Moonshot AI's Xu Tiantian Appears at Caixin Summit: AI Capabilities Are Far from Hitting the Ceiling; Three-Dimensional Curves Are Driving Intelligence to Accelerate Continuously

At a time when the pace of AI development is sparking widespread discussion, where exactly are the capability boundaries of large models? At the 17th Caixin Summit held on November 12, Xu Tiantian, Head of Financial Industry and Ecosystem at Moonshot AI, shared a professional view, clearly pointing out that the current capability boundaries and intelligence ceiling of AI are far from reaching their limits, and the industry as a whole is still on the fast track of accelerated evolution.

Three overlapping growth curves: the underlying logic driving the continuous evolution of large models

In her keynote speech, Xu Tiantian analyzed in depth the driving forces behind AI's evolution. She noted that the rapid leap in current AI capabilities is not caused by a single factor, but is the result of three key growth curves stacking and resonating together:

First: the traditional pretraining curve. This covers the continued growth of traditional dimensions such as computing infrastructure, energy supply, and model parameter scale. This foundational curve is still maintaining an upward momentum.

Second: growth on the post-training and inference side. By introducing Chain of Thought and specialized training for the inference side, large models have achieved a qualitative leap in logical thinking and the decomposition of complex problems.

Third: the system design of agents. As agents themselves and their complex control frameworks (Harness engineering) continue to iterate and upgrade, the upper limit of collaboration for the entire system is further raised.

A deeper look at industry transformation: technological iteration far outpaces institutional design

As large-model technology advances at breakneck speed, Xu Tiantian also pointed out a major practical challenge facing the industry in real-world applications and ecosystem deployment: the pace at which enterprises advance workflow reshaping and internal institutional design is currently far behind the pace of the models' own iteration.

This speed gap between technological development and organizational change means that when embracing AI-driven transformation, enterprises cannot merely stop at integrating tools at the surface level. They need to make deep adaptations in organizational structure and business processes to truly convert the explosive power of underlying technology into tangible business value and productivity.