The research paradigm of artificial intelligence is undergoing a profound transformation. At the recent 2026 Zhongguancun Forum in Beijing, Yang Zilin, founder of Moonshot AI, a leading company in large models, delivered a keynote speech, raising a key issue that has drawn industry attention: AI development is entering the third phase of "AI-led research."
In his speech, Yang Zilin pointed out that starting from 2026, the way of researching in the field of artificial intelligence will undergo a qualitative leap. In the past, the iteration of AI models highly relied on the rule design and fine-tuning by human researchers. Now, this model is being broken. With technological advancement, AI systems have begun to play the role of "leader" in the research process, accelerating the entire R&D process significantly through deeper self-evolution and self-optimization. In this new stage, open-source models have gradually established industry standards and become the core driving force for technology diffusion and innovation.
The theme of this Zhongguancun Forum is "Deep Integration of Scientific and Technological Innovation with Industrial Innovation." It not only focuses on breakthroughs in technology itself but also pays attention to how AI empowers various industries. According to the information, the forum lasts five days and includes more than 100 events across five major sections. Thousands of industry experts and business guests gathered on-site, building a broad platform connecting technology research and development with industrial application through more than 20 high-density technology对接 activities and over 500 technology projects' roadshows.
Yang Zilin's view not only outlines the technical roadmap of AI evolution but also sends a clear message to developers and enterprises: as AI begins to possess the ability of "leading research," the efficiency and depth of large model training will experience a new boom, and the barriers and paradigms of AI development will be restructured accordingly.
