---
title: "鲁棒性的假象：聚合准确率掩盖任务无关上下文下的预测翻转"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.12963v1"
permalink: "https://feed.astralor.com/items/itm_df8c3bcab736af1a5fcf86b9"
published: "2026-07-14T17:01:12.000Z"
score: 78
channel: "ai"
tags: ["伪词","大语言模型","尾部风险","聚合准确率","预测稳定性","鲁棒性评估"]
---

# 鲁棒性的假象：聚合准确率掩盖任务无关上下文下的预测翻转

## AI 点评

> 这篇论文戳破了LLM鲁棒性领域的一个常见直觉：整体准确率稳定 ≠ 模型可靠。它用简洁的实验证明了聚合指标可能隐藏严重的个体样本风险，对AI工程师和评估研究者都是当头棒喝。关注部署安全的朋友尤其值得一读。

## 速览

最新研究表明，最先进的大语言模型在聚合层面上对任务无关上下文表现出鲁棒性，整体准确率几乎不变。然而，这种稳定性掩盖了个体样本上的预测翻转：即使是随机组合字符的无意义伪词，也能使少数样本的预测发生显著变化，既有恶化也有改善。这种双面效应广泛存在于多种模型和数据集中，但受影响的样本高度依赖特定模型。上下文类型、长度、测试时计算量和模型发展阶段均会影响不稳定性。该发现揭示了聚合指标掩盖的尾部风险，呼吁对语言模型进行逐样本可靠性评估。

## 关键点

- 聚合准确率稳定，但个体样本预测可能存在显著翻转。
- 无意义的伪词也能引发预测变化，且影响具有双面性。
- 受影响的样本因模型而异，具有高度模型特异性。
- 上下文类型、长度、测试时计算和模型发展阶段均会影响不稳定性。

## 全文

As large language models (LLMs) grow more capable, they are increasingly deployed in context-rich settings where task inputs are often accompanied by long, partially irrelevant context. In a controlled setting, we find that state-of-the-art models often appear robust to task-irrelevant context at the aggregate level: prepending it to benchmark questions causes little change in overall accuracy. This aggregate stability, however, masks significant per-example instability. Even semantically meaningless pseudo-words, formed by randomly combining characters, can markedly shift model predictions on a small fraction of examples, degrading performance on some while improving it on others. This two-sided effect holds consistently across a wide range of models and datasets, yet the affected examples are largely model-specific. We further show that this instability is modulated by context type, context length, test-time compute, and model development stage. Together, our findings reveal context-induced tail risks concealed by aggregate accuracy, motivating per-example reliability evaluation of language models.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_df8c3bcab736af1a5fcf86b9