CENTEK(SZ:000931) Huan Qiu Wang·2025-03-31 09:24Core Insights - The 2025 Zhongguancun Forum showcased the AI agent AutoGLM, which integrates deep research capabilities with practical operations, marking a shift towards "thinking while doing" in AI technology [1][3]. Group 1: AI Agent Capabilities - AutoGLM is designed to perform complex tasks by understanding user intent through simple text or voice commands, enabling it to simulate human actions such as web browsing, data retrieval, and report generation [3][4]. - The AI agent's capabilities are powered by the "thinking model," which employs reinforcement learning to enable self-critique, reflection, and deeper thinking for improved outcomes [3][4]. Group 2: Technological Foundation - The thinking model is built on a proprietary full-stack large model technology developed by Zhiyuan AI, combining the general capabilities of GLM-4, reflective abilities of GLM-Z1, and the execution capabilities of AutoGLM [4]. - The thinking function has been officially launched on Zhiyuan's web, PC, and mobile platforms, providing free and unlimited access to global users, representing a significant advancement in deep research capabilities [4]. Group 3: Future Outlook - The CEO of Zhiyuan AI, Zhang Peng, predicts that 2025 will be a breakthrough year for AI agents, with ongoing development of models that possess logical reasoning and deep thinking capabilities [4].