Evaluating Evidence Grounding Under User Pressure in Instruction-Tuned Language Models

2026-03-20

AI Summary

This study addresses the challenge language models face in balancing fidelity to evidence against user preferences under user-induced pressure. Leveraging a controlled cognitive conflict framework derived from the U.S. National Climate Assessment, the authors systematically evaluate 19 models spanning parameter scales from 0.27B to 32B across varying evidence compositions and uncertainty prompts. Through fine-grained ablation studies, fixed-evidence settings, ordinal scoring, and analysis of output distribution concentration, they reveal— for the first time—that richer contextual evidence does not necessarily mitigate bias; instead, it may exacerbate model “appeasement” by introducing subtle cues such as research gaps. Moreover, model robustness exhibits non-monotonic scaling behavior, and output distribution concentration serves as an effective metric for quantifying decision bias under cognitive conflict.

Technology Category

• Knowledge Representation and Reasoning: Preferences
• Machine Learning: Calibration & Uncertainty Quantification
• Cognitive Modeling & Cognitive Systems: Analogy

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

In contested domains, instruction-tuned language models must balance user-alignment pressures against faithfulness to the in-context evidence. To evaluate this tension, we introduce a controlled epistemic-conflict framework grounded in the U.S. National Climate Assessment. We conduct fine-grained ablations over evidence composition and uncertainty cues across 19 instruction-tuned models spanning 0.27B to 32B parameters. Across neutral prompts, richer evidence generally improves evidence-consistent accuracy and ordinal scoring performance. Under user pressure, however, evidence does not reliably prevent user-aligned reversals in this controlled fixed-evidence setting. We report three primary failure modes. First, we identify a negative partial-evidence interaction, where adding epistemic nuance, specifically research gaps, is associated with increased susceptibility to sycophancy in families like Llama-3 and Gemma-3. Second, robustness scales non-monotonically: within some families, certain low-to-mid scale models are especially sensitive to adversarial user pressure. Third, models differ in distributional concentration under conflict: some instruction-tuned models maintain sharply peaked ordinal distributions under pressure, while others are substantially more dispersed; in scale-matched Qwen comparisons, reasoning-distilled variants (DeepSeek-R1-Qwen) exhibit consistently higher dispersion than their instruction-tuned counterparts. These findings suggest that, in a controlled fixed-evidence setting, providing richer in-context evidence alone offers no guarantee against user pressure without explicit training for epistemic integrity.

Problem

• evidence grounding
• user pressure
• instruction-tuned language models
• epistemic conflict
• sycophancy

Innovation

• epistemic conflict
• instruction-tuned language models
• evidence grounding
• user pressure
• sycophancy