摘要：TMTPOST – The market share of open-source models is set to exceed that of closed-source models after the release of DeepSeek, even

TMTPOST – The market share of open-source models is set to exceed that of closed-source models after the release of DeepSeek, even if the latter’s market share is far more than that of the former, said Larry Shum, a Foreign Member of the U.S. National Academy of Engineering.

“DeepSeek has showcased the triumph of the open-source community. By open-sourcing an extraordinary model, more people have the opportunity to build upon it and achieve even greater things,“ Shum noted.

His speech delved into how open-source models are reshaping the AI landscape, pushing innovation forward by allowing developers to iterate and enhance existing technologies.

Shum also pointed out a shift from the GPT paradigm to the "Reasoner paradigm," which focuses on iterative, task-oriented problem solving, similar to human "slow thinking,” adding that this new approach emphasizes reasoning and correction rather than pre-trained predictions, aligning more closely with real-world problem-solving methods. He stressed that this shift would lead to AI systems that more effectively augment human intelligence rather than replacing it.

The Global Developer Conference (GDC), held in Xuhui, Shanghai from Friday to Sunday, brought together leaders from industries such as finance, healthcare, education, and manufacturing to explore the future of AI and technology. The event focused on the commercialization of large AI models, open-source ecosystems, and multimodal AI technologies.

Prominent speakers included Shum, who is also the Chairman of the Hong Kong University of Science and Technology's Council; Qi Yuan, Director of the Shanghai Artificial Intelligence Laboratory; Jiang Daxin, the CEO of StepAI; and Yang Fan, the co-founder of SenseTime.

The conference focused on the rapid growth and future trajectory of AI technologies, with particular emphasis on large models. During the opening ceremony, Shum delivered a keynote titled "Innovation and Reflections in the Era of Large Models." In his speech, he highlighted the transformative potential of open-source AI and praised its ability to foster global collaboration.

One of the most discussed topics at the GDC was the DeepSeek open-source large model, which has garnered significant attention in recent months. Qi spoke about DeepSeek’s success in revolutionizing AI applications. The model's open-source nature and low inference costs have made it a preferred choice for developers worldwide. DeepSeek’s success was underscored by its remarkable achievement of reaching 100 million users in just seven days, breaking previous records set by models like ChatGPT.

Qi explained the economic implications of such open-source initiatives. While inference costs per token are decreasing, the overall usage is increasing as more people adopt it, Qi said. This phenomenon is consistent with Jevons Paradox, where higher efficiency and lower costs lead to increased consumption.

Qi emphasized that open-source technologies could dramatically accelerate the adoption of AI across industries by reducing entry barriers and fostering trust within the ecosystem.

SenseTime: Advancing AI Agent Applications and Infrastructure

Another major highlight of the conference was SenseTime’s announcement of its open-source development framework, LazyLLM, designed to support multi-agent AI applications. SenseTime, a leader in AI innovation, unveiled its LazyLLM framework on the Wanxiang Platform, which is set to become a key player in the growing AI ecosystem. LazyLLM offers developers a one-stop solution for building and deploying AI agent applications, addressing a wide range of industry needs.

Yang shared insights on the company’s future direction, emphasizing the importance of "large models + large devices" in the development of artificial general intelligence (AGI).

Fan’s keynote at the conference elaborated on the potential of combining cutting-edge AI models with robust computing infrastructure to accelerate AGI’s growth.

Technology is advancing so rapidly that it’s difficult to predict the exact trajectory, but expansion of application scenarios and breakthroughs in domestic computing power are essential for future developments, Fan noted.

SenseTime’s multi-agent AI model, Code Raccoon 2.0, and Office Raccoon 2.0, represent a shift from traditional Copilot applications to more sophisticated agent-driven solutions. Code Raccoon 2.0, for example, enhances programming efficiency by integrating multimodal reasoning and improving software development workflows. Meanwhile, Office Raccoon 2.0 leverages AI’s capabilities to assist with daily tasks, from document processing to complex data analysis, improving productivity for individual and enterprise users alike.

With these advancements, SenseTime has positioned itself at the forefront of AI deployment, catering to over 300 enterprise clients and more than a million individual users. The company also revealed that LazyLLM would facilitate continuous data iteration during the development of AI applications, optimizing their performance across various use cases. By offering a comprehensive suite of development tools and resources, the platform has quickly gained traction across industries such as finance, healthcare, and telecommunications.

As Shanghai continues to foster innovation in AI and technology, the city’s ambitions to become a global hub for AI research and development are becoming more evident. In 2024, Shanghai’s AI industry surpassed 450 billion yuan, with a growth rate of over 7%. By the end of 2025, the city plans to establish a world-class AI industrial ecosystem, with an emphasis on computing power, linguistic data, models, and applications.

Shum, in his keynote, emphasized that China is both a beneficiary and a key contributor to the international open-source community. He pointed to the collaboration and knowledge exchange fostered by open-source technologies as a driving force behind the country’s rapid advancements in AI.

Shum said that China’s focus on innovation, along with its increasing contributions to the open-source community, has strengthened its position as a leader in AI.

The Future of AI: Human-Centered AI and Ethics

As the conference, discussions about the ethical implications of AI took center stage. Shum noted that while AI’s capabilities are expanding, it’s crucial to maintain a focus on "human-centered AI."

This approach emphasizes the augmentation of human intelligence, rather than its replacement, fostering collaboration between machines and people. Shum also stressed the importance of addressing AI ethics to ensure responsible development and deployment of AI technologies.

Jiang echoed these views, explaining that how open-source AI has transformed his company’s products. StepAI, a leader in AI-driven video generation, has benefited from the open-source ecosystem, seeing significant growth in the use of its models across various creative and technical communities. Jiang also announced that StepAI would open-source its image-to-video products in March, further contributing to the growing open-source ecosystem.

The 2025 GDC underscored AI’s potential to drive significant change across industries, from healthcare and finance to education and manufacturing. As companies like DeepSeek, SenseTime, and StepAI continue to advance AI technologies and open-source models, the world stands on the cusp of a new era in AI-powered solutions. The conference’s discussions highlighted the growing importance of collaboration, open-source development, and ethical considerations in shaping the future of AI.

Xiong Jijun, Vice Minister of the Ministry of Industry and Information Technology (MIIT), said that the country will adhere to an innovation-driven and application-oriented approach to create a favorable development environment for developers and provide strong support for achieving new industrialization.

来源：大力财经