Since the launch of ChatGPT-4 in March this year, leading domestic science and technology enterprises have intensively launched large models of artificial intelligence. Baidu "ERNIE Bot", Alibaba Cloud "Tongyi Qianwen", Huawei "Pangu", 360 "Zhinao", Kunlun Wanwei "Tiangong", JD "Lingxi", iFLYTEK "Spark", Tencent "Hunyuan", Shangtang "Ririxin" and other large models have appeared successively, showing a trend of blooming and rapid development. As of early October 2023, there have been over 200 publicly available AI models in China. Currently, big models are becoming an important force in promoting a new round of technological innovation, industrial upgrading, and productivity leap. With cutting-edge technology driving the upgrading of the entire industry chain, big models may be ubiquitous in the future, from fields to production lines, from laboratories to convenience stores. "Compared with the early AI models, this year China's big models have achieved a qualitative leap in the number of parameters, improved the ability to model complex tasks as a whole, had stronger learning ability and generalization, and had a higher level of cognitive interaction ability." Chen Xiaohua, executive director of the Yuanzhou Industrial Collaborative Innovation Center in the Science Park of Beijing University of Posts, said. However, behind the rapid development of large models, there are also a series of problems that urgently need to be solved. Firstly, there are doubts about the large model casing. Recently, from the falsification of Google's newly released demo video of the AI model Gemini to the ByteDance, it was revealed that some engineers used OpenAI technology to develop their own big language models, which violated the terms of OpenAI service, so they were suspended from their accounts... The fierce competition of the "100 model war" and the chaos caused by it also made the industry put forward new topics on the evaluation standards and unified norms of big models. Due to the complexity of large models and the diversity of application scenarios, evaluating their performance and effectiveness has become a highly challenging issue in the next stage. Therefore, establishing a scientific, reasonable, and comprehensive evaluation system is crucial for the application of large models. It can promote technological progress, promote industrial development, and ensure the compliance and social benefits of technological applications. Secondly, with the continuous adjustment of large model parameters by internet and technology companies, the demand for digital transformation in traditional enterprises has increased, leading to an increase in computing power demand. However, due to the difficulty in obtaining high-end chips, the research and development costs of large models have further increased. Professor Lin Zhouchen, Vice Dean of the School of Intelligence at Peking University, proposed the necessity of developing and utilizing large models more efficiently through collaborative sharing of resources and data, as well as innovative training methods and reward mechanisms. He proposed encouraging companies to share data and computing power through shares and special reward mechanisms, where shares can be allocated based on financial contributions or model accuracy contributions. Individuals can also provide data or participate in the training process to own shares through data quality and fine-tuning effects. This may help small and medium-sized enterprises move away from the two mountains of "cost" and "technology" in front of them. In terms of algorithms, there is no fundamental difference between domestically produced large models and international advanced levels. Most of them focus on basic algorithms such as deep neural networks, attention mechanisms, and manual tuning for model structure or local algorithm optimization. However, there is a certain gap in computing power, data, and other aspects compared to international advanced levels. IResearch Consulting