question answering, and natural language comprehension tasks.
These enhancements correlate with demonstrable improvements in confidence calibration, error detection, and the precision of response revisions. Furthermore, this work conducts a comprehensive examination of more than ten families of large language models and agentic frameworks, systematically identifying enduring challenges, including overconfidence, susceptibility to hallucinations, limited error awareness, and unstable self-reflection processes. This research furnishes an analytical basis for elucidating the conditions under which metacognitive mechanisms either enhance or diminish LLM reliability. Moreover, it delineates prospective research avenues aimed at developing scalable, trustworthy, and human-centered artificial intelligence systems. This foundation is established through the synthesis of evaluation protocols, benchmarking methodologies, and comparative evidence.
BrownT, et al.Language models are few-shot learners.Adv Neural Inf Process Syst2020; 33: 1877–1901.
2.
IqbalS. The intelligence spectrum: Unraveling the path from ani to asi.Journal of Computing & Biomedical Informatics2024; 7.
3.
NeisserU. Cognitive psychology: Classic edition. London: Taylor & Francis, 2014.
4.
AndersonJR. Cognitive psychology and its implications. New York: Worth Publishers (Macmillan), 2005.
5.
IqbalS. The intelligence spectrum: Unraveling the path from ani to asi.Journal of Computing & Biomedical Informatics2024; 7.
6.
PiagetJ. The origins of intelligence in children.. New York: International Universities Press, 1952.
7.
VygotskyLS. Mind in society: The development of higher psychological processes. Vol. 86. Cambridge, MA: Harvard university press, 1978.
8.
EvansJSBStanovichKE. Dual-process theories of higher cognition: advancing the debate.Perspect Psychol Sci2013; 8: 223–241.
9.
FlavellJH. Metacognition and cognitive monitoring: a new area of cognitive–developmental inquiry.Am Psychol1979; 34: 906.
10.
JiaXLiWCaoL. The role of metacognitive components in creative thinking.Front Psychol2019; 10: 2404.
11.
BasuSDixitS. Role of metacognition in explaining decision-making styles: a study of knowledge about cognition and regulation of cognition.Pers Individ Dif2022; 185: 111318.
12.
YeungNSummerfieldC. Metacognition in human decision-making: confidence and error monitoring.Philos Trans R Soc, B2012; 367: 1310–1321.
13.
JostJTKruglanskiAWNelsonTO. Social metacognition: an expansionist reviewMetacognition. London: Psychology Press (Taylor & Francis), 2013, pp.137–154.
14.
McKinneySM, et al.International evaluation of an AI system for breast cancer screening.Nature2020; 577: 89–94.
15.
BansalPKockelmanKM. Forecasting Americans’ long-term adoption of connected and autonomous vehicle technologies.Transp Res A Policy Pract2017; 95: 49–63.
16.
ShinnNCassanoFGopinathA, et al.Reflexion: language agents with verbal reinforcement learning. Adv Neural Inf Process Syst2023; 36: 8634–8652. Wei, Jason, et al. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems 35 (2022): 24824–24837.
17.
Doshi-VelezFKimB. Towards A Rigorous Science of Interpretable Machine Learning. arXiv: Machine Learning (2017): n. pag.
18.
AndersonJRLebiereCJ. The atomic components of thought. London: Psychology Press (Taylor & Francis), 2014.
19.
LairdJENewellARosenbloomPS. Soar: an architecture for general intelligence.Artif Intell1987; 33: 1–64.
20.
KahnemanD. Thinking, fast and slow. New York: Farrar, Straus and Giroux, 2011.
21.
VaswaniA. Attention is all you need. Adv Neural Inf Process Syst2017; 30: 5998–5608
22.
SchneiderJ. Improving Next Tokens via Second-to-Last Predictions with Generate and RefineInternational Symposium on Intelligent Data Analysis. Cham: Springer Nature Switzerland, 2025, pp.223–233.
23.
DevlinJ, et al.Bert: Pre-training of deep bidirectional transformers for language understandingProceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers). Minneapolis, MN: Association for Computational Linguistics (ACL), 2019, pp.4171–4186.
24.
HeP, et al.Deberta: Decoding-enhanced bert with disentangled attention. arXiv preprint arXiv:2006.03654. 2020.
25.
RaffelC, et al.Exploring the limits of transfer learning with a unified text-to-text transformer.J Mach Learn Res2020; 21: 1–67.
26.
XueL, et al.mT5: A massively multilingual pre-trained text-to-text transformer. arXiv preprint arXiv:2010.11934. 2020.
27.
DaiZ, et al.Transformer-xl: Attentive language models beyond a fixed-length context. arXiv preprint arXiv:1901.02860. 2019.
28.
BorgeaudS, et al.Improving language models by retrieving from trillions of tokensInternational conference on machine learning. Baltimore, MD: Proceedings of Machine Learning Research (PMLR), 2022, pp.2206–2240.
29.
JiangAQ, et al.Mixtral of experts. arXiv preprint arXiv:2401.04088. 2024.
30.
BalmauO, et al.Accelerating MoE Model Inference with Expert ShardingProceedings of the 5th Workshop on Machine Learning and Systems. Machine Learning and Systems (MLSys Workshop), 2025, pp.192–198.
31.
HuffMUlakçıE. Metacognitive Monitoring: A Human Ability Beyond Generative Artificial Intelligence. arXiv preprint arXiv:2410.13392 (2024).
32.
ZhouH, et al.Self-Monitoring Large Language Models for Click-Through Rate Prediction. New York: ACM Transactions on Information Systems, 2025.
33.
ZhouY, et al.Metacognitive retrieval-augmented large language models.Proceedings of the ACM on Web Conference 2024. Singapore: ACM (Association for Computing Machinery), 2024, pp.1453–1463.
34.
WangYZhaoY. Metacognitive prompting improves understanding in large language models. arXiv preprint arXiv:2308.05342 (2023).
35.
DherinB, et al.Learning without training: The implicit dynamics of in-context learning.arXiv preprint arXiv:2507.16003. 2025.
36.
WangX, et al.Self-consistency improves chain of thought reasoning in language models. arXiv preprint arXiv:2203.11171. 2022.
37.
BenderEM, et al.On the dangers of stochastic parrots: Can language models be too big?Proceedings of the 2021 ACM conference on fairness, accountability, and transparency. Virtual Event: ACM (Association for Computing Machinery), 2021, pp.610–623.
38.
KadavathS, et al.Language models (mostly) know what they know. arXiv preprint arXiv:2207.05221. 2022.
39.
RadfordA, et al.Language models are unsupervised multitask learners.OpenAI Blog2019; 1: 9.
40.
TangT, et al.Context-tuning: Learning contextualized prompts for natural language generation. arXiv preprint arXiv:2201.08670 (2022).
41.
AdiwardanaD, et al.Towards a human-like open-domain chatbot. arXiv preprint arXiv:2001.09977 (2020).
42.
ZhangR, et al.Contrastive data and learning for natural language processingProceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies: Tutorial Abstracts. Seattle, WA: Association for Computational Linguistics (ACL), 2022, pp.39–47.
43.
HuangQ, et al.Selective Prompting Tuning for Personalized Conversations with LLMsAnnual Meeting of the Association for Computational Linguistics. Bangkok: Association for Computational Linguistics (ACL), 2024, pp.16212–16226.
44.
KongA, et al.Self-prompt tuning: Enable autonomous role-playing in llms. arXiv preprint arXiv:2407.08995 (2024).
45.
DidolkarA, et al.Metacognitive Capabilities of LLMs: An Exploration in Mathematical Problem Solving. arXiv preprint arXiv:2405.12205 (2024).
46.
OhtaniRSakuraiYOyamaS. Does Metacognitive Prompting Improve Causal Inference in Large Language Models?2024 IEEE Conference on Artificial Intelligence (CAI). New York: IEEE, 2024, pp.458–459.
47.
RenzeMGuvenE. Self-Reflection in LLM Agents: Effects on Problem-Solving Performance. arXiv preprint arXiv:2405.06682 (2024).
48.
ToyJMacAdamJTaborP. Metacognition is all you need? using introspection in generative agents to improve goal-directed behavior. arXiv preprint arXiv:2401.10910 (2024).
49.
TanZ, et al.Tuning-Free Accountable Intervention for LLM Deployment–A Metacognitive Approach. arXiv preprint arXiv:2403.05636 (2024).
50.
GriotM, et al.Large language models lack essential metacognition for reliable medical reasoning.Nat Commun2025; 16: 642.
51.
BerkeAMcClellandJLAbbottJT. MetaCOG: Meta-cognitive object detection via Bayesian inference of perceptual reliabilityProceedings of Machine Learning Research. Vienna: Proceedings of Machine Learning Research (PMLR), 2024, pp.349–359. https://proceedings.mlr.press/v244/berke24a.html.
52.
PipitoneNHouir AlamiG. Legalbench-rag: A benchmark for retrieval-augmented generation in the legal domain. arXiv preprint arXiv:2408.10343 (2024). 2024.
53.
MadaanATandonNGuptaP, et al.Self-refine: iterative refinement with self-feedback. Adv Neural Inf Process Syst2023; 36: 46534–46594.
54.
ElBYuksekgonulMZouJ. Inefficiencies of Meta Agents for Agent Design. arXiv preprint arXiv:2510.06711. 2025.
55.
MadaanATandonNGuptaP, et al.Self-refine: iterative refinement with self-feedback. Adv Neural Inf Process Syst2023; 36: 46534–46594.
56.
AbbottJTMcClellandJL. MetaGen: Learning object representations and perceptual reliability via meta-cognitive generative modeling. arXiv preprint. https://arxiv.org/abs/2011.15067. 2021.
57.
WeberTDyerCBahdanauD. Meta-learning compositionality through guided training. Washington, DC: National Academy of Sciences (PNAS), 2023, https://pubmed.ncbi.nlm.nih.gov/37880371.
WuQ, et al.Autogen: Enabling next-gen LLM applications via multi-agent conversations.First Conference on Language Modeling2024.
77.
MavroudisV. LangChain. 2024.
78.
VenkadeshPDivyaSVKumarKS. Unlocking ai creativity: a multi-agent approach with crewai.Journal of Trends in Computer Science Smart Technology2024; 6: 338–356.
HongS, et al.MetaGPT: Meta programming for a multi-agent collaborative frameworkThe Twelfth International Conference on Learning Representations. Kigali: International Conference on Learning Representations (ICLR), 2023.
85.
WuQ, et al.Autogen: Enabling next-gen LLM applications via multi-agent conversationsFirst Conference on Language Modeling. Conference on Language Modeling (COLM), 2024.
86.
GoyalDGautamA. Introduction to LangChain frameworkTextual intelligence: large language models and their real-world applications. Wiley: Hoboken, NJ, 2025, pp.253–285.
87.
SapkotaRRoumeliotisKIKarkeeM. Ai agents vs. agentic ai: A conceptual taxonomy, applications and challenges. arXiv preprint arXiv:2505.10468 (2025). 2025.
88.
BeaulieuS, et al.Learning to continually learn. arXiv preprint arXiv:2002.09571. 2020.
89.
AbumallohRANilashiM. Agentic artificial intelligence: autonomy, decision-making, and responsibility in the age of intelligent systems. Journal of Soft Computing and Decision Support Systems2025; 12: 14–20.
90.
XuJ. Agentic Metacognition: Designing a” Self-Aware” Low-Code Agent for Failure Prediction and Human Handoff. arXiv preprint arXiv:2509.19783. 2025.
91.
CuadronA, et al.The danger of overthinking: Examining the reasoning-action dilemma in agentic tasks. arXiv preprint arXiv:2502.08235. 2025.
