单词学习：![[b2d6f319-c310-4471-b15c-1c9ca17f8ee9.mp3]]

新闻播报：![[831fa5dd-fe65-413d-8a33-65a5789f948f.mp3]]

七部门发布“人工智能+交通”蓝图，明确2030年发展目标

简报：

• 交通运输部等7部门联合发布实施意见，提出到2027年，人工智能在交通运输行业实现广泛应用，综合交通运输大模型体系落地部署。
• 最终目标是到2030年，实现人工智能深度融入交通运输行业，全面推进智能综合立体交通网的建设。
• 重点任务包括开展智能驾驶大模型提升行动，加快智能驾驶系统、远程驾驶座舱、智能船舶和智慧列车等装备的研发。
• 意见支持在京津冀、长三角、粤港澳大湾区等地推进智能辅助驾驶技术示范应用。

相关链接：

• 关于“人工智能+交通运输”的实施意见

Seven Ministries Release "AI+Transportation" Blueprint, Setting Clear 2030 Development Goals 🛣️

Brief:

• The Ministry of Transport and six other departments have jointly issued implementation guidelines, proposing that by 2027, AI will achieve widespread application across the transportation sector, with a comprehensive large model system for integrated transportation deployed.
• The ultimate goal by 2030 is to achieve deep integration of AI into the transportation sector, fully advancing the construction of an intelligent, integrated, multi-dimensional transportation network. 🚀
• Key tasks include launching a "Smart Driving Large Model Enhancement Action" and accelerating the R&D of equipment such as intelligent driving systems, remote driving cockpits, smart vessels, and smart trains. 🚗
• The guidelines support promoting demonstration applications of intelligent assisted driving technology in regions like Beijing-Tianjin-Hebei, the Yangtze River Delta, and the Guangdong-Hong Kong-Macao Greater Bay Area.

Related Link:

• Implementation Guidelines on "Artificial Intelligence + Transportation"

斑马智行联合阿里高通，发布车载端侧大模型方案Auto Omni

简报：

• 在2025云栖大会上，斑马智行联合阿里通义及高通，全球首发全模态端侧大模型实车方案Auto Omni，旨在将智能座舱体验从指令式交互提升至主动服务。
• 该方案为车企提供交钥匙方案，包含独家深度定制的Qwen Omni模型和对高通8397等平台的适配。
• Auto Omni已获得国内头部车企定点合作，计划于2026年正式进入量产阶段。
• 斑马智行还联合网易云音乐、喜马拉雅等伙伴共同发起行业首个“AI车载平台服务联盟”，涵盖数字娱乐、本地生活、车辆服务三大场景。

相关链接：

• 独家绑定阿里通义及高通，斑马智行全球首发全模态端侧大模型实车方案 Auto Omni

Banma Zhixing Joins Forces with Alibaba and Qualcomm to Unveil Auto Omni, an In-Vehicle Edge Large Model Solution

Briefing:

• At the 2025 Apsara Conference, Banma Zhixing, in collaboration with Alibaba Tongyi and Qualcomm, globally launched Auto Omni 🚀, the world's first full-modal edge large model solution for vehicles. This aims to elevate the intelligent cockpit experience from command-based interaction to proactive service.
• This solution provides a turnkey package for automakers, featuring the exclusively deep-customized Qwen Omni model and adaptation to platforms like Qualcomm 8397.
• Auto Omni has already secured designated cooperation with leading domestic automotive manufacturers and is scheduled to officially enter mass production in 2026 🗓️.
• Banma Zhixing also partnered with NetEase Cloud Music, Ximalaya, and other collaborators to launch the industry's first "AI In-Vehicle Platform Service Alliance" 🤝, covering three major scenarios: digital entertainment, local life, and vehicle services.

Related Links:

• 独家绑定阿里通义及高通，斑马智行全球首发全模态端侧大模型实车方案 Auto Omni

OpenAI新测试：GPT-5在部分专业任务中表现接近人类专家

简报：

• OpenAI发布名为GDPval的新基准测试，旨在评估其AI模型在医疗、金融等九个主要行业中与人类专家的表现差异。
• 该测试涵盖了软件工程师、记者等44种职业，通过让专业人士对比AI与同行生成的报告来评估工作质量。
• 结果显示，增强版GPT-5在40.6%的任务中表现优于或持平于行业专家，而Anthropic的Claude Opus4.1则达到49%。
• OpenAI承认，当前测试范围有限，仅涵盖人类工作中的一小部分任务，并计划未来开发更全面的测试。

相关链接：

• OpenAI 最新基准测试显示 GPT-5 在多个行业中逐渐逼近人类专家
• OpenAI says GPT-5 stacks up to humans in a wide range of jobs
• OpenAI官方博客：GDPval

OpenAI's New Test: GPT-5 Approaches Human Expert Performance in Some Professional Tasks

Brief:

• OpenAI has launched a new benchmark test called GDPval, designed to evaluate how its AI models perform against human experts across nine major industries, including healthcare and finance. 🩺
• The test covers 44 professions, such as software engineers and journalists. It assesses work quality by having professionals compare reports generated by AI with those from their human peers.
• Results indicate that an enhanced version of GPT-5 performed better than or on par with industry experts in 40.6% of tasks, while Anthropic's Claude Opus 4.1 achieved 49%. 📈
• OpenAI acknowledges that the current test has a limited scope, covering only a small fraction of human work tasks, and plans to develop more comprehensive tests in the future. 🚀

Related Links:

• OpenAI 最新基准测试显示 GPT-5 在多个行业中逐渐逼近人类专家
• OpenAI says GPT-5 stacks up to humans in a wide range of jobs
• OpenAI Official Blog: GDPval

xAI起诉OpenAI，指控其通过挖角窃取商业机密

简报：

• 马斯克的xAI公司已对OpenAI发起诉讼，指控其通过“有组织的计划”挖角员工以获取商业机密。
• 诉讼文件显示，OpenAI被指挖走了八名xAI员工，其中包括多名工程师以及仅任职三个月的前首席财务官Mike Liberatore。
• xAI在诉讼文件中包含一张截图，显示其律师向一名离职高管发送指控其违约的邮件，对方在回复中使用了粗俗语言。
• 目前，OpenAI尚未对此事发表评论。

相关链接：

• 马斯克 xAI 起诉 OpenAI，指控后者通过挖角手段窃取商业机密

xAI Sues OpenAI, Alleges Trade Secret Theft Through Poaching

Brief:

• Elon Musk's xAI has filed a lawsuit against OpenAI, accusing it of an "organized scheme" to poach employees to gain access to trade secrets. 🕵️‍♀️
• According to the lawsuit documents, OpenAI allegedly poached eight xAI employees, including several engineers and former CFO Mike Liberatore, who had only been with xAI for three months.
• xAI included a screenshot in the lawsuit showing its lawyers sending an email to a departing executive accusing them of breach of contract, to which the executive reportedly replied with vulgar language. 😠
• OpenAI has not yet commented on the matter.

Related Links:

• Elon Musk's xAI Sues OpenAI, Alleging Trade Secret Theft Through Poaching ⚖️

DeepSeek V3.1-Terminus发布，修复关键Bug为V4铺路

简报：

• 国产AI大模型DeepSeek发布V3.1-Terminus版本，其命名“Terminus”被解读为V3系列的收官之作。
• 新版本重点修复了此前存在的“极你太美”异常输出、中英文混杂及异常字符等技术问题，显著提升了模型稳定性。
• V3.1版本对Code Agent和Search Agent模块进行了深度优化，提升了代码生成和搜索功能的准确性。
• 部分用户在编程竞赛测试中发现，新版本在某些复杂算法问题上表现略有下降，分析称这可能是为提升安全性进行的权衡。
• 业界普遍预期，采用全新架构的V4版本或代号为R2的重大更新可能在年底发布。

相关链接：

• DeepSeek V3.1终结版发布：修复关键漏洞提升稳定性，V4新架构呼之欲出

DeepSeek V3.1-Terminus Released, Fixing Critical Bugs and Paving the Way for V4

Briefing:

• Chinese domestic AI large model DeepSeek has released its V3.1-Terminus version, with its name 'Terminus' interpreted as the final iteration of the V3 series. 🚀
• The new version primarily addresses previous technical issues such as the 'Jǐ Nǐ Tài Měi' (a common meme output) abnormal output, mixed Chinese-English text, and unusual characters, significantly enhancing model stability. ✨
• V3.1 brings deep optimizations to its Code Agent and Search Agent modules, improving the accuracy of code generation and search functionalities.
• Some users participating in programming contest tests noted a slight decrease in performance for certain complex algorithmic problems in the new version, with analyses suggesting this might be a trade-off for enhanced security.
• The industry widely anticipates that a V4 version featuring a completely new architecture, or a major update codenamed R2, could be released by year-end. 🔜

Related Links:

• DeepSeek V3.1 Final Version Released: Critical Vulnerabilities Fixed, Stability Improved, V4 New Architecture Imminent

国家发改委发文，加速构建全国一体化算力网络

简报：

• 国家发改委于9月26日印发《关于加强数字经济创新型企业培育的若干措施》通知，提出加快构建全国一体化算力网。
• 措施旨在推动全国算力资源有序池化、并网运行，打造集算力统筹监测、统一调度、弹性供给、安全保障于一体的新型算力基础设施。
• 鼓励国家枢纽节点面向数字创新企业提供低成本、广覆盖、可靠安全的算力服务，以降低企业使用算力的门槛。

相关链接：

• 国家发改委印发《关于加强数字经济创新型企业培育的若干措施》的通知
• 国家发改委：加快构建全国一体化算力网

NDRC Issues Document, Accelerating the Construction of a National Integrated Computing Network

Brief:

• On September 26, the National Development and Reform Commission (NDRC) issued the "Several Measures on Strengthening the Cultivation of Innovative Enterprises in the Digital Economy" notice, proposing to accelerate the construction of a national integrated computing network. 🚀
• The measures aim to promote the orderly pooling and network operation of national computing resources, creating a new type of computing infrastructure that integrates comprehensive monitoring, unified scheduling, elastic supply, and security guarantees. 💡
• It encourages national hub nodes to provide low-cost, wide-coverage, reliable, and secure computing services to digital innovation enterprises, thereby lowering the barrier for enterprises to use computing power. 🌐

Related Links:

• NDRC Issues Notice on 'Several Measures on Strengthening the Cultivation of Innovative Enterprises in the Digital Economy'
• NDRC: Accelerating the Construction of a National Integrated Computing Network

Meta发布代码世界模型，引入预测能力革新AI编程

简报：

• Meta公司于2025年9月25日正式发布名为“代码世界模型”（Code World Model，简称CWM）的大型语言模型，旨在提升AI的代码生成能力。
• CWM引入“世界模型”理念，能够在生成代码的同时预测指令执行可能产生的效果，使其具备类似人类的规划和推理能力。
• 该模型针对Python和Bash（Linux和macOS的命令行语言）进行了专门训练，能模拟Python程序在Bash环境中的执行过程。
• 在多个编程基准测试中，32B参数的小型CWM模型表现出色，例如在SWE-bench Verified测试中获得65.8%的成绩，在Math-500测试中达到96.6%。

相关链接：

• Meta重磅发布代码世界模型，AI编程将进入新纪元

Meta Unveils Code World Model, Revolutionizing AI Programming with Predictive Capabilities 💻

Brief:

• On September 25, 2025, Meta officially released its Large Language Model named “Code World Model” (CWM), aiming to enhance AI's code generation capabilities.
• CWM incorporates the “world model” concept, allowing it to predict the potential outcomes of instruction execution while generating code, thus endowing it with human-like planning and reasoning abilities. 💡
• The model has been specifically trained on Python and Bash (the command-line language for Linux and macOS), capable of simulating the execution of Python programs within a Bash environment.
• In various programming benchmarks, the smaller 32B-parameter CWM model demonstrated impressive performance, achieving 65.8% in the SWE-bench Verified test and 96.6% in the Math-500 test. 🚀

Related Links:

• Meta's Groundbreaking Code World Model: A New Era for AI Programming