Chinese AI Giants Debate AGI

On January 10, the AGI-Next summit, initiated by Tsinghua University’s Basic Model Beijing Key Laboratory and Zhipu, convened industry leaders including Zhipu founder Tang Jie, Kimi founder Yang Zhilin, Tencent’s Chief AI Scientist Yao Shunyu, and Alibaba’s Qwen technology head Lin Junyang. The summit addressed the core challenges in the development of AGI (Artificial General Intelligence).

Currently, China demonstrates strong competitive power in the AGI innovation wave. Technologically, large model capabilities continue to break through, and the open-source ecosystem is thriving. In the industry, Zhipu and MiniMax have listed on the Hong Kong Stock Exchange, and Kimi has secured significant financing, accelerating the industrialization and capitalization of AI. On the policy front, the series of national standards for “Artificial Intelligence Large Models” has been officially implemented, pushing the industry into a new phase of “scientific authority and unified standards.”

As technology rapidly iterates and the ecosystem matures, the summit focused on key questions: What will the next paradigm of AGI be? What core development directions will future large models take? How can China establish its coordinates and pathways within this context?

Exploring the Path for Next-Generation AI Technology

“Can we lead a new paradigm? This may be the only issue that China’s AI industry needs to solve today. In other areas, whether in business, industrial design, or engineering, we have reached a top level to some extent,” said Yao Shunyu.

This cautious optimism reflects a clear awareness among China’s AI academia and industry regarding their positioning. Tang Jie stated, “We have achieved some successes in open-source, and some may feel that China’s large models have ‘reached the peak,’ but the real answer is that the gap with the world’s top level has not yet narrowed.”

Yang Zhilin believes that AGI/ASI is not just an ordinary tool; it is an amplifier that can change human civilization. It can address problems that we cannot solve alone today, such as issues in healthcare, energy, and climate change. We should not abandon the research and development of AGI/ASI, as doing so means giving up the pursuit of the upper limits of human civilization.

On the basis of recognizing the gaps and missions, the guests discussed possible paths for the next generation of AI paradigms. According to Tang Jie, 2025 will merely be an “adaptation period” for multimodal capabilities. The future’s key will be whether models can integrate visual, auditory, tactile, and other information in a native and unified manner, forming a holistic perception capability, which is central to achieving a leap in multimodal abilities.

Deeper challenges lie in memory, continuous learning, and self-awareness. “Current models lack a layered memory structure. How to expand personal memory into a long-term, reflective human collective knowledge base, and explore the model’s ‘self-awareness’ based on that, will be the most challenging and worthwhile direction to invest in for the next stage,” Tang Jie stated.

Regarding specific breakthrough directions, Yao Shunyu highlighted that “autonomous learning” is an important signal. Some teams are already attempting to learn in real-time using the latest user data, but the biggest challenge they face may be a lack of imagination.

Lin Junyang proposed two core directions: first, the autonomous evolution of AI, exploring how models can avoid becoming “dumber” and achieve self-updating through human interaction; second, enhancing AI’s proactivity, enabling it to think and act independently.

Tang Jie is confident that a paradigm shift will occur by 2026, driven by two major trends: first, the gap in computing power and innovation between academia and industry is narrowing, with universities now capable of incubating innovative seeds; second, the development of large models faces efficiency bottlenecks, with diminishing marginal returns from data and computing power accumulation. The industry needs to pursue higher “intelligent efficiency,” achieving greater intelligence increments with less input, which will compel the birth of new paradigms.

What is the Real Path for AI Agents?

The development path of AI Agents, as an important carrier of AI applications, became another core topic of the summit. The industry anticipates that 2026 may be a critical year for AI Agents to create substantial economic value.

“Compared to the model itself, the Agent is a broader concept, capable of autonomously using tools like a human to complete tasks in an environment. This is the direction AI should take,” Lin Junyang stated.

AI scientist and member of the Royal Society of Canada, Yang Qiang, indicated that the future direction is to enable large models to define their own goals and make plans, becoming an “endogenous native system.”

Lin Junyang believes that the true capability of a general Agent lies in solving the “long tail problem.” Addressing common head needs is relatively easy, but the value of AGI is precisely reflected in its ability to tackle personalized and complex problems that users struggle to find answers for.

However, the commercialization of Agents cannot escape the iron laws of reality. Tang Jie pointed out three decisive factors: value, cost, and speed. First, Agents must address genuinely valuable human affairs; second, costs must be controllable—“if an API can solve the problem but the cost of the Agent is particularly high, that creates a contradiction”; finally, response and execution speed. The balance of these three factors will determine whether Agent products can transition from concept to scalable commercial success.

Looking ahead, Tang Jie remarked, “The opportunity for China’s AI industry lies in smart and daring young people, as well as a continuously improving business environment.” Yao Shunyu also believes, “Once a new business model is discovered, it can be quickly replicated in China, sometimes even done better in certain areas. This has been repeatedly validated in manufacturing and new energy vehicles, and it continues to happen.”

As Tang Jie stated, the key lies in whether everyone in the industry can “persist, dare to act, and take risks on a single path.” On the dual track of pursuing paradigm leadership and achieving commercial value, China’s AI industry stands at a new historical starting point, ready to turn the page on a new chapter of development.