GPT-4 as an Effective Zero-Shot Evaluator for Scientific Figure Captions

Keyphrases

• Figure Captions 100%
• Scientific Figures 100%
• Zero-shot 100%
• GPT-4 100%
• Human Evaluation 66%
• Large Language Models 66%
• Relevant Context 33%
• Computer Science 33%
• ArXiv 33%
• Informatics 33%
• Evaluation Dataset 33%
• Human Judgment 33%
• Kendall Correlation 33%
• Correlation Score 33%
• Student Ranking 33%
• Academic Expertise 33%
• GPT-3 33%
• Automatic Evaluation 33%
• Machine-made 33%

Social Sciences

• Generative Pre-Trained Transformer 4 100%
• Large Language Model 66%
• Computer Science 33%
• Authors 33%

Computer Science

• Generative Pre-Trained Transformer 3 33%