2023 is undoubtedly the "explosive year" of artificial intelligence. Since the beginning of this year, the AI whirlwind has swept the world. Not only that, ChatGPT is still evolving at an extremely fast pace, and even the founder and CEO of Qiji Innovation, Lu Qi, bluntly stated that they cannot keep up with the "crazy" speed of the big model era. The first meeting of the 20th Central Finance and Economics Commission held not long ago pointed out that we should grasp the wave of new scientific and technological revolution such as artificial intelligence, adapt to the requirements of harmonious coexistence between man and nature, and build a modern industrial system with integrity, progressiveness and security. According to Guangdong's plan, by 2025, several internationally competitive artificial intelligence innovative industrial clusters will be formed, with a core industry scale exceeding 150 billion yuan and driving the scale of related industries to exceed trillion yuan. Can Guangdong, which is in the first tier in China, rank first in the "Chinese version of ChatGPT"? How to avoid being "hot headed" in the domestic AI big model of "Immortal Fight"? Is it possible to achieve "overtaking on curves" or even "charging straight"? Huang Renxun, founder and CEO of Nvidia, a "first tier" enterprise that invests billions of dollars annually, couldn't conceal his excitement in a speech this year. "Computing is developing at the speed of light." Less than two months after his speech, on the early morning of May 19th, OpenAI, an artificial intelligence laboratory in the United States, released the ChatGPT mobile version application for the first time. The emergence of large models represented by ChatGPT has ushered in a "shining moment" for the AI industry. AI large model refers to a deep neural network model with parameter scales of over millions. After specialized training, such models can perform complex processing and task processing on massive amounts of data. The reason why ChatGPT has strong dialogue and interaction capabilities relies on an AI pre training model called "GPT". Suddenly, like a night of spring breeze, the wind triggered by ChatGPT quickly swept from abroad into China. For a while, the domestic AI large model track can be described as a "divine fight". First, Baidu's big language model "ERNIE Bot" was open for internal testing, and then Alibaba, Shangtang Technology and Kunlun Wanwei released their own big model achievements. IFLYTEK, Tencent, 360, ByteDance and other enterprises are also "rubbing their fists". The "AI Big Model Entrepreneurship Corps", led by the co founder of Meituan Wang Huiwen, the founder of Innovation Works Li Kaifu, the founder of Sogou Wang Xiaochuan and other enterprise executives, also aims at this opening. But competing for the AI big model is not easy. The annual cost of $50 million to $100 million is a ticket to training billions of large-scale models, "said Fang Han, CEO of Kunlun Wanwei. The high training costs and research and development investment of AI large-scale models also make many entrants feel intimidated. In fact, AI large models have both "large-scale" and "pre trained" attributes. Before modeling for practical tasks, it is necessary to pre train on massive universal data in order to significantly improve the universality and generalization of AI. However, the AI large model requires a very large amount of parameters and data. Taking ChatGPT as an example, the parameter quantity of GPT-1 is only 117 million. By the time GPT-3 was released in 2020,