Zhao Hejuan, the founder, Chairperson, and CEO of TMTPost, cautions against the optimistic view that China can quickly catch up in AI following the release of GPT-3 and subsequent models for several reasons.

Firstly, she believes that the timeframe of one to two years often used to describe the gap between China and the United States in AI is misleading. This timeframe corresponds to generational innovation cycles rather than actual capability gaps. For example, China might be able to develop a model similar to GPT-5 within a few years of its release, but that doesn't mean China's underlying innovation capabilities are on par with those of the U.S.

Secondly, she highlights the need for China to acknowledge its status as a follower in the AI race and dedicate efforts to diligent learning instead of overestimating its capabilities. She points out that China is not the largest market for AI patents and talent globally, despite the common belief. The number of top-tier papers and researchers in China is significantly less compared to the U.S.

Thirdly, she argues that the main obstacle for China's AI development is not just the "bottlenecked computing power" but also the lack of model innovation capability and data capability. She emphasizes that China lacks a mature data market, which is crucial for AI development.

Lastly, she calls for a shift in focus from debates over open-source vs. closed-source models to building up China's capability for innovation and originality. She also stresses the importance of enhancing fundamental innovation and fostering long-term capacity building in AI.

Therefore, Zhao Hejuan's caution against the optimistic view stems from her realistic assessment of China's current standing in the AI race and the need for strategic efforts to bridge the gap.