China's AI Breakthrough
ChatGPT-5 was launched on August 8, 2025. However, on August 9, OpenAI relaunched GPT-4o.
The reason is that ChatGPT-5 was criticized for its slow response speed and frequent errors in problem-solving. Moreover, third-party tests showed that GPT-5 had only a negligible leading edge. Behind the diminishing marginal returns of large models, two unavoidable dilemmas have emerged:
1) Data depletion
2) Limitations of computing power cost
The AI wave is reshaping the global economic landscape and the map of national competition with unprecedented force. But when OpenAI reignited the topic with GPT-5 and NVIDIA's GPUs built a computing power barrier, the bottlenecks in AI development have become increasingly clear - high-quality training data is nearly exhausted, there is an energy consumption black hole, and there are also security concerns.
China is facing up to challenges such as the siege of computing power, data depletion, and security governance with an open and inclusive attitude and the blade of algorithm innovation.
China's scientific and technological strength is not confined by high walls.
Meanwhile, China's core competitiveness lies in actively integrating artificial intelligence into the real economy and replicating on a large scale more efficient and innovative new - quality production models.
DeepSeek - V3 has broken through the iron curtain of closed - source models at extremely low cost; domestic chips such as Huawei Ascend and Cambricon are advancing in the huge domestic circular market. China is exploring a new paradigm for AI development with "algorithm breakthrough as the engine, open - source and inclusiveness as the path, industrial application as the foundation, and security and controllability as the bottom line."
Against the above background, this article in Starship Knowledge mainly explores -
● What are the main technical dilemmas currently faced by the development of artificial intelligence?
● What are China's advantages, contributions, and challenges?
This article is partially excerpted and edited from Starship Knowledge: Trends and Prospects of Future AI Artificial Intelligence Technology Development written by Dr. Qian Hongsheng, a professor - level senior engineer in the communication industry and an expert in the ITU communication standard technology group.
The author starts from the technical dilemmas currently faced by the development of artificial intelligence and believes that when Silicon Valley is deeply involved in the controversy of the "data desert" and the shadow of OpenAI's losses hangs over Wall Street, the world's artificial intelligence is accelerating its eastward look -
China is rewriting the rules of AI competition.
source: unsplash
In the midsummer of 2025, at the opening ceremony of the World Artificial Intelligence Conference in Shanghai, Premier Li Qiang of the State Council said:
No matter how technology changes, it should be utilized and controlled by humans and develop in a direction that is beneficial and inclusive. Artificial intelligence should also become an international public good that benefits humanity. We should adhere to openness and sharing, and equal access to intelligence, so that more countries and groups can benefit from it. China is willing to jointly carry out technological research with other countries, increase the intensity of open - source and inclusiveness, and jointly promote the development of artificial intelligence to a higher level.
source: pixabay
An AI revolution driven by algorithm innovation, chip self - sufficiency, and the explosion of application scenarios is rising from the East.
In order to achieve the grand goal of China's AI development, we must have an understanding of the international development path of AI technology in the next few years.
Main Technical Dilemmas Currently Faced by the Development of Artificial Intelligence
● Algorithm bias and dilemmas of large - scale generative models
OpenAI released its latest artificial intelligence model, GPT - 5, in August. According to the OpenAI official website, this is the most powerful artificial intelligence system launched by the organization to date. However, after its launch, it was criticized for being "slow in response and frequently making mistakes in problem - solving."
Currently, whether the algorithms of generative artificial intelligence represented by ChatGPT are scientific and effective has always been the focus of debate among scientists. Experts' criticisms of generative artificial intelligence generally focus on the following aspects: 1. It can produce systematic hallucinations; 2. Limited logical reasoning ability; 3. Data dependence; 4. Biased and unexplainable output information; 5. Security and ethical risks, etc.
● The end of AI computing power growth is the problem of energy consumption
AI systems, especially deep - learning models, need to perform a large number of matrix operations and intensive computing tasks, which all rely on the support of high - performance computers and graphics processing chips (GPUs).
Under the highest load condition, the peak power consumption of a single NVIDIA H100 chip can reach 700W. The total power consumption of 100,000 H100 chips will be close to the total output power of a small power plant.
The energy consumption problem of computing power chips is one of the main technical difficulties restricting the development of AI, and all effective measures must be taken to address it.
● Thoughts triggered by the depletion of large - model training data
Data, as the "fossil fuel" of AI, is becoming increasingly depleted. This trend will force a change in the current pre - training methods of large - scale AI models.
The reasons for data depletion are as follows👇
● Security management and risk control of artificial intelligence
The Collingridge Dilemma is a problem that all innovative countries cannot avoid.
The "Collingridge Dilemma" was proposed by British technology philosopher David Collingridge in 1980.
It mainly states that in the early stage of technology development, due to insufficient understanding of the potential social negative impacts of new technologies, it is difficult to implement effective control measures. When the negative impacts begin to appear, the technology itself has penetrated into all aspects of society, and the cost and difficulty of change have increased significantly.
Today, AI has become the main theme of scientific and technological development, and its Collingridge Dilemma is manifested as follows:
On the one hand, the development of AI technology is bringing about technological changes and social progress;
On the other hand, the current exponential development of AI is deeply trapped in this dilemma - the capabilities of models and the boundaries of applications are expanding rapidly, while the potential social and ethical risks are becoming increasingly difficult to predict and control.
AI technology should be based on security and be restricted by the generally recognized social security and ethical guarantee mechanisms.
The possible security and ethical risks in the development of artificial intelligence generally include the following aspects👇
● Controversy over the open - source of AI system development software
The use of open - source software is not necessarily free. Based on the GPL rules, there are the following charging models👇
1) Charge a certain development cost through distribution;
2) Charge a certain fee by providing guarantee clauses;
3) For software application systems developed on the basis of open - source software, corresponding trademark licensing fees can be charged;
4) In addition, certain fees can be charged for technical support services and personnel training provided for open - source software.
Regarding the issue of open - sourcing AI software code, there are differences in the views and ideas of all parties. This issue will continue to be discussed. The author believes that the opening of the source code of AI system software is an inevitable path in historical development. Although there will be twists and turns, the opening of AI software code is the general trend of social progress and technological development.
source: unsplash
Combining the views of some international experts and scholars and the judgments of well - known international forecasting institutions, Starship Knowledge explores and forecasts several possible development trends of AI technology and applications in the next two to three years as follows.
Welcome like - minded people to criticize, supplement, and improve, and offer suggestions for the smooth development of the country's AI.
Algorithm breakthroughs change the bottlenecks in AI development, and open - source reflects China's open - minded spirit in AI
Over the past year, China has made significant contributions to the leap - forward improvement of global large - scale AI models -
● Open - source models such as China's DeepSeek - V3 (MoE architecture + MLA mechanism) have halved the training cost through innovative algorithms, achieving high performance at low cost (training cost < $6 million), promoting the popularization of global AI.
China's innovation has turned "success through brute force" into "success through ingenious force."
● Open - source breaks monopolies and promotes global innovation. The open - source spirit is becoming China's AI calling card.
On July 23 this year, Alibaba open - sourced its brand - new Tongyi Qianwen AI programming large - scale model, Qwen3 - Coder, which outperforms all open - source models in performance. In actual use, its performance is comparable to that of closed - source models such as Claude and GPT4.1. Chinese enterprises have directly open - sourced the ideal programming large - scale model that everyone has in mind.
More and more Chinese technology enterprises are becoming a microcosm of China's artificial intelligence being open, innovative, and willing to jointly carry out technological research with other countries👇
China's DeepSeek v3 is the highest - ranked open - source licensed model to date. Its training cost is less than $6 million. The DeepSeek - V3 large - scale model adopts a unique innovative architecture with 671 billion parameters, of which only 37 billion parameters are actively running. It has been pre - trained on 14.8 trillion tokens.
Overall, DeepSeek - V3 is still a mature, stable, and continuously evolving mid - term version that has not been completely replaced by the next generation. Its breakthroughs in AI algorithms are mainly reflected in the following aspects👇
In 2024, it broke the monopoly of OpenAI. It has greatly saved pre - training time and cost investment, and reduced the extensive AI development model that relies on the accumulation of AI chips.
It emphasizes the support ability of innovative algorithms for large - scale models at the native level, reduces the dependence of AI computing on hardware requirements, and thus improves the operating efficiency of the entire large - scale AI model. DeepSeek has achieved a breakthrough in cost control, providing new development ideas for the development of global large - scale AI models.
source: unsplash
Chinese technology enterprise Alibaba Cloud has achieved what the American enterprise Anthropic failed to do, directly open - sourcing the ideal large - scale model that everyone has in mind.
In actual use, the performance of the Tongyi Qianwen AI programming large - scale model Qwen3 - Coder is comparable to that of closed - source models such as Claude and GPT4.1.
1) From the perspective of technology diffusion laws and business operations, open - source large - scale models can more quickly achieve a positive cycle of "a surge in the number of users - faster adjustment and adaptation to market demand and user feedback - iteration of large - scale model capabilities - further growth in the number of users."
2) In the case of inconvenience in using overseas programming tools, more and more Chinese developers will eventually turn to an AI programming model that is capable, reliable, and convenient. A huge domestic market will provide a foundation for Alibaba Cloud's artificial intelligence programming.
China's AI has demonstrated a true spirit of open - research, and its research results have made profound contributions to the AI community -
This low - cost and high - efficiency model will greatly promote the popularization of AI technology, enabling more teams and enterprises to afford the development of cutting - edge AI models.
In the history of human development, knowledge is so