Inference in large language models resists repeated context influence
Not All Answers Are Contextually Persuadable: Inference Dynamics in Large Language Models under Contextual Influence
Artificial Intelligence
Summary
Sometimes, when you give a large language AI model extra hints or context repeatedly, it doesn't keep changing its answers as you might expect. The authors show that these models tend to settle into stable ways of answering based on the question, so repeating the same hints over and over won't necessarily change the model’s prediction. However, in some cases, a change in answer is inevitable. They back this up with experiments that match their mathematical analysis. This helps us better understand how and when models use extra context during thinking.
What this means in practice
- •For ai developers: Design more reliable prompts knowing that repeating the same context does not always change model outputs.
- •For chatbot designers: Improve chatbot response consistency by understanding when repeated context inputs will not sway model answers.
Authors
Zongye Hu, Weiqing Luo, Yanjie Fu, Yu Gan, Haofeng Zhang, Ziyi Huang
Abstract
At the core of modern prompting techniques is contextual sensitivity, the ability of large language models to adapt their predictions based on inference-time context. Despite its central role, inference behavior under strong contextual influence remains poorly understood, particularly at the level of internal inference dynamics. We introduce a theoretical framework for analyzing contextual influence through inference dynamics, enabling quantitative characterization of inference behavior beyond output-level answer changes. Our analysis shows that inference dynamics do not exhibit unbounded drift under repeated contextual assertions. Instead, predictive representations converge to stable, query-dependent regimes that fundamentally constrain whether contextual signals can alter a model's prediction. This leads to a surprising finding: Repeated contextual assertions do not act as accumulating evidence during inference and may therefore fail to alter a model's prediction even under unbounded repetition, while in other cases a prediction change becomes inevitable. We empirically validate our theoretical predictions, demonstrating strong alignment between theory and observed inference behavior. These contributions offer a principled pathway toward characterizing the limits of contextual influence during inference, providing practical implications for model development.