在苏箐看来，这一阶段的核心命题，是将现有技术的潜力发挥到极致，比如地平线将持续提升芯片算力 与模型容量，以统一范式推进L2到L4的融合，让城市L2从20万级车型下探至10万级市场，让准L4系统 以平民化价格走进大众。同时，强化工程与组织能力，应对海量长尾场景的打磨，这才是穿越周期的关 键。 "自动驾驶的终极目标，是造出能替代人类司机的机器。这场长跑，在范式革命之后，考验的是行业沉 下心来做'精活'的耐力。在未来几年内，能够把L4级别的车，以平民化的价格送到用户手上。这才是我 们所有人辛苦了这20年做这一行的意义所在。" （文章来源：中国经营报） "未来三年，自动驾驶行业将告别范式迭代的狂飙，进入极致优化的'苦日子'。" 12月9日，在"2025地平线技术生态大会"上，作为深耕自动驾驶20年的老兵，一向"反共识"的地平线副 总裁&首席架构师苏箐分享了对自动驾驶未来趋势的判断。 值得一提的是，对于当下，苏箐则给出了冷静的判断："行业需要清醒。当前深度学习已显露天花板迹 象，AGI基础理论暂无突破信号，下一轮内核重构至少还需5—20年的技术沉淀。未来三年，自动驾驶 行业将告别范式迭代的狂飙，进入极致优化的'苦日子' ...

Core Insights - "Zijing Zhikang" has completed nearly 100 million yuan in angel round financing, led by Xinglian Capital, with the funds primarily allocated for the development, iteration, and upgrade of the Zijing AI Hospital system [1] Company Overview - Established in September 2024, Zijing Zhikang was incubated by Tsinghua University's Intelligent Industry Research Institute and initiated by Professor Liu Yang [1] - The company aims to leverage cutting-edge large model AI technology to develop a medical virtual world system and promote its application and optimization in the real world, thereby empowering smart healthcare [1] Technology and Innovation - The core logic of the Zijing AI Hospital is to simulate real hospital facilities and processes, particularly by creating highly human-like, widely distributed, and diverse AI patients to meet initial training data needs [1] - The AI hospital has constructed over 500,000 AI patients covering various countries, age groups, and disease types, serving as an important supplement for training AI doctor agents [2] AI Doctor Development - The team has designed specific memory and reflection algorithms that allow AI doctors to accumulate "experience" during the consultation loop, with validated experiences entering the AI doctor's database as "exclusive memory" [3] - The AI doctors have achieved an accuracy rate exceeding 96% on the MedQA dataset, surpassing the average level of human doctors [3] Product Design and Functionality - The Zijing AI system features three ports: a patient app, a doctor workstation, and a hospital system, facilitating full-cycle health management from pre-diagnosis to post-diagnosis [3] - Patients can register online, engage in intelligent pre-consultation, and generate structured medical records, while doctors can access these records to save time and focus on critical medical decisions [3] Future Plans - The Zijing AI Hospital system is set to launch on June 30, 2025, with internal testing already underway in various departments at Tsinghua University Hospital [4] - Public testing is planned for the end of 2025, expanding from Beijing to more cities and covering various hospital levels and departments [4] Investor Perspectives - Investors highlight the innovative breakthroughs in technology and the potential for Zijing AI Hospital to reshape healthcare efficiency and accessibility [5][6] - The project is seen as a significant force in promoting smart healthcare infrastructure transformation, supported by national policies favoring AI in healthcare [6]

Core Viewpoint - The article discusses the rapid rise of large model technology globally, emphasizing Alibaba's Tongyi Qwen model's international success and its strategic focus on multilingual capabilities to cater to a global audience [2][3]. Group 1: Multilingual Strategy - Tongyi Qwen supports 119 languages, with a core strategy prioritizing multilingual data optimization from the outset to ensure equitable access to AI technology for global users [2][3]. - The team has developed a complex cultural annotation system to address the challenges of multilingual safety and cultural alignment, covering thousands of detailed categories to ensure compliance and effectiveness across different regions [3][12]. - The current industry faces a "multilingual reasoning challenge," where models often mix languages during processing, leading to inconsistencies. The team has adopted a compromise strategy to use native languages for strong languages and English for low-resource languages to maintain output stability [3][16]. Group 2: Scaling Law and Knowledge Density - The article highlights the importance of scaling model size and data volume while also focusing on increasing "knowledge density," which refers to the concentration of useful knowledge within the training data [19][20]. - Recent trends show that smaller models with higher knowledge density can outperform larger models, indicating a shift in focus from merely increasing data volume to refining data quality [20][21]. - The team is exploring data synthesis methods to enhance training data quality, which includes generating new knowledge and filtering redundant data to improve knowledge density [22][23]. Group 3: AI Integration and Future Prospects - The integration of AI models into various devices, such as smart glasses and earphones, is a growing trend, with the company planning to release smaller model versions optimized for these applications [28][30]. - The article discusses the potential for AI to enhance user experiences in everyday tasks, such as real-time translation and contextual assistance, although challenges remain in achieving seamless integration [30][32]. - The company acknowledges the importance of balancing the use of synthetic data with human-generated content to maintain diversity and avoid narrowing the model's knowledge base [25][26].

Core Insights - The company, Zhiyuan (Beijing Zhiyuan Huazhang Technology Co., Ltd.), officially launched the intelligent agent "AutoGLM Rumination" at the Zhongguancun Forum on March 31, showcasing its deep research capabilities and practical operations, marking a transition to "thinking while doing" in AI agents [1] - CEO Zhang Peng highlighted the presence of a Scaling Law in agents, indicating that by expanding inference compute during training, agents demonstrate enhanced performance [1] - The technology evolution path of AutoGLM includes the GLM-4 base model, GLM-Z1 inference model, GLM-Z1-Rumination model, and the AutoGLM model, with core models and technologies set to be open-sourced on April 14 to promote industry ecosystem development [1] Model Development - Based on recent technological advancements, the company retrained a base model called GLM-4-Air-0414 with 32 billion parameters, incorporating more code and reasoning data during the pre-training phase and optimizing for agent capabilities during the alignment phase, significantly enhancing its abilities in tool invocation and online search tasks [2]