Researchers have conducted an enlightening comparison between Gemini Pro and GPT-4V, specifically in the realm of education. The study focused on visual question answering (VQA) to evaluate the models’ ability to automatically score text-based rubrics and student-drawn scientific models. The use of NERIF (Notation-Enhanced Rubrics for Image Feedback) prompts further enhanced the process.

Key findings from the study include:

• GPT-4V’s lead in scoring accuracy, according to quantitative analysis.
• Higher Quadratic Weighted Kappa for GPT-4V, indicating better consistency in assessment.
• Qualitative analysis suggests GPT-4V is more adept at handling fine-grained texts within images.
• When image input size was reduced, Gemini Pro’s performance still did not match GPT-4V’s efficiency.

This research underscores the importance of GPT-4V in educational applications where multimodal data interpretation is essential. It’s a significant step forward in AI-driven education. GPT-4V’s nuanced understanding of complex visual and textual data could usher in a new era of automated assessment tools in academic settings. Further investigations could extend these findings to other disciplines or explore the integration of such AI models into broader educational platforms.