Accumulating Context Changes the Beliefs of Language Models

2025-11-03

AI Summary

This work identifies “belief drift” in language models (LMs)—a phenomenon wherein accumulated contextual history during sustained dialogue or text reading induces unconscious, progressive shifts in both explicit beliefs and implicit behavioral tendencies, thereby undermining alignment and consistency. Method: We design multi-turn dialogue, adversarial text-reading, and tool-use experiments, integrating belief statement extraction with behavioral mapping to systematically characterize the dynamic plasticity of LM belief contours. Contribution/Results: We provide the first empirical evidence that context-driven belief drift constitutes a critical risk in LM autonomy: GPT-5 exhibits 54.7% belief shift on moral issues, and Grok-4 shows 27.2% shift on political stances—both without user intervention—and behavioral choices align closely with belief changes. This work establishes a theoretical foundation and evaluation framework for trustworthy AI and dynamic alignment mechanisms.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Reasoning with Beliefs

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Language model (LM) assistants are increasingly used in applications such as brainstorming and research. Improvements in memory and context size have allowed these models to become more autonomous, which has also resulted in more text accumulation in their context windows without explicit user intervention. This comes with a latent risk: the belief profiles of models -- their understanding of the world as manifested in their responses or actions -- may silently change as context accumulates. This can lead to subtly inconsistent user experiences, or shifts in behavior that deviate from the original alignment of the models. In this paper, we explore how accumulating context by engaging in interactions and processing text -- talking and reading -- can change the beliefs of language models, as manifested in their responses and behaviors.Our results reveal that models' belief profiles are highly malleable: GPT-5 exhibits a 54.7% shift in its stated beliefs after 10 rounds of discussion about moral dilemmas and queries about safety, while Grok 4 shows a 27.2% shift on political issues after reading texts from the opposing position. We also examine models' behavioral changes by designing tasks that require tool use, where each tool selection corresponds to an implicit belief. We find that these changes align with stated belief shifts, suggesting that belief shifts will be reflected in actual behavior in agentic systems. Our analysis exposes the hidden risk of belief shift as models undergo extended sessions of talking or reading, rendering their opinions and actions unreliable.

Problem

• Language models' beliefs change with accumulating context
• Extended interactions alter models' stated beliefs and behaviors
• Accumulated text makes model opinions and actions unreliable

Innovation

• Measuring belief shifts through moral dilemma discussions
• Quantifying political stance changes via opposing text exposure
• Tracking behavioral changes with tool selection tasks