Former Alibaba VP Zhang Kaifu: AI Enters 'Self-Improvement' Phase, Exponential Growth Expected in 1–2 Years

Huo Xing Cai Jing reports that on August 29, at the event "AI × New Finance — Innovation Global Tour Hong Kong Station and the Yangtze Star Program Launch," Zhang Kaifu, founder of Xiaopiti AI and former vice president of Alibaba, stated that the core evolution of AI over the past year has been the transition of models from chat tools to Agents capable of continuously completing complex tasks. The next phase, he emphasized, is even more significant: AI is beginning to participate in its own research, training, and inference optimization. Zhang believes this "recursive self-improvement" has already emerged in certain research efforts. If AI in the future can continuously optimize models and training efficiency like top-tier researchers, the pace of technological advancement could experience another exponential leap, making the next one to two years a critical variable in AI development. He also predicts that as model capabilities continue to improve, AI Agents will shift from passive responses to always-on, proactive task execution, driving new infrastructure demands in cloud computing, identity authentication, and cybersecurity.