AI Takes the Wheel: Yang Zhilin on the Third Stage of Large Model Training

The way we research artificial intelligence is undergoing a seismic shift. At the 2026 Zhongguancun Forum in Beijing, Yang Zhilin, founder of Moonshot AI, dropped a bombshell: AI development has entered its third phase—one where AI itself takes the lead in research.

From Human-Led to AI-Led Research

For years, training large language models was a painstaking process. Human researchers would meticulously design rules, tweak parameters, and fine-tune every aspect. But starting this year, that model is being turned on its head. According to Yang, AI systems are now capable of driving their own evolution. They can self-optimize, learn from their mistakes, and accelerate the entire R&D cycle—all without constant human intervention.

“We’re moving from a world where humans tell AI what to do, to one where AI tells us what’s possible,” Yang said during his keynote. This shift isn’t just about speed; it’s about depth. AI can explore solution spaces that humans might never consider, leading to breakthroughs that were previously out of reach.

Open Source: The New Standard

A key driver of this transformation is the rise of open-source models. Yang noted that open-source frameworks have become the de facto standard for AI development. They allow researchers and companies to build on each other’s work, accelerating innovation across the board. “Open source is no longer just an option—it’s the engine of progress,” he emphasized.

This democratization of AI means that even small teams can now compete with tech giants. The barriers to entry are crumbling, and the pace of innovation is only going to increase.

The Forum: Where Tech Meets Industry

The Zhongguancun Forum, themed “Deep Integration of Scientific and Technological Innovation with Industrial Innovation,” was the perfect backdrop for Yang’s announcement. Over five days, the forum hosted more than 100 events, including over 20 high-density tech matchmaking sessions and 500 project roadshows. Thousands of industry experts and business leaders gathered to bridge the gap between lab research and real-world applications.

Yang’s message resonated deeply with the audience. For developers and enterprises, it was a clear signal: the rules of the game are changing. If you want to stay ahead, you need to embrace AI-led research and open-source collaboration.

What This Means for the Future

Yang’s vision paints a picture of an AI landscape that is more dynamic, more efficient, and more inclusive. With AI leading the charge, the next generation of large models will be trained faster, smarter, and with less human effort. But this also raises questions: What happens to the role of human researchers? How do we ensure that AI-led research remains aligned with human values?

For now, the industry is buzzing with excitement. The third stage of AI development is here, and it’s bringing a whole new set of possibilities.

Key Points

AI research has entered a third phase where AI systems lead the research process.
Open-source models are becoming the industry standard, driving innovation and lowering barriers.
The 2026 Zhongguancun Forum highlighted the integration of AI with industrial applications.
Yang Zhilin’s keynote signals a paradigm shift for developers and enterprises.

AI Takes the Wheel: Yang Zhilin on the Third Stage of Large Model Training