In the drama "The Bright Sword," Li Yunlong uses the phrase "brave wins when meeting on a narrow path" to inspire morale, emphasizing the importance of courageously facing strong enemies rather than just pursuing victory or defeat. Last year, the rise of large-scale general models represented by ChatGPT sparked a new wave of industrial revolution, even earning praises from Bill Gates. In 2024, global AI competition has entered a "systemic" competition stage, with the main battlefield focusing on the struggle for large-scale general models.
As the saying goes, only when the tide goes out do you discover who's been swimming naked. After undergoing the brutal baptism of the "Hundred Models Battle" in vertical domains in 2023, China's domestic large-scale general models have entered the "elimination round" in 2024. Those able to "stay" in the arena possess hardcore capabilities as "high-energy players." For China to lead the global development of large-scale model industry, it must seize the initiative and leadership in the industrial chain. How to plan the overall situation at the top policy level is crucial for winning this "tough battle."
During this year's National People's Congress (NPC) and Chinese People's Political Consultative Conference (CPPCC) sessions, Liu Qingfeng, the chairman of iFlytek, who has been a National People's Congress delegate for 22 years and has been deeply involved in the AI field, proposed a systematic solution that resonated with the industry. He suggested focusing on developing autonomous and controllable large-scale models and formulating a national "General Artificial Intelligence Development Plan."
Liu Qingfeng believes that in this current competition, where large-scale models represent general artificial intelligence, only by building a domestically controlled and autonomous computational infrastructure can a country control its own destiny and narrow the international gap. Taking the example of China's first nationwide open large-scale model, the iFlytek Starfire Model, which is based on domestically produced computational power, it is estimated that within six months, it could reach the current best level of GPT-4/4V. However, with the release of GPT-5, this gap may widen to over a year. If resources are organized efficiently in terms of computational power, data, and model training, there is hope that this gap can be narrowed to a comparable level within 1-2 years. At the same time, China is also forming a comparative advantage in leading internationally in areas such as speech and medical large-scale models.
From a global perspective, amidst the current decoupling of technology between China and the United States, the U.S. is employing a dual strategy, implementing technology isolation from hardware such as chips to software like artificial intelligence. Even companies like iFlytek are inevitably included in entity lists.
In the face of this reality, Liu Qingfeng believes that we must both acknowledge the gap and accelerate our catching up. He suggests drawing inspiration from the spirit of China's first atomic bomb explosion 60 years ago, leveraging the spirit of "two bombs, one satellite," and the advantage of "concentrating our efforts to accomplish great things," to systematically propose suggestions for competition in the era of general artificial intelligence. General artificial intelligence is the "atomic bomb" of the digital age, and China must have its own.
Liu Qingfeng suggests focusing on the "main battlefield" of general large models, integrating resources from all parties and increasing investment continuously. He presents nine suggestions around dimensions such as independent and controllable general large model research and development, forward-looking layout of source technology, construction of computing power ecosystem, open and shared high-quality data, establishment of scientific evaluation standards, talent cultivation, legislation, and ethical humanities.
The development of artificial intelligence requires policy support. Events like the outbreak of general artificial intelligence this time propel the application of artificial intelligence to a higher dimension. Referring back to 2017, when the State Council issued and implemented the "New Generation Artificial Intelligence Development Plan" (hereinafter referred to as the "Plan"), elevating artificial intelligence to a strategic technology leading the future, and as a major strategy to enhance national competitiveness and safeguard national security. This "Two Sessions," Liu Qingfeng suggests that based on the 2017 version of the "Plan," we should design around the shortcomings that need to be addressed in the development of China's general artificial intelligence, focusing on dimensions such as the construction of an independent and controllable computing power ecosystem, and the open sharing of high-quality data, to systematically formulate the national "General Artificial Intelligence Development Plan."
As the ancients often said, to learn from the barbarians to master their skills, those who are not good at learning from the barbarians will be mastered by the barbarians. There is a gap between China and the United States in the field of large models. The ban on the sale of high-end graphics cards like NVIDIA in China has also led to a sharp rise in the substitution of domestic computing power. Although China is constrained and has shortcomings, it also has the conditions to catch up vigorously. For example, the first domestic computing power platform "Feixing No. 1" launched by iFlytek in cooperation with Huawei provides a good demonstration. Through the integration of software and hardware capabilities and the construction of network systems, it fills the current gap in single-chip computing power and lays a foundation for building an independent and controllable general artificial intelligence computing base. Liu Qingfeng believes that China is expected to become the second pole of global wisdom emergence, providing the world with a second choice.
During this year's "Two Sessions," "artificial intelligence+" and "new quality productivity" became high-frequency hot words and hit the hot search. Behind this is the reflection that only by promoting the integration of artificial intelligence and industry can we accelerate the formation of new quality productivity, play the leading role of technological innovation, and help the country achieve high-quality development.
Sixty years ago, China successfully conducted its first atomic bomb test in the Lop Nur region, not only breaking through Western technological blockade but also forging a sharp sword to defend national dignity with its own strength. Today, thanks to the Chinese characteristic advantage of "concentrating our efforts to accomplish great things," the country will surely achieve self-reliance and self-improvement in the era of general artificial intelligence. As a representative of the national team in the era of general artificial intelligence, iFlytek firmly innovates in source technology, dares to face the gap, and accelerates catching up while pursuing ChatGPT at the international first-class level!
