Do LLMs Overcome Shortcut Learning? An Evaluation of Shortcut Challenges in Large Language Models

TL;DR

This paper introduces Shortcut Suite, a comprehensive evaluation framework that assesses how large language models rely on dataset shortcuts, revealing their tendencies, prompting effects, and overconfidence issues, with implications for improving robustness.

Contribution

The paper presents Shortcut Suite, a novel test suite for evaluating shortcut reliance in LLMs, including new metrics, prompting strategies, and extensive experimental insights.

Findings

Larger LLMs rely more on shortcuts in zero-shot and few-shot settings.

Chain-of-thought prompting reduces shortcut reliance.

LLMs often overconfident and have lower explanation quality on shortcut datasets.

Abstract

Large Language Models (LLMs) have shown remarkable capabilities in various natural language processing tasks. However, LLMs may rely on dataset biases as shortcuts for prediction, which can significantly impair their robustness and generalization capabilities. This paper presents Shortcut Suite, a comprehensive test suite designed to evaluate the impact of shortcuts on LLMs' performance, incorporating six shortcut types, five evaluation metrics, and four prompting strategies. Our extensive experiments yield several key findings: 1) LLMs demonstrate varying reliance on shortcuts for downstream tasks, significantly impairing their performance. 2) Larger LLMs are more likely to utilize shortcuts under zero-shot and few-shot in-context learning prompts. 3) Chain-of-thought prompting notably reduces shortcut reliance and outperforms other prompting strategies, while few-shot prompts generally underperform compared to zero-shot prompts. 4) LLMs often exhibit overconfidence in their predictions, especially when dealing with datasets that contain shortcuts. 5) LLMs generally have a lower explanation quality in shortcut-laden datasets, with errors falling into three types: distraction, disguised comprehension, and logical fallacy. Our findings offer new insights for evaluating robustness and generalization in LLMs and suggest potential directions for mitigating the reliance on shortcuts. The code is available at \url {https://github.com/yyhappier/ShortcutSuite.git}.