Can Large Language Model speak Art?
Recent advances in generative multimodal Large Language Models (LLMs) offer new opportunities for Digital Art History, yet their potential remains largely unexplored. The aim of this project is to investigate the capabilities and limitations of multimodal LLMs in addressing art historical questions with images and texts. With this project, I aim to evaluate what general-purpose models know about Art History, focusing on their bias and hallucination. Then, to explore how domain adaptation techniques, e.g., Retrieval-Augmented Generation (RAG) or continual pretraining can enhance model performances by integrating curated art history data. The study proposes a structured evaluation methodology and aims to develop flexible tools to support complex, large-scale art historical research.
