English

An Empirical Analysis on Large Language Models in Debate Evaluation

Computation and Language 2024-06-05 v2 Artificial Intelligence

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.

Keywords

Cite

@article{arxiv.2406.00050,
  title  = {An Empirical Analysis on Large Language Models in Debate Evaluation},
  author = {Xinyi Liu and Pinxin Liu and Hangfeng He},
  journal= {arXiv preprint arXiv:2406.00050},
  year   = {2024}
}

Comments

Accepted to ACL 2024 main

R2 v1 2026-06-28T16:48:55.645Z