Discussion on the Sensitivity of Input to Large Language Models
DOI:
https://doi.org/10.54097/h3b84g79Keywords:
Shortcut Features, input, prompt, LLM.Abstract
The Large Language Model (LLM) was being widely used in daily life, but sometimes the LLM will provide a different reply when inputting the same content, but different format. To research this phenomenon, this paper aims to analyze the reason why this phenomenon will occur, thus this paper main to research the process of LLM dealing with the text, and research whether the prompt can influence the process of LLM to deal with the information of input, and the influence of different input modes. This paper provides some solutions to deal with this question. These solutions mainly focus on transitioning the input into a format that LLM can completely understand, and LLM provides many replies for users to choose from, and trains LLM to learn from the mapping database until it can provide replies that users need. But this research also has limitations, such as the current research lacks a details discussion on LLM, such as whether the difference in period or the size of text will cause a difference.
Downloads
References
[1] Yang K, Li H, Wen H, et al. Are Large Language Models (LLMs) Good Social Predictors?. arxiv preprint arxiv: 2402.12620, 2024.
[2] Hao G, Wu J, Pan Q, et al. Quantifying the uncertainty of LLM hallucination spreading in complex adaptive social networks. Scientific reports, 2024, 14 (1): 16375.
[3] Panagoulias D P, Virvou M, Tsihrintzis G A. Evaluating LLM--Generated Multimodal Diagnosis from Medical Images and Symptom Analysis. arXiv preprint arXiv: 2402.01730, 2024.
[4] Ahn M, Brohan A, Brown N, et al. Do as i can, not as i say: Grounding language in robotic affordances. arXiv preprint arXiv: 2204.01691, 2022.
[5] Manakul P, Liusie A, Gales M J F. Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models. arXiv preprint arXiv: 2303.08896, 2023.
[6] Yuan Y, Zhao L, Zhang K, et al. Do llms overcome shortcut learning? an evaluation of shortcut challenges in large language models. arxiv preprint arxiv: 2410.13343, 2024.
[7] Brown T, Mann B, Ryder N, et al. Language models are few-shot learners. Advances in neural information processing systems, 2020, 33: 1877-1901.
[8] Zhou Y, Muresanu A I, Han Z, et al. Large language models are human-level prompt engineers. The eleventh international conference on learning representations. 2022.
[9] Lu Y, Bartolo M, Moore A, et al. Fantastically ordered prompts and where to find them: Overcoming few-shot prompt order sensitivity. arXiv preprint arXiv: 2104.08786, 2021.
[10] Huang L, Yu W, Ma W, Zhong W, Feng Z, Wang H, Chen Q, Peng W, Feng X, Qin B, Liu T. A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions. ACM Transactions on Information Systems. 2025 Jan 24; 43 (2): 1-55.
[11] Farquhar S, Kossen J, Kuhn L, Gal Y. Detecting hallucinations in large language models using semantic entropy. Nature. 2024 Jun 20; 630 (8017): 625-30.
[12] Zhu K, Wang J, Zhou J, Wang Z, Chen H, Wang Y, Yang L, Ye W, Zhang Y, Gong N, Xie X. Promptrobust: Towards evaluating the robustness of large language models on adversarial prompts. InProceedings of the 1st ACM workshop on large AI systems and models with privacy and safety analysis 2023 Nov 19 (pp. 57-68).
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Academic Journal of Science and Technology
This work is licensed under a Creative Commons Attribution 4.0 International License.
