AI summary
Summary of article:
The AI dialogue at the Wuzhen Summit focused on the open source ecosystem. Representatives from DeepSeek, Alibaba and MiniMax advocated using open source to accelerate innovation and AGI development, and also reminded that security governance needs to be strengthened in medical and other scenarios.
The main stage of the 2026 World Internet Conference Wuzhen Summit was held on the afternoon of the 3rd.artificial intelligence “Wuzen Dialogue” invites the world’s leading technology companies, experts and scholars to interact and discuss building an open source ecosystem for artificial intelligence (AI). Gao Xi, a research engineer at DeepSeek, said that AI is developing rapidly, and the real competitor of the industry is not other big model companies, but time. Open source can help gather more power and accelerate the advent of general artificial intelligence (AGI).
In addition to Gao Xi, the organizer also invitedalibaba Liu Deiheng, leader of the Qianwenda language model project of ATH Business Group, Feng Wen, leader of the Minimax open platform, and John Higgins CBE, president of the International Artificial Intelligence Governance Association, and others participated.
Gao Xi said that outsiders often ask why DeepSeek made the results of its hard work and development open source. “Actually, the answer is very simple.” This is for AGI, so open source is an inevitable choice. AI has made rapid progress in recent years. From initial processing of simple tasks to having deep reasoning abilities, answering exam and competition questions, aiding program development and solving complex mathematical problems, the pace of technological progress is accelerating year by year. He believes that industry players should compete with the times to promote AGI rather than hinder competitors by setting up technological barriers, as this may also delay the technological progress of the entire industry.
Gao Xi said that if a new architecture is discovered during the research and development process that can reduce the cost of model training and inference, or a new method that can improve the model effect, it should be opened as soon as possible for other researchers to test, verify and implement, which will accelerate technological breakthroughs. He also pointed out that it is not most important which company ultimately implements AGI. The main thing is whether the technology can be developed as quickly as possible so that more people can benefit.
Liu Deiheng said that the Qianwen team has continued the open source model, after completing the pre-training of the first generation 7B model more than three years ago. Currently, it has more than 460 models open source, with cumulative downloads of more than 4 billion times worldwide. He believes that open source has two important meanings, one is to accelerate AI innovation, and the other is to allow more people to use AI technology.
Liu Deiheng said that open source gives developers, researchers and students around the world the opportunity to participate in model innovation. It is no longer just a few companies that master the core technology. In particular, smaller models can reduce the computing power and cost required for research, allowing the academic community to conduct experiments and feed research results back into the open source ecosystem. He also said that open source models can help meet the needs of small languages, specialized industries, and specific hardware and allow companies to deploy models locally and maintain autonomy in technology selection and data security.
Feng Wen said that in addition to open source models of various sizes and modes, large model manufacturers will also open up experimental experience, technological changes, equipment and related information during the model development process, so that developers can refer to the research and development process and experience to further promote technological innovation.
However, the expanded application of the open source model also brings security and governance issues. Higgins pointed out that open source is not inherently less secure than closed models, but the open source ecosystem needs to establish a common assessment, testing and certification mechanism so that different countries and markets can rely on relevant standards. When open source models are applied in fields such as hospitals, aviation, railways, and automobiles, we cannot rely solely on the spontaneity of the developer community. We need systems and related institutions to provide reliable evaluation mechanisms.
Summary FAQ
-
Gao Xi said that the real competition of the industry is the times, not other big model companies. If new architectures and new methods can be made open source as soon as possible so that more researchers can test and verify them, then more efforts can be made to accelerate the advent of AGI.
-
Liu Deiheng pointed out that after completing the pre-training of the first generation 7B model, the Qianwen team continued to open source. So far, it has open sourced more than 460 models, been downloaded more than 4 billion times worldwide, and has been used to promote more people to participate in AI innovation.
-
Higgins believes that open source is no less secure than the closed model, but in high-risk areas such as hospitals, aviation, railways, and automobiles, joint assessment, testing and certification standards are still needed and cannot be maintained by the community alone.
