Exploring Boundary of GPT-4V on Marine Analysis: A Preliminary Case Study

Large language models (LLMs) have demonstrated a powerful ability to answervarious queries as a general-purpose assistant. The continuous multi-modallarge language models (MLLM) empower LLMs with the ability to perceive visualsignals. The launch of GPT-4 (Generative Pre-trained Transformers) hasgenerated significant interest in the research communities. GPT-4V(ison) hasdemonstrated significant power in both academia and industry fields, as a focalpoint in a new artificial intelligence generation. Though significant successwas achieved by GPT-4V, exploring MLLMs in domain-specific analysis (e.g.,marine analysis) that required domain-specific knowledge and expertise hasgained less attention. In this study, we carry out the preliminary andcomprehensive case study of utilizing GPT-4V for marine analysis. This reportconducts a systematic evaluation of existing GPT-4V, assessing the performanceof GPT-4V on marine research and also setting a new standard for futuredevelopments in MLLMs. The experimental results of GPT-4V show that theresponses generated by GPT-4V are still far away from satisfying thedomain-specific requirements of the marine professions. All images and promptsused in this study will be available athttps://github.com/hkust-vgd/Marine_GPT-4V_Eval