Humans or LLMs as the Judge? A Study on Judgement Biases
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2024
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866929515467374592 |
|---|---|
| author | Chen, Guiming Hardy Chen, Shunian Liu, Ziche Jiang, Feng Wang, Benyou |
| author_facet | Chen, Guiming Hardy Chen, Shunian Liu, Ziche Jiang, Feng Wang, Benyou |
| contents | Adopting human and large language models (LLM) as judges (a.k.a human- and LLM-as-a-judge) for evaluating the performance of LLMs has recently gained attention. Nonetheless, this approach concurrently introduces potential biases from human and LLMs, questioning the reliability of the evaluation results. In this paper, we propose a novel framework that is free from referencing groundtruth annotations for investigating Misinformation Oversight Bias, Gender Bias, Authority Bias and Beauty Bias on LLM and human judges. We curate a dataset referring to the revised Bloom's Taxonomy and conduct thousands of evaluations. Results show that human and LLM judges are vulnerable to perturbations to various degrees, and that even the cutting-edge judges possess considerable biases. We further exploit these biases to conduct attacks on LLM judges. We hope that our work can notify the community of the bias and vulnerability of human- and LLM-as-a-judge, as well as the urgency of developing robust evaluation systems. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2402_10669 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Humans or LLMs as the Judge? A Study on Judgement Biases Chen, Guiming Hardy Chen, Shunian Liu, Ziche Jiang, Feng Wang, Benyou Computation and Language Adopting human and large language models (LLM) as judges (a.k.a human- and LLM-as-a-judge) for evaluating the performance of LLMs has recently gained attention. Nonetheless, this approach concurrently introduces potential biases from human and LLMs, questioning the reliability of the evaluation results. In this paper, we propose a novel framework that is free from referencing groundtruth annotations for investigating Misinformation Oversight Bias, Gender Bias, Authority Bias and Beauty Bias on LLM and human judges. We curate a dataset referring to the revised Bloom's Taxonomy and conduct thousands of evaluations. Results show that human and LLM judges are vulnerable to perturbations to various degrees, and that even the cutting-edge judges possess considerable biases. We further exploit these biases to conduct attacks on LLM judges. We hope that our work can notify the community of the bias and vulnerability of human- and LLM-as-a-judge, as well as the urgency of developing robust evaluation systems. |
| title | Humans or LLMs as the Judge? A Study on Judgement Biases |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2402.10669 |