Author
Pinxin Liu
Mentors
Hangfeng He and Chenliang Xu
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. 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 Positional biases, Lexical biases, Order biases, etc. LLMs Evaluation: Pro side is the winner to prompt design. We also uncover a lexical bias 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. Our findings also reveal that Prompt engineering strategies are not effective in alleviating biases.