Artificial intelligence (AI) has evolved from rule-based symbolic systems to data-driven, neural network–based probabilistic modeling. In particular, the Transformer architecture, proposed in 2017, has become the foundational infrastructure for AI across language, vision, and multimodal domains. This talk reviews the major technological milestones in AI and examines the learning framework of large language models (LLMs) such as ChatGPT—covering pre-training, fine-tuning, and reinforcement learning from human feedback (RLHF)—to illustrate the structural advances in modern natural language processing (NLP). Furthermore, we explore how LLMs are increasingly being applied to scientific discovery and problem-solving, including applications in biomedical research, physics simulations, materials design, and climate modeling, highlighting their potential to accelerate hypothesis generation and interdisciplinary collaboration in the scientific community.