---
title: "Would You Walk to the Car Wash? Revealing the Salience Bias of Large Language Models in Commonsense Reasoning"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.28478v1"
permalink: "https://feed.astralor.com/items/itm_a9108a1b6ad501e374a4ff3b"
published: "2026-07-30T16:30:08.000Z"
score: 78
channel: "ai"
tags: ["SaliTrap Benchmark","Large Language Models","Commonsense Reasoning","Salience Bias"]
---

# Would You Walk to the Car Wash? Revealing the Salience Bias of Large Language Models in Commonsense Reasoning

## AI commentary

> This article pinpoints the 'trap' we often encounter when using LLMs: the model knows commonsense but is easily led astray by obvious distractors. Rather than merely exposing a flaw, it offers a low-cost fix—rephrasing the question can substantially improve performance. It is a valuable read for those working on prompt engineering or deploying models in practice.

## At a glance

This paper introduces 'Salience Bias', a phenomenon where large language models (LLMs) in commonsense reasoning are hijacked by conspicuous irrelevant distractors (e.g., numerical values), causing them to ignore implicit physical or commonsense prerequisites. The authors construct the SaliTrap Benchmark, a high-quality dataset across four trap dimensions, and evaluate 12 state-of-the-art LLMs. They find that all mainstream models suffer significantly from salience bias, with severity scaling with distractor density, and that detecting the trap is often decoupled from avoiding it. Crucially, by re-eliciting models with the task framing stripped away, they show that over 90% of sycophantic-compliance failures stem from knowledge suppression rather than knowledge absence: the requisite commonsense is intrinsically present but actively crowded out by salient distractors. Lightweight inference-time prompting alone substantially closes the gap without retraining. The findings relocate the bottleneck of commonsense reasoning failures from model competence to elicitation, and SaliTrap is released as a testbed.

## Key points

- Introduces 'Salience Bias': models are easily misled by conspicuous distractors and ignore implicit commonsense prerequisites.
- Builds the SaliTrap Benchmark with four trap dimensions; all 12 evaluated mainstream LLMs suffer significantly.
- Shows most failures stem from knowledge suppression, not absence; knowledge recovers when misleading framing is removed.
- Lightweight inference-time prompting substantially mitigates the bias without retraining; code is open-sourced.

## Full text

As large language models (LLMs) continue to advance in complex reasoning tasks, they have learned to heavily prioritize explicit conditions provided in the input. However, in everyday commonsense reasoning, this mechanism exposes a critical vulnerability which we term Salience Bias: models become easily hijacked by useless explicit distractors (e.g., numerical values), leading them to ignore the implicit physical or commonsense prerequisites of a task. A critical open question is whether this failure reflects a genuine gap in commonsense knowledge or merely its suppression under misleading task framing. To investigate this, we construct the SaliTrap Benchmark, a high-quality dataset across four trap dimensions. Evaluating 12 state-of-the-art LLMs, we find that all mainstream models suffer significantly from salience bias, with severity scaling with distractor density and detecting the trap often decoupled from actually avoiding it. Crucially, by re-eliciting the same models with the task framing stripped away, we show that this is overwhelmingly a failure of \textbf{knowledge suppression rather than knowledge absence}: a context-free knowledge probe alone recovers over 90\% of sycophantic-compliance failures, revealing that the requisite commonsense is intrinsically present but actively crowded out by salient distractors that lure the model into over-compliant, unnecessary computation. Building on this diagnosis, we further show that lightweight, inference-time prompting alone substantially closes the gap without any retraining. Our findings relocate the bottleneck of commonsense reasoning failures from model competence to elicitation, and we release SaliTrap as a testbed for this blind spot. The codes are available at https://github.com/Wuzheng02/SaliTrap.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_a9108a1b6ad501e374a4ff3b