Large Language Models are Inconsistent and Biased Evaluators
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_ | 1866917656253169664 |
|---|---|
| author | Stureborg, Rickard Alikaniotis, Dimitris Suhara, Yoshi |
| author_facet | Stureborg, Rickard Alikaniotis, Dimitris Suhara, Yoshi |
| contents | The zero-shot capability of Large Language Models (LLMs) has enabled highly flexible, reference-free metrics for various tasks, making LLM evaluators common tools in NLP. However, the robustness of these LLM evaluators remains relatively understudied; existing work mainly pursued optimal performance in terms of correlating LLM scores with human expert scores. In this paper, we conduct a series of analyses using the SummEval dataset and confirm that LLMs are biased evaluators as they: (1) exhibit familiarity bias-a preference for text with lower perplexity, (2) show skewed and biased distributions of ratings, and (3) experience anchoring effects for multi-attribute judgments. We also found that LLMs are inconsistent evaluators, showing low "inter-sample" agreement and sensitivity to prompt differences that are insignificant to human understanding of text quality. Furthermore, we share recipes for configuring LLM evaluators to mitigate these limitations. Experimental results on the RoSE dataset demonstrate improvements over the state-of-the-art LLM evaluators. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2405_01724 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Large Language Models are Inconsistent and Biased Evaluators Stureborg, Rickard Alikaniotis, Dimitris Suhara, Yoshi Computation and Language Artificial Intelligence 68T50 (Primary) 68T01, 68T37, 91F20 (Secondary) I.2; I.2.7; I.7 The zero-shot capability of Large Language Models (LLMs) has enabled highly flexible, reference-free metrics for various tasks, making LLM evaluators common tools in NLP. However, the robustness of these LLM evaluators remains relatively understudied; existing work mainly pursued optimal performance in terms of correlating LLM scores with human expert scores. In this paper, we conduct a series of analyses using the SummEval dataset and confirm that LLMs are biased evaluators as they: (1) exhibit familiarity bias-a preference for text with lower perplexity, (2) show skewed and biased distributions of ratings, and (3) experience anchoring effects for multi-attribute judgments. We also found that LLMs are inconsistent evaluators, showing low "inter-sample" agreement and sensitivity to prompt differences that are insignificant to human understanding of text quality. Furthermore, we share recipes for configuring LLM evaluators to mitigate these limitations. Experimental results on the RoSE dataset demonstrate improvements over the state-of-the-art LLM evaluators. |
| title | Large Language Models are Inconsistent and Biased Evaluators |
| topic | Computation and Language Artificial Intelligence 68T50 (Primary) 68T01, 68T37, 91F20 (Secondary) I.2; I.2.7; I.7 |
| url | https://arxiv.org/abs/2405.01724 |