Data+AI: LLM4Data and Data4LLM
Guoliang Li, Jiayi Wang, Chenyang Zhang, Jiannan Wang · 2025
Large language models (LLMs) have revolutionized traditional data management systems by their natural language processing capabilities (e.g., understanding, reasoning, generation, few-shot and zero-shot learning), while data management techniques play a vital role in optimizing AI models (e.g., data preparation, data efficient training and inference). This establishes a two-way relationship where AI enhances data management, and data boosts AI capabilities. This tutorial covers recent advancements and challenges at this intersection, focusing on LLM4Data and Data4LLM over four parts. Initially, we discuss the background and motivations for integrating data and AI, and present the challenges and principles of LLM4Data and Data4LLM. We then explore how LLMs optimize data management by offering practical applications like managing unstructured data. We also examine how data management optimizes AI, particularly in training and fine-tuning LLMs, showcasing techniques for data preparation and inference. Finally, we provide open challenges and future research directions, aiming to drive innovation in both fields.