Liu Tieyan, chairman of the Zhongguancun Artificial Intelligence Research Institute, pointed out while attending the 2026 World Artificial Intelligence Conference Wuwei Xinqiong Forum that as the pre-training data for large models is exhausted and the marginal effects of Scaling Law decrease, the future of the smart body lies not in the size of the initial model, but in the slope of its daily self-evolution, or “recursive self-improvement.” In this context, he proposed that smart bodies need to expand the boundaries of software, verification, and adaptation to form an “empirical flywheel.”

Liu Tieyan, chairman of the Zhongguancun Artificial Intelligence Research Institute, pointed out while attending the 2026 World Artificial Intelligence Conference Wuwei Xinqiong Forum that as the pre-training data for large models is exhausted and the marginal effects of Scaling Law decrease, the future of the smart body lies not in the size of the initial model, but in the slope of its daily self-evolution, or “recursive self-improvement.” In this context, he proposed that smart bodies need to expand the boundaries of software, verification, and adaptation to form an “empirical flywheel.”