Decoding-Level Taboo: A Diagnostic Stress Test for LLM Robustness

• URL: http://arxiv.org/abs/2608.09900v2
• Date: Tue, 11 Aug 2026 18:46:25 GMT
• Title: Decoding-Level Taboo: A Diagnostic Stress Test for LLM Robustness
• Abstract summary: Decoding-Level Taboo is a zero-prompt diagnostic stress test that intervenes directly in logit space at runtime.<n>By dynamically masking primary candidate tokens at word boundaries, Taboo forces machine circumlocution.<n>Off-path robustness is heavily influenced by both parameter scale and post-training instruction alignment.
• Score: 8.2408248134396
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large language model evaluations typically focus on performance under nominal conditions, creating an illusion of capability where models comfortably walk a narrow, highly optimized generation corridor. In real-world deployments, however, complex system prompts, safety guardrails, and structural constraints continuously force models off this nominal path, driving a divergence between benchmark scores and deployment performance. To address this issue, we introduce Decoding-Level Taboo, a zero-prompt diagnostic stress test that intervenes directly in logit space at runtime, forcing models out of their nominal paths. By dynamically masking primary candidate tokens at word boundaries, Taboo forces machine circumlocution. Evaluating Taboo across several open-weight model families reveals that off-path robustness is heavily influenced by both parameter scale and post-training instruction alignment, with robustness generally improving with model size and alignment. Beyond the results presented in this paper, Taboo provides a novel primitive for generating diverse synthetic datasets, stress-testing runtime safety guardrails, and auditing model reliability prior to real-world deployment.

Related papers

• Dance Across Shifts: Forward-Facilitation Continual Test-Time Adaptation through Dynamic Style Bridging [65.11234199720428]
  Continual Test-Time Adaptation aims to empower perception systems to handle dynamic distribution shifts encountered after deployment.<n>We introduce a novel forward-facilitation paradigm through a method termed Dynamic Style Bridging.<n>During test time, to mitigate inherent generative bias and adapt these proxies to incoming data, we propose a multi-level bridging mechanism.
  arXiv  Detail & Related papers  (2026-05-18T16:18:17Z)
• VLA-ATTC: Adaptive Test-Time Compute for VLA Models with Relative Action Critic Model [54.35791816657227]
  We introduce textbfVLA-ATTC', a framework that endows VLA models with adaptive test-time compute.<n>VLA-ATTC employs an uncertainty-based cognitive clutch'' to dynamically transition from reflexive execution to a TTC deliberation phase.<n>On the LIBERO-LONG benchmark, VLA-ATTC reduces the failure rate of the SOTA model PI0.5 by over 50%.
  arXiv  Detail & Related papers  (2026-05-02T02:13:11Z)
• ABLE: Representing and Mapping LLMs via Attribution-Based Large-model Embedding [23.27523995613204]
  We propose ABLE (Attribution-Based Large-model Embedding), a framework that leverages the interpretability space to construct model representations.<n>By aggregating gradient-based feature attributions via a tokenizer-agnostic word-level alignment, ABLE captures model-specific input-sensitivity patterns.<n>Our training-free approach achieves competitive or superior performance in relation prediction, model routing, and benchmark score prediction.
  arXiv  Detail & Related papers  (2026-04-20T04:55:11Z)
• Dynamic analysis enhances issue resolution [53.50448142467294]
  DAIRA (Dynamic Analysis-enhanced Issue Resolution Agent) is an automated repair framework that embeds dynamic analysis into the agent's reasoning cycle.<n> Driven by a Test Tracing-Driven methodology, DAIRA utilizes lightweight monitors to extract critical runtime data.<n>When powered by Gemini 3 Flash Preview, DAIRA establishes a new state-of-the-art (SOTA) performance, achieving a 79.4% resolution rate on the SWE-bench Verified dataset.
  arXiv  Detail & Related papers  (2026-03-23T14:48:54Z)
• Not All Preferences Are Created Equal: Stability-Aware and Gradient-Efficient Alignment for Reasoning Models [52.48582333951919]
  We propose a dynamic framework designed to enhance alignment reliability by maximizing the Signal-to-Noise Ratio of policy updates.<n>SAGE (Stability-Aware Gradient Efficiency) integrates a coarse-grained curriculum mechanism that refreshes candidate pools based on model competence.<n> Experiments on multiple mathematical reasoning benchmarks demonstrate that SAGE significantly accelerates convergence and outperforms static baselines.
  arXiv  Detail & Related papers  (2026-02-01T12:56:10Z)
• Robust LLM-based Column Type Annotation via Prompt Augmentation with LoRA Tuning [0.749377967268953]
  Column Type.<n>CTA is a fundamental step towards enabling schema alignment and semantic understanding.<n>Existing encoder-only language models achieve high accuracy when fine-tuned on labeled columns.<n>We present a parameter-efficient framework for CTA that trains models over prompt-augmented data.
  arXiv  Detail & Related papers  (2025-12-28T02:04:17Z)
• 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)
• CMT-Bench: Cricket Multi-Table Generation Benchmark for Probing Robustness in Large Language Models [11.167804698594866]
  We present CMT-Bench, a diagnostic benchmark built from live cricket commentary.<n>We find large drops without extractive summaries, monotonic degradation with input length, and consistent accuracy drop under entity-form changes.
  arXiv  Detail & Related papers  (2025-10-20T23:51:28Z)
• SALMAN: Stability Analysis of Language Models Through the Maps Between Graph-based Manifolds [11.373585987937913]
  We propose a unified, local (sample-level) robustness framework (SALMAN) that evaluates model stability without modifying internal parameters or resorting to complex perturbations.<n>Central to our approach is a novel Distance Mapping Distortion (DMD) measure, which ranks each sample's susceptibility by comparing input-to-output distance mappings in a near-linear manner.<n>By demonstrating significant gains in attack efficiency and robust training, we position our framework as a practical, model-agnostic tool for advancing the reliability of transformer-based NLP systems.
  arXiv  Detail & Related papers  (2025-08-23T02:50:55Z)
• Well Begun is Half Done: Low-resource Preference Alignment by Weak-to-Strong Decoding [26.416630784362525]
  Large Language Models (LLMs) require alignment with human preferences to avoid generating offensive, false, or meaningless content.<n>We propose a novel framework, Weak-to-Strong Decoding (WSD), to enhance the alignment ability of base models.<n>We also collect a new dataset, GenAligner, to fine-tune a small-sized Pilot-3B as the draft model.
  arXiv  Detail & Related papers  (2025-06-09T05:21:22Z)
• Towards Continual Learning Desiderata via HSIC-Bottleneck Orthogonalization and Equiangular Embedding [55.107555305760954]
  We propose a conceptually simple yet effective method that attributes forgetting to layer-wise parameter overwriting and the resulting decision boundary distortion. Our method achieves competitive accuracy performance, even with absolute superiority of zero exemplar buffer and 1.02x the base model.
  arXiv  Detail & Related papers  (2024-01-17T09:01:29Z)

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.