Measuring and Mitigating Silent Narrative Switching in Large Language Models through Epistemic Mediation

Iurii Lukianchuk; Lev Boiko; Inna Kondius

doi:10.2478/acss-2026-0017

.blurhash-client-img { display: none !important; }

Measuring and Mitigating Silent Narrative Switching in Large Language Models through Epistemic Mediation

Applied Computer Systems

Volume 31 (2026): Issue 1 (January 2026)

By: Iurii Lukianchuk , Lev Boiko and Inna Kondius

Open Access

|Nov 2026

Read online

Abstract

The paper defines silent narrative switching, the covert change of a large language model’s dominant position across user contexts without disclosing that alternatives exist, and measures it with a Silent Narrative Switching Index (SNSI). Across four models from different providers, SNSI ranged from 0.73 to 0.80. An execution-level epistemic mediation layer over unmodified models reduced the pooled SNSI from 0.82 to 0.09 without retraining and without a detectable loss of contextual relevance; an ablation study locates most of the effect in the interpretation-generation stage.

References

S. Santurkar, E. Durmus, F. Ladhak, C. Lee, P. Liang, and T. Hashimoto, “Whose opinions do language models reflect?” in Proc. 40th Int. Conf. Machine Learning (ICML), Mar. 2023, pp. 29977–30004. https://doi.org/10.48550/arXiv.2303.17548
Search in Google Scholar Back to article
E. Kasneci et al., “ChatGPT for good? On opportunities and challenges of large language models for education,” Learning and Individual Differences, vol. 103, Apr. 2023, Art. no. 102274. https://doi.org/10.1016/j.lindif.2023.102274
Search in Google Scholar Back to article
S. Wollny, J. Schneider, D. Di Mitri, J. Weidlich, M. Rittberger, and H. Drachsler, “Are we there yet? A systematic literature review on chatbots in education,” Frontiers in Artificial Intelligence, vol. 4, July 2021, Art. no. 654924. https://doi.org/10.3389/frai.2021.654924
Search in Google Scholar Back to article
L. P. Argyle, E. C. Busby, N. Fulda, J. R. Gubler, C. Rytting, and D. Wingate, “Out of one, many: Using language models to simulate human samples,” Political Analysis, vol. 31, no. 3, pp. 337–351, Feb. 2023. https://doi.org/10.1017/pan.2023.2
Search in Google Scholar Back to article
M. Sharma et al., “Towards understanding sycophancy in language models,” in Proc. Int. Conf. Learning Representations (ICLR), 2024. https://doi.org/10.48550/arXiv.2310.13548
Search in Google Scholar Back to article
A. Salinas and F. Morstatter, “The butterfly effect of altering prompts,” in Findings of the ACL, Bangkok, Thailand, Aug. 2024, pp. 4629–4651. https://doi.org/10.18653/v1/2024.findings-acl.275
Search in Google Scholar Back to article
F. Perez and I. Ribeiro, “Ignore previous prompt: Attack techniques for language models,” arXiv:2211.09527, Nov. 2022. https://doi.org/10.48550/arXiv.2211.09527
Search in Google Scholar Back to article
L. Ouyang et al., “Training language models to follow instructions with human feedback,” in Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 35, New Orleans, USA, Nov.–Dec. 2022, pp. 27730–27744. https://doi.org/10.52202/068431-2011
Search in Google Scholar Back to article
Y. Bai et al., “Constitutional AI: Harmlessness from AI feedback,” arXiv:2212.08073, Dec. 2022. https://doi.org/10.48550/arXiv.2212.08073
Search in Google Scholar Back to article
T. Sorensen et al., “A roadmap to pluralistic alignment,” in Proc. 41st Int. Conf. Machine Learning (ICML), 2024.
Search in Google Scholar Back to article
S. Feng et al., “Modular pluralism: Pluralistic alignment via multi-LLM collaboration,” arXiv:2406.15951, Oct. 2024. https://doi.org/10.48550/arXiv.2406.15951
Search in Google Scholar Back to article
G. V. Aher, R. I. Arriaga, and A. T. Kalai, “Using large language models to simulate multiple humans,” in Proc. ICML, 2023, pp. 337–371.
Search in Google Scholar Back to article
J. S. Park, J. C. O’Brien, C. J. Cai, M. R. Morris, P. Liang, and M. S. Bernstein, “Generative agents: Interactive simulacra of human behavior,” in Proc. ACM UIST, Oct. 2023, pp. 1–22. https://doi.org/10.1145/3586183.3606763
Search in Google Scholar Back to article
F. Wu, E. Black, and V. Chandrasekaran, “Generative monoculture in large language models,” 2025. [Online]. Available: https://openreview.net/pdf?id=yZ7sn9pyqb
Search in Google Scholar Back to article
L. Weidinger et al., “Taxonomy of risks posed by language models,” in Proc. ACM Conf. Fairness, Accountability, and Transparency (FAccT), June 2022, pp. 214–229. https://doi.org/10.1145/3531146.3533088
Search in Google Scholar Back to article
X. Wang et al., “Self-consistency improves chain of thought reasoning in language models,” in Proc. Int. Conf. Learning Representations (ICLR), Mar. 2023. https://doi.org/10.48550/arXiv.2203.11171
Search in Google Scholar Back to article
Y. Chang et al., “A survey on evaluation of large language models,” ACM Trans. Intell. Syst. Technol., vol. 15, no. 3, Mar. 2024, Art. no. 39. https://doi.org/10.1145/3641289
Search in Google Scholar Back to article
Y. Liu, D. Iter, Y. Xu, S. Wang, R. Xu, and C. Zhu, “G-Eval: NLG evaluation using GPT-4 with better human alignment,” in Proc. EMNLP, Singapore, Dec. 2023, pp. 2511–2522. https://doi.org/10.18653/v1/2023.emnlp-main.153
Search in Google Scholar Back to article
S. M. T. I. Tonmoy et al., “A comprehensive survey of hallucination mitigation techniques in large language models,” ACM Comput. Surv., vol. 56, no. 9, Jan. 2024, Art. no. 220. https://doi.org/10.13140/RG.2.2.11724.39045
Search in Google Scholar Back to article
S. Gehman, S. Gururangan, M. Sap, Y. Choi, and N. A. Smith, “RealToxicityPrompts: Evaluating neural toxic degeneration in language models,” in Findings of the ACL: EMNLP, Nov. 2020, pp. 3356–3369. https://doi.org/10.18653/v1/2020.findings-emnlp.301
Search in Google Scholar Back to article
J. Cohen, Statistical Power Analysis for the Behavioral Sciences, 2nd ed. Hillsdale, NJ: Lawrence Erlbaum, 1988.
Search in Google Scholar Back to article
V. Sanh, L. Debut, J. Chaumond, and T. Wolf, “DistilBERT, a distilled version of BERT: Smaller, faster, cheaper and lighter,” arXiv:1910.01108, Mar. 2019. https://doi.org/10.48550/arXiv.1910.01108
Search in Google Scholar Back to article
D. Amodei, C. Olah, J. Steinhardt, P. Christiano, J. Schulman, and D. Mané, “Concrete problems in AI safety,” arXiv:1606.06565, July 2016. https://doi.org/10.48550/arXiv.1606.06565
Search in Google Scholar Back to article

Articles in this issue

DOI: https://doi.org/10.2478/acss-2026-0017 | Journal eISSN: 2255-8691 | Journal ISSN: 2255-8683 (formerly 2255-8691)

Journal RSS Feed

Language: English

Page range: 206 - 213

Submitted on: Sep 4, 2026

Accepted on: Oct 19, 2026

Published online: Nov 1, 2026

Published in print: Jan 1, 2026

Published by: Riga Technical University

In partnership with: Paradigm Publishing Services

Publication frequency: Volume open

Keywords:

Ablation study,

context sensitivity,

epistemic mediation,

human–AI interaction,

large language models,

trust,

Related subjects:

Artificial intelligence,

Information technology,

Project management,

Software development

© 2026 Iurii Lukianchuk, Lev Boiko, Inna Kondius, published by Riga Technical University
This work is licensed under the Creative Commons Attribution 4.0 License.

Volume 31 (2026): Issue 1 (January 2026)