Self-Evaluating Large Language Models: A Hybrid Least-Squares Learning and Reinforcement Planning Architecture for Reliable AI Decision Support
Keywords:
large language models, self-evaluation, least-squares learning, reinforcement planning, hybrid architecture, AI reliability, decision support systems, robustness, fairness, governanceAbstract
Large language models have demonstrated remarkable fluency and reasoning capabilities, yet their deployment in high-stakes decision support systems remains fraught with challenges related to reliability, calibration, and alignment. This paper proposes a hybrid architecture that integrates least-squares learning with reinforcement planning to enable self-evaluating large language models. The framework leverages a least-squares component for efficient uncertainty quantification and parameter estimation from observed outcomes, while a reinforcement planning module provides deliberative, goal-directed reasoning over long horizons. By coupling these two paradigms within a unified system, the model can continuously evaluate its own predictions, detect potential errors, and adjust its planning strategy in real time. We examine the structural trade-offs inherent in such an architecture, including the balance between computational efficiency and representational capacity, the tension between exploration and exploitation in planning, and the implications for robustness under distributional shift. Deployment considerations are discussed with respect to infrastructure requirements, latency constraints, and energy sustainability. Governance and fairness challenges are analyzed through the lens of algorithmic accountability, bias propagation, and the need for transparent self-evaluation mechanisms. Cross-domain comparisons from autonomous driving, clinical decision support, and financial risk assessment illustrate the transferability of the hybrid approach. The paper concludes with a forward-looking discussion on the policy implications of self-evaluating AI systems and the necessity of multi-stakeholder oversight.
References
1. Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., ... & Polosukhin, I. (2017). Attention is all you need. Advances in Neural Information Processing Systems, 30.
2. Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D., Dhariwal, P., ... & Amodei, D. (2020). Language models are few-shot learners. Advances in Neural Information Processing Systems, 33, 1877-1901.
3. Schulman, J., Wolski, F., Dhariwal, P., Radford, A., & Klimov, O. (2017). Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347.
4. Sutton, R. S., & Barto, A. G. (2018). Reinforcement learning: An introduction (2nd ed.). MIT Press.
5. Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., ... & Zhang, Y. (2023). Sparks of artificial general intelligence: Early experiments with GPT-4. arXiv preprint arXiv:2303.12712.
6. Gao, H., Zeng, W., Zhang, J., & Liang, Y. (2025, December). A large model API response quality prediction model based on least squares vector machine and SHAP interpretability analysis. In 2025 5th International Symposium on Artificial Intelligence and Big Data (AIBDF) (pp. 438-442). IEEE.
7. Russell, S., & Norvig, P. (2021). Artificial intelligence: A modern approach (4th ed.). Pearson.
8. Shui, Y., Jin, R., Dou, Z., & Gao, Z. (2026). ProtoGuard-SL: Prototype Consistency Based Backdoor Defense for Vertical Split Learning. arXiv preprint arXiv:2604.03595.
9. Zhou, D. (2025, December). M-VP2: Microservice-Oriented Vulnerability Patch Planning-A Cost-Aware Approachusing Multi-Agent Reinforcement Learning. In 2025 5th International Conference on Computer, Internet of Things and Control Engineering (CITCE) (pp. 248-254). IEEE.
10. Amodei, D., Olah, C., Steinhardt, J., Christiano, P., Schulman, J., & Mané, D. (2016). Concrete problems in AI safety. arXiv preprint arXiv:1606.06565.
11. Christiano, P., Shlegeris, B., & Amodei, D. (2018). Supervising strong learners by amplifying weak experts. arXiv preprint arXiv:1810.08575.
12. Hendrycks, D., Burns, C., Basart, S., Zou, A., Mazeika, M., Song, D., & Steinhardt, J. (2021). Measuring massive multitask language understanding. arXiv preprint arXiv:2009.03300.
13. Bai, Y., Jones, A., Ndousse, K., Askell, A., Chen, A., DasSarma, N., ... & Kaplan, J. (2022). Training a helpful and harmless assistant from human feedback. arXiv preprint arXiv:2204.05862.
14. Dou, Z., Zhao, Q., Wan, Z., Zhang, D., Wang, W., Raiyan, T., ... & Biswas, S. (2025). Plan Then Action: High-Level Planning Guidance Reinforcement Learning for LLM Reasoning. arXiv preprint arXiv:2510.01833.
15. Liang, P., Bommasani, R., Lee, T., Tsipras, D., Soylu, D., Yasunaga, M., ... & Hashimoto, T. (2022). Holistic evaluation of language models. arXiv preprint arXiv:2211.09110.
16. Wang, A., Pruksachatkun, Y., Nangia, N., Singh, A., Michael, J., Hill, F., ... & Bowman, S. R. (2019). SuperGLUE: A stickier benchmark for general-purpose language understanding systems. Advances in Neural Information Processing Systems, 32.
17. Kojima, T., Gu, S. S., Reid, M., Matsuo, Y., & Iwasawa, Y. (2022). Large language models are zero-shot reasoners. Advances in Neural Information Processing Systems, 35, 22199-22213.
18. Wei, J., Wang, X., Schuurmans, D., Bosma, M., Xia, F., Chi, E., ... & Zhou, D. (2022). Chain-of-thought prompting elicits reasoning in large language models. Advances in Neural Information Processing Systems, 35, 24824-24837.
19. Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., ... & Lowe, R. (2022). Training language models to follow instructions with human feedback. Advances in Neural Information Processing Systems, 35, 27730-27744.
20. Hendrycks, D., & Mazeika, M. (2022). X-risk analysis for AI research. arXiv preprint arXiv:2206.05862.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Journal of Data Intelligence and AI Systems
This work is licensed under a Creative Commons Attribution 4.0 International License.
This article is published under the Creative Commons Attribution 4.0 International License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
