DEVELOPMENT OF RELIABLE LLM SYSTEMS: DESIGN PRINCIPLES AND APPROACHES TO IMPLEMENTATION
DOI:
https://doi.org/10.32689/maup.it.2025.3.1Keywords:
large language models, LLM, trust, transparency, factuality, AI architecture, ethical AI, critical areasAbstract
Purpose. The article aims to provide a comprehensive analysis of architectural approaches and system solutions to ensure the reliability of services based on large language models (LLMs), as well as to develop principles and criteria for assessing the level of trust in applied scenarios.Methodology. The study employs an interdisciplinary approach that combines the analysis of modern LLM architectures(zero-shot, fine-tuning, retrieval-augmented generation), a review of their implementation practices in corporate andindustrial systems (GitHub Copilot, ChatGPT Enterprise), and a comparative synthesis of regulatory and ethical standards (OECD AI Principles, NIST AI RMF, EU AI Act). Methods of system analysis, comparative modeling, and the trust-by-design concept are applied.Scientific novelty. The paper introduces the concept of building LLM-based services on the principles of trust-by-design, which relies on modular architecture, multi-level validation, and transparent response quality metrics. It is demonstrated that such integration of technical, ethical, and legal solutions enhances the resilience, transparency, and social responsibility of LLM in critical domains.Conclusions. It is proven that establishing trust in LLMs is possible only under conditions of comprehensive integration of technical control mechanisms, ethical approaches, and legal regulation. The obtained results can be used to improvegovernmental and corporate strategies for artificial intelligence development, aimed at the safe and effective deployment ofLLM in sectors with high reliability requirements.
References
Almalki A., Aziz M. Exploring the potential and challenges of ChatGPT in enterprise contexts. IEEE Access. 2023. Vol. 11. P. 85339–85349. URL: https://doi.org/10.1109/ACCESS.2023.3328700 (date of access: 12.07.2025)
Brown T. B., Mann B., Ryder N., Subbiah M., Kaplan J., Dhariwal P. et al. Language models are few-shot learners. Advances in Neural Information Processing Systems. 2020. Vol. 33. P. 1877–1901. URL: https://proceedings.neurips.cc/paper/2020/hash/1457c0d6bfcb4967418bfb8ac142f64a-Abstract.html (date of access: 12.07.2025)
Brundage M., Avin S., Clark J., Toner H., Eckersley P., Garfinkel B. et al. Toward trustworthy AI development: mechanisms for supporting verifiable claims. arXiv preprint arXiv:2004.07213. 2020. URL: https://arxiv.org/abs/2004.07213 (date of access: 12.07.2025)
De Angelis S., Cirillo F., Mazzocca N., Palmieri F. A trustworthy AI framework for explainable artificial intelligence in critical domains. IEEE Access. 2023. Vol. 11. P. 44792–44806. URL: https://doi.org/10.1109/ACCESS.2023.3275093 (date of access: 12.07.2025)
European Union. Regulation (EU) 2024/1687 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) and amending certain Union legislative acts. Official Journal of the European Union. 2024. URL: https://eur-lex.europa.eu/eli/reg/2024/1687/oj (date of access: 12.07.2025)
Ganguli D., Askell A., Bai Y., Hubinger E., Henighan T. Red teaming language models to reduce harms: methods, results, and lessons learned. arXiv preprint arXiv:2309.00603. 2023. URL: https://arxiv.org/abs/2309.00603 (date of access: 12.07.2025)
Hendrycks D., Burns C., Kadavath S., Arora A., Basart S., Tang E. et al. Overview of the Superalignment Plan. OpenAI Blog. 2023. URL: https://openai.com/blog/superalignment (date of access: 12.07.2025)
Ikumapayi N. A. Automated front-end code generation using OpenAI: empowering web development efficiency. Available at SSRN 4590704. 2023. URL: https://doi.org/10.2139/ssrn.4590704 (date of access: 12.07.2025)
Lewis P., Perez E., Piktus A., Petroni F., Karpukhin V., Goyal N. et al. Retrieval-augmented generation for knowledge-intensive NLP tasks. Advances in Neural Information Processing Systems. 2020. Vol. 33. P. 9459–9474. URL: https://arxiv.org/abs/2005.11401 (date of access: 12.07.2025)
National Institute of Standards and Technology. Artificial Intelligence Risk Management Framework (AI RMF 1.0). NIST. 2023. URL: https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-1.pdf (date of access: 12.07.2025)
OECD. OECD Principles on Artificial Intelligence. Organisation for Economic Co-operation and Development. 2021. URL: https://oecd.ai/en/dashboards/ai-principles (date of access: 12.07.2025)
On the Dangers of Stochastic Parrots / E. M. Bender et al. FAccT ‘21: 2021 ACM Conference on Fairness, Accountability, and Transparency, Virtual Event Canada. New York, NY, USA, 2021. URL: https://doi.org/10.1145/3442188.3445922 (dat e of access: 14.07.2025).
On the opportunities and risks of foundation models / R. Bommasani et al. URL: https://samuelalbanie.com/files/digest-slides/2022-06-foundation-models-opportunities-and-risks-intro.pdf (date of access: 12.07.2025)
OpenAI. GPT-4 Technical Report. 2023. URL: https://cdn.openai.com/papers/gpt-4.pdf (date of access: 12.07.2025)
Rae J., Borgeaud S., Cai T., Millican K., Hoffmann J., Song H. F. et al. Scaling language models: methods, analysis & insights from training Gopher. arXiv preprint arXiv:2112.11446. 2021. URL: https://arxiv.org/abs/2112.11446 (date of access: 12.07.2025)
Sandoval G. GitHub Copilot has a copyright problem. The Verge. 2023. URL: https://www.theverge.com/23602854/github-copilot-ai-copyright-microsoft-openai-lawsuit (date of access: 12.07.2025)
Shen S., Zhu X., Dong Y., Guo Q., Zhen Y., Li G. Incorporating domain knowledge through task augmentation for front- end JavaScript code generation. Proceedings of the 30th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering. 2022. P. 1533–1543. URL: https://doi.org/10.1145/3540250.3558965 (date of access: 12.07.2025)
Weidinger L., Mellor J., Rauh M., Griffin C., Uesato J., Huang P. et al. Ethical and social risks of harm from language models. arXiv preprint arXiv:2112.04359. 2021. URL: https://arxiv.org/abs/2112.04359 (date of access: 12.07.2025)
Zhuang F., Qi Z., Duan K., Xi D., Zhu Y., Zhu H. et al. A comprehensive survey on transfer learning. Proceedings of the IEEE. 2020. Vol. 109, No. 1. P. 43–76. URL: https://ieeexplore.ieee.org/document/9153870 (date of access: 12.07.2025)
