The significant significance, challenges and opportunities of driving innovative development in manufacturing industry with artificial intelligence technologies including large language models are explored in this article. The technological system of large language models is analyzed, its basic engineering concepts are clarified, and the pan-LC theory --- a scientific explanation for next token prediction --- is presented. Based on the theory, the causes and consequences of some weird behaviors of large language models are explained, giving a more comprehensive and in-depth understanding of large language models. On the basis, three main requirements for artificial intelligence technologies in manufacturing industry are sorted out, and the core difficulties in the integration of large language models and the artificial intelligence brute-force technology are revealed. A closedness-based solution is proposed for the construction of artificial intelligence systems in manufacturing sectors, such that these systems satisfy the main requirements of specialization, logical validity and knowledge ability, as well as explainability and controllability. Finally, the trend of shifting from "industrial application of new technologies" to "sector innovation driven by new technologies" in the high-quality development of manufacturing industry is discussed briefly.