TPCD: Tone-Pressure Contrastive Decoding and the Label-Free Gating Bottleneck in Vision-Language Models
arXiv:2607.26536
The paper introduces Tone‑Pressure Contrastive Decoding (TPCD), which subtracts logits from high‑pressure prompts from those of neutral prompts to reduce commitment bias in vision‑language models, and evaluates simple gating mechanisms that preserve correct answers while lowering attack success rates.
Abstract
High-pressure prompts can push vision-language models (VLMs) into unsupported commitments, such as reading illegible text, reporting indeterminate times, or affirming absent objects. This paper asks whether the pressure-induced distribution itself can serve as a contrastive-decoding negative branch. Tone-pressure contrastive decoding (TPCD) subtracts logits produced under a high-pressure instruction from logits produced under a safe neutral instruction. On the 800-example tone-matters benchmark, LLaVA-1.5-7B under pressure reaches 66.75% attack success rate (ASR); safe neutralization reduces ASR to 9.88%; full TPCD reaches 0.50% but collapses positives to 15.56%. A benchmark-specific task-prior/disagreement gate preserves measured positive accuracy (54.44%) while lowering ASR to 1.63% on LLaVA. Treating this LLaVA analysis as the design split, full negative and matched-positive held-out runs on GLM-4.6V and Llama-3.2-Vision show that simple gates can improve over safe neutralization, with sensitivity analyses bounding the weak time-positive subtask. A category-prior-free answer-disagreement router reduces held-out aggregate ASR to 6.93%, improving over both safe neutralization (10.98%) and branch disagreement (9.67%) while matching branch disagreement's 79.94% positive accuracy, although it remains post-hoc and surface-form based. We conclude that pressure is a useful probe of commitment bias and a viable mitigation signal, but the current gates are not yet independently validated grounding-aware detectors.