The Chinese artificial intelligence landscape is undergoing a silent but profound bifurcation. While the initial global race was defined by a frantic grab for GPUs and massive datasets, the leading edge of China's intellectual elite is now questioning the wisdom of merely shadowing Silicon Valley. Liu Zhiyuan, a preeminent scholar at the Beijing Academy of Artificial Intelligence (BAAI) and co-founder of ModelBest, argues that China’s path to dominance lies not in being a 'second OpenAI,' but in redefining the medium of intelligence itself.
Central to Liu’s vision is the 'Densing Law,' a phenomenon he and his team documented in the journal Nature Machine Intelligence. Their research suggests that 'intelligence density'—the amount of cognitive capability per parameter—is doubling roughly every 3.5 months. This implies that every 100 days, the same level of capability can be achieved with half the parameters. For China, this shift represents a strategic escape from the cloud-only paradigm dominated by American hyperscalers, moving intelligence into 'edge' devices like smartphones, automobiles, and wearable tech.
Liu remains candid about the current global hierarchy. While Chinese labs have demonstrated a remarkable ability to refine and surpass Western models in efficiency—exemplified by the recent global waves made by DeepSeek—he warns of a persistent 'original innovation gap.' The United States remains the primary generator of 'zero-to-one' breakthroughs, such as the initial validation of deep reasoning. China, by contrast, has mastered the 'one-to-one hundred' phase, optimizing architectures and data governance with surgical precision after the path has been cleared.
To bridge this gap, Liu highlights the importance of institutional hybrids like BAAI. These organizations operate in the 'middle ground' between the ivory tower of academia and the high-pressure environment of commercial enterprise. Free from the immediate demands of quarterly KPIs, these incubators allow for the 'outward competition' (waijuan) that Liu advocates: entering real-world industrial scenarios to solve problems that Silicon Valley has not yet defined, rather than fighting over-saturated Western categories.
Ultimately, the shift toward smaller, high-density models suggests a future where AI is pervasive, localized, and private. Liu envisions a world of 'One-Person Companies' where individual humans act as coordinators for fleets of specialized AI agents. This transition moves the focus of human education away from rote labor and toward the higher-order skills of organizational architecture and critical judgment, ensuring that as AI shrinks in size, its impact on human productivity expands exponentially.