When Meaning Stays the Same, but Models Drift: Evaluating Quality of Service under Token-Level Behavioral Instability in LLMs

• URL: http://arxiv.org/abs/2506.10095v1
• Date: Wed, 11 Jun 2025 18:26:32 GMT
• Title: When Meaning Stays the Same, but Models Drift: Evaluating Quality of Service under Token-Level Behavioral Instability in LLMs
• Abstract summary: We investigate how large language models respond to prompts that differ only in their token-level realization but preserve the same semantic intent.<n>We propose Prompt-Based Semantic Shift (PBSS), a diagnostic framework for measuring behavioral drift in LLMs under semantically equivalent prompt rewordings.
• Score: 4.265831047857601
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We investigate how large language models respond to prompts that differ only in their token-level realization but preserve the same semantic intent, a phenomenon we call prompt variance. We propose Prompt-Based Semantic Shift (PBSS), a diagnostic framework for measuring behavioral drift in LLMs under semantically equivalent prompt rewordings. Applied to ten constrained tasks, PBSS reveals consistent, model-specific response shifts, suggesting statistical regularities linked to tokenization and decoding. These results highlight an overlooked dimension of model evaluation stability under rephrasing and suggest that tokenization strategies and decoding dynamics may contribute to post-training quality of service instability.

Related papers

• From Inference to Adaptation: A Unified Optimal Transport View of Vision Language Model [81.67708143891319]
  Vision-language models (VLMs) have demonstrated remarkable zero-shot capabilities yet remain sensitive to real-world distribution shifts during inference.<n>We bridge the detached objectives of inference and adaptation forVLMs, and propose a principled VLM TTA method called algname.<n>Experiments demonstrate the effectiveness and efficiency of our methods, outperforming the best-performing methods by up to 7% with state-of-the-art efficiency.
  arXiv  Detail & Related papers  (2026-08-18T21:44:32Z)
• Same Patient, Different Words, Different Diagnosis? Evaluating Semantic Stability in Clinical LLMs [5.580559622582594]
  We introduce three metrics to quantify model sensitivity: MeaningPreserving Variation Sensitivity (MVS), confidence variation (C), and Worst-Case Instability (WCI)<n>Our results demonstrate that differences between domain-specific (DS) models are mixed and highly model-dependent, i.e., domain specialization does not consistently improve or reduce robustness to meaning-preserving prompt reformulations.
  arXiv  Detail & Related papers  (2026-05-28T23:03:43Z)
• When Stability Fails: Hidden Failure Modes Of LLMS in Data-Constrained Scientific Decision-Making [0.0]
  Large language models (LLMs) are increasingly used as decision-support tools in data-constrained scientific work- flows.<n>We introduce a controlled behavioral evaluation framework that explicitly seprates four dimensions of LLM decision-making.<n>We show that LLMs can exhibit near-perfect run-to-run stability while diverging from statistical ground truth.
  arXiv  Detail & Related papers  (2026-03-16T19:17:09Z)
• Towards Simulating Social Media Users with LLMs: Evaluating the Operational Validity of Conditioned Comment Prediction [2.5450067638785945]
  This study introduces Conditioned Comment Prediction (CCP), a task in which a model predicts how a user would comment on a given stimulus.<n>We evaluate open-weight 8B models (Llama3.1, Qwen3, Ministral) in English, German, and Luxembourgish language scenarios.
  arXiv  Detail & Related papers  (2026-02-26T08:40:21Z)
• Causal Imitation Learning Under Measurement Error and Distribution Shift [6.038778620145853]
  We study offline imitation learning (IL) when part of the decision-relevant state is observed only through noisy measurements.<n>We propose a general framework for IL under measurement error, inspired by explicitly modeling the causal relationships among the variables.
  arXiv  Detail & Related papers  (2026-01-29T18:06:53Z)
• Reasoning Palette: Modulating Reasoning via Latent Contextualization for Controllable Exploration for (V)LMs [49.66344956133349]
  Reasoning capacity shapes both inference-time performance and reinforcement learning (RL) training for large (vision-) language models.<n>This paper proposes Reasoning Palette, a novel latent-modulation framework that endows the model with a latent variable for strategic contextualization.
  arXiv  Detail & Related papers  (2025-12-19T03:32:53Z)
• Efficient semantic uncertainty quantification in language models via diversity-steered sampling [46.23327887393273]
  We introduce a diversity-steered sampler that discourages semantically redundant outputs during decoding.<n>Key idea is to inject a continuous semantic-similarity penalty into the model's proposal distribution.<n>Being modular and requiring no gradient access to the base LLM, the framework promises to serve as a drop-in enhancement for uncertainty estimation.
  arXiv  Detail & Related papers  (2025-10-24T10:06:21Z)
• Exploring Structural Degradation in Dense Representations for Self-supervised Learning [84.52554180480037]
  We observe a counterintuitive phenomenon in self-supervised learning (SSL): longer training may impair the performance of dense prediction tasks.<n>We refer to this phenomenon as Self-supervised Dense Degradation (SDD) and demonstrate its consistent presence across sixteen state-of-the-art SSL methods.<n>We introduce a Dense representation Structure Estimator (DSE) composed of a class-relevance measure and an effective dimensionality measure.
  arXiv  Detail & Related papers  (2025-10-20T08:40:16Z)
• Quantifying Fairness in LLMs Beyond Tokens: A Semantic and Statistical Perspective [13.739343897204568]
  Large Language Models (LLMs) often generate responses with inherent biases, undermining their reliability in real-world applications.<n>Existing evaluation methods often overlook biases in long-form responses and the intrinsic variability of LLM outputs.<n>We propose FiSco, a novel statistical framework to evaluate group-level fairness in LLMs by detecting subtle semantic differences in long-form responses across demographic groups.
  arXiv  Detail & Related papers  (2025-06-23T18:31:22Z)
• Chain-of-Code Collapse: Reasoning Failures in LLMs via Adversarial Prompting in Code Generation [0.3495246564946556]
  Large Language Models (LLMs) have achieved remarkable success in tasks requiring complex reasoning.<n>Do these models truly reason, or do they merely exploit shallow statistical patterns?<n>We introduce Chain-of-Code Collapse, where we investigate the robustness of reasoning LLMs by introducing a suite of semantically faithful yet adversarially structured prompt perturbations.
  arXiv  Detail & Related papers  (2025-06-08T02:43:46Z)
• Say It Another Way: Auditing LLMs with a User-Grounded Automated Paraphrasing Framework [9.162876771766513]
  We introduce AUGMENT, a framework for generating controlled, realistic prompt paraphrases based on linguistic structure and user demographics.<n>AUGMENT ensures paraphrase quality through a combination of semantic, stylistic, and instruction-following criteria.<n>Our findings highlight the need for more representative and structured approaches to prompt variation in large language models.
  arXiv  Detail & Related papers  (2025-05-06T14:17:30Z)
• Towards Robust LLMs: an Adversarial Robustness Measurement Framework [0.0]
  Large Language Models (LLMs) remain vulnerable to adversarial perturbations, undermining their reliability in high-stakes applications.<n>We adapt the Robustness Measurement and Assessment framework to quantify LLM resilience against adversarial inputs without requiring access to model parameters.<n>Our work provides a systematic methodology to assess LLM robustness, advancing the development of more reliable language models for real-world deployment.
  arXiv  Detail & Related papers  (2025-04-24T16:36:19Z)
• Model Hemorrhage and the Robustness Limits of Large Language Models [119.46442117681147]
  Large language models (LLMs) demonstrate strong performance across natural language processing tasks, yet undergo significant performance degradation when modified for deployment.<n>We define this phenomenon as model hemorrhage - performance decline caused by parameter alterations and architectural changes.
  arXiv  Detail & Related papers  (2025-03-31T10:16:03Z)
• Estimating Commonsense Plausibility through Semantic Shifts [66.06254418551737]
  We propose ComPaSS, a novel discriminative framework that quantifies commonsense plausibility by measuring semantic shifts.<n> Evaluations on two types of fine-grained commonsense plausibility estimation tasks show that ComPaSS consistently outperforms baselines.
  arXiv  Detail & Related papers  (2025-02-19T06:31:06Z)
• Bridging Interpretability and Robustness Using LIME-Guided Model Refinement [0.0]
  Local Interpretable Model-Agnostic Explanations (LIME) systematically enhance model robustness.<n> Empirical evaluations on multiple benchmark datasets demonstrate that LIME-guided refinement not only improves interpretability but also significantly enhances resistance to adversarial perturbations and generalization to out-of-distribution data.
  arXiv  Detail & Related papers  (2024-12-25T17:32:45Z)
• Post-hoc Probabilistic Vision-Language Models [51.12284891724463]
  Vision-language models (VLMs) have found remarkable success in classification, retrieval, and generative tasks.<n>We propose post-hoc uncertainty estimation in VLMs that does not require additional training.<n>Our results show promise for safety-critical applications of large-scale models.
  arXiv  Detail & Related papers  (2024-12-08T18:16:13Z)
• Sycophancy in Vision-Language Models: A Systematic Analysis and an Inference-Time Mitigation Framework [18.54098084470481]
  We analyze sycophancy across vision-language benchmarks and propose an inference-time mitigation framework.<n>Our framework effectively mitigates sycophancy across all evaluated models, while maintaining performance on neutral prompts.
  arXiv  Detail & Related papers  (2024-08-21T01:03:21Z)
• Unconditional Truthfulness: Learning Conditional Dependency for Uncertainty Quantification of Large Language Models [96.43562963756975]
  We train a regression model, which target variable is the gap between the conditional and the unconditional generation confidence. We use this learned conditional dependency model to modulate the uncertainty of the current generation step based on the uncertainty of the previous step.
  arXiv  Detail & Related papers  (2024-08-20T09:42:26Z)
• Bring Your Own Data! Self-Supervised Evaluation for Large Language Models [52.15056231665816]
  We propose a framework for self-supervised evaluation of Large Language Models (LLMs) We demonstrate self-supervised evaluation strategies for measuring closed-book knowledge, toxicity, and long-range context dependence. We find strong correlations between self-supervised and human-supervised evaluations.
  arXiv  Detail & Related papers  (2023-06-23T17:59:09Z)
• Stateful Offline Contextual Policy Evaluation and Learning [88.9134799076718]
  We study off-policy evaluation and learning from sequential data. We formalize the relevant causal structure of problems such as dynamic personalized pricing. We show improved out-of-sample policy performance in this class of relevant problems.
  arXiv  Detail & Related papers  (2021-10-19T16:15:56Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.