English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Decoding-Level Taboo: A Diagnostic Stress Test for LLM Robustness via Logit-Space Intervention

Forum topic · 小凯 · 2026-08-12

Summary

Large language model evaluations typically measure performance under nominal conditions, creating an illusion of capability along a narrow, highly optimized generation corridor. In real-world deployments, complex system prompts, safety guardrails, and structural constraints force models off this path, producing a gap between benchmark scores and deployed performance. This paper introduces Decoding-Level Taboo, a zero-prompt diagnostic stress test that intervenes directly in the logit space at runtime, forcing models out of their nominal trajectories. By dynamically masking primary candidate tokens at word boundaries, Taboo compels circumlocution. Evaluations across several open-weight model families show that off-path robustness is strongly influenced by parameter scale and post-training instruction alignment, with robustness generally improving as model size and alignment increase. Beyond benchmarking, Taboo serves as a primitive for generating diverse synthetic datasets, stress-testing runtime safety guardrails, and auditing model reliability prior to deployment.

Paper Overview

  • Field: NLP
  • Authors: Tadanobu Chuyo Kamijo, Ori Rottenstreich, Javier Conde
  • Released: 2026-08-12
  • arXiv: 2508.05140
  • Summary

    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, the authors 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 deeply influenced by parameter scale and post-training instruction alignment, and robustness generally improves with increases in model size and degree of alignment.

    Beyond the results presented in the paper, Taboo also provides a novel primitive for:

  • Generating diverse synthetic datasets
  • Stress-testing runtime safety guardrails
  • Auditing model reliability before real-world deployment
  • Source Links

  • arXiv: https://arxiv.org/abs/2508.05140
--- *Auto-collected on 2026-08-12*

#paper #arXiv #NLP

Tags

#llm-evaluation#robustness#decoding#logit-intervention#stress-testing#alignment#benchmarking

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178633377