An Empirical Analysis on Large Language Models in Debate Evaluation

TL;DR

This paper evaluates large language models like GPT-3.5 and GPT-4 in debate assessment, revealing they outperform humans and existing methods but exhibit biases such as positional, lexical, and end-of-discussion biases affecting their judgments.

Contribution

It provides the first comprehensive analysis of LLMs' debate evaluation capabilities and uncovers inherent biases influencing their performance.

Findings

LLMs outperform humans and state-of-the-art methods in debate evaluation.

Identifies positional, lexical, and end-of-discussion biases in LLM judgments.

Biases are influenced by prompt design and label verbalizer choices.

Abstract

In this study, we investigate the capabilities and inherent biases of advanced large language models (LLMs) such as GPT-3.5 and GPT-4 in the context of debate evaluation. We discover that LLM's performance exceeds humans and surpasses the performance of state-of-the-art methods fine-tuned on extensive datasets in debate evaluation. We additionally explore and analyze biases present in LLMs, including positional bias, lexical bias, order bias, which may affect their evaluative judgments. Our findings reveal a consistent bias in both GPT-3.5 and GPT-4 towards the second candidate response presented, attributed to prompt design. We also uncover lexical biases in both GPT-3.5 and GPT-4, especially when label sets carry connotations such as numerical or sequential, highlighting the critical need for careful label verbalizer selection in prompt design. Additionally, our analysis indicates a tendency of both models to favor the debate's concluding side as the winner, suggesting an end-of-discussion bias.