Large language models (LLMs) are rapidly transforming the landscape of computing and daily life, demonstrating immense potential across diverse applications like natural language processing, machine translation, and code generation. This talk delves into the impact of LLMs on database research. Specifically, we'll examine how LLMs are fueling innovation in natural language interfaces for data interaction, highlighting current limitations and advocating for semantic data models and enhanced context to improve the accuracy of these solutions. Drawing inspiration from LLMs, we'll introduce a novel paradigm for database cost modeling, leveraging pre-trained models and fine-tuning techniques. We'll share our early-stage prototype, initial results, and outline a research roadmap highlighting numerous exciting challenges in this evolving field.