单词学习：![[5f5f4a83-30e9-44ab-9864-d715ac2001db.mp3]]

新闻播报：![[3ea7ca2f-2d67-4128-838f-1b276ea17bbd.mp3]]

蚂蚁集团设立AI特别激励，加速全面AI化转型

简报：

• 蚂蚁集团CEO韩歆毅通过内部信宣布，推出“AI Credit”特别激励方案，为在AI方向有开创性贡献的团队和个人提供额外激励。
• 韩歆毅在信中提出“业务和组织都要全面AI化”的目标，并指出公司虽取得“小胜”，但仍是“追赶者”，需拒绝小胜即安。
• 该激励方案旨在推动AI技术深入核心业务，将与相关业务未来两年为公司创造的价值挂钩。
• 内部信同时回顾了公司近期在AI应用（如“蚂蚁阿福”月活超3000万）、AGI探索（百灵大模型进入万亿参数行列）以及支付和全球化领域取得的新进展。

相关链接：

• 蚂蚁集团推出“AI Credit”激励方案，推动业务和组织“全面AI化”

Ant Group Establishes Special AI Incentive, Accelerating All-AI Transformation

Brief:

• Ant Group CEO Han Xinyi announced via an internal letter the launch of a "AI Credit" special incentive program, offering additional rewards for teams and individuals making pioneering contributions in AI. 🚀
• In the letter, Han Xinyi proposed the goal of "making both business and organization fully AI-driven," emphasizing that while the company has achieved "small victories," it remains a "chaser" and must reject complacency.
• The incentive program aims to integrate AI technology deeper into core businesses, with rewards linked to the value created by relevant businesses for the company over the next two years.
• The internal letter also reviewed the company's recent advancements in AI applications (e.g., "Ant Afu" exceeding 30 million monthly active users 🤖), AGI exploration (the Bailian large model entering the trillion-parameter league), and new progress in payments and globalization. 🌍

Related links:

• Ant Group Launches "AI Credit" Incentive Program to Drive "All-AI Transformation" of Business and Organization

马斯克xAI发布Grok Imagine 1.0，视频生成能力再升级

简报：马斯克旗下的人工智能公司xAI于2月2日宣布全面发布Grok Imagine 1.0，该模型能够生成长达10秒、分辨率为720p并具有更高音频质量的视频。

相关链接：

• 马斯克旗下 xAI 发布 Grok Imagine 1.0，可生成 10 秒、720p 分辨率视频

Musk's xAI Unveils Grok Imagine 1.0, Video Generation Capabilities Further Enhanced 🚀

Brief: Musk's artificial intelligence company xAI announced the full release of Grok Imagine 1.0 on February 2nd. This model is capable of generating videos up to 10 seconds long, with 720p resolution, and enhanced audio quality. ✨

Related Links:

• Musk's xAI Unveils Grok Imagine 1.0, Capable of Generating 10-Second, 720p Resolution Videos 🎬

特斯拉人形机器人迈向大规模量产，2026年亮相并预计年产百万

简报：

• 特斯拉官方宣布第三代人形机器人将于2026年亮相，是其首款为量产而设计的人形机器人。
• 该机器人基于第一性原理重新设计，能够通过观察人类行为来学习新技能。
• 投产后，预计年产量将达到百万台，每台成本控制在2万美元以内。

相关链接：

• 特斯拉第三代人形机器人官宣 2026 年亮相：支持观察人类行为学习新技能，预计年产百万台

Tesla Humanoid Robot Nears Mass Production, Set for 2026 Debut with Million-Unit Annual Output Forecast

Briefing:

• Tesla officially announced that its third-generation humanoid robot will debut in 2026, marking its first humanoid robot designed for mass production. 🤖
• The robot has been redesigned based on first principles, capable of learning new skills by observing human behavior.
• After going into production, the estimated annual output is expected to reach millions of units, with the cost per unit controlled under $20,000. 💰✨

Related Link:

• Tesla's third-generation humanoid robot officially announced for 2026 debut: supports learning new skills by observing human behavior, estimated annual output of millions

英伟达开源三款AI气象模型，提升预报效率与精度

简报： 英伟达近日发布了名为“Earth-2”系列的三款开源人工智能模型，旨在通过比传统数值模拟更快、成本效益更高的方式，提供包括全球15天天气和美国6小时强风暴在内的精准预报，推动气象行业向AI与数据驱动转型。

相关链接：

• 英伟达发布开源AI模型：让气象预报更快、更省、更精准

NVIDIA Open-Sources Three AI Weather Models, Boosting Forecasting Efficiency and Accuracy

Brief: NVIDIA recently unveiled three open-source artificial intelligence models from its "Earth-2" series 🌍. These models are designed to provide precise forecasts, including global 15-day weather and 6-hour severe storms in the U.S., faster ⚡ and more cost-effectively than traditional numerical simulations. This initiative aims to drive the meteorological industry towards an AI and data-driven transformation 🚀.

Related links:

• NVIDIA Releases Open-Source AI Models: Making Weather Forecasts Faster, Cheaper, and More Accurate

阶跃星辰开源新模型Step 3.5 Flash，专为Agent打造

简报：

• 阶跃星辰于2月2日发布了其最新开源基座模型Step 3.5 Flash，宣称该模型为Agent而生。
• 官方介绍，该模型具备高达350 TPS的推理速度，在Agent场景和数学任务上可媲美闭源模型。
• Step 3.5 Flash采用稀疏MoE架构，总参数量1960亿，但每个token仅激活约110亿参数，并能高效处理256K上下文。
• 该模型现已全量上线，同时官方已启动下一代Step 4模型的训练。

相关链接：

• 为 Agent 而生，阶跃星辰发布开源基座模型 Step 3.5 Flash

Jiesu Xingchen Open-Sources New Model Step 3.5 Flash, Built Specifically for Agents

Brief:

• Jiesu Xingchen released its latest open-source foundational model, Step 3.5 Flash, on February 2nd, claiming the model was "born for Agents." 🤖
• According to official introductions, the model boasts an inference speed of up to 350 TPS and rivals closed-source models in Agent scenarios and mathematical tasks.
• Step 3.5 Flash adopts a sparse MoE architecture with a total of 196 billion parameters, yet only activates approximately 11 billion parameters per token, while efficiently handling 256K context.
• The model is now fully launched, and official training for the next-generation Step 4 model has already commenced. 🚀✨

Related Links:

• Born for Agents: Jiesu Xingchen Releases Open-Source Foundational Model Step 3.5 Flash

优必选开源机器人“大脑”Thinker，旨在解决数据难题

简报：

• 优必选公司宣布开源其为工业人形机器人设计的具身智能大模型Thinker，旨在解决当前机器人在空间理解和视觉感知等任务上存在的精度不足、参数庞大及实时性差的问题。
• Thinker模型提出“精炼提纯-自动化标注-数据驱动训练”的全链路解决方案，能够将20B的原始数据经过处理提纯至10M，以提升数据质量。
• 该模型采用“弱监督+自监督+少量人工校验”的自动化标注体系，可将数据标注成本降低99%，并通过持续的闭环迭代提升模型准确性。

相关链接：

• 优必选开源 Thinker 大模型，赋能工业人形机器人新未来

UBTECH Open-Sources Robot 'Brain' Thinker to Tackle Data Challenges 🧠

Brief:

• UBTECH announced the open-sourcing of Thinker, its large embodied AI model designed for industrial humanoid robots. The model aims to address current challenges in tasks like spatial understanding and visual perception, which often suffer from insufficient accuracy, enormous parameter counts, and poor real-time performance. 🤖
• The Thinker model proposes a full-lifecycle solution encompassing "refinement and purification - automated labeling - data-driven training." This approach can process 20 billion raw data points, purifying them down to 10 million, significantly enhancing data quality.
• The model employs an automated labeling system that combines "weak supervision + self-supervision + minimal human validation." This innovative system is projected to reduce data labeling costs by 99% and continuously enhance model accuracy through ongoing closed-loop iterations. ✨

Related Links:

• UBTECH Open-Sources Thinker Large Model, Empowering a New Future for Industrial Humanoid Robots

AI赋能传统建材：东南大学发布混凝土科学大模型

简报：

• 东南大学于2026年2月1日发布了全球首个名为“砼真砼知”的混凝土材料科学大模型。
• 该模型由东南大学、重大基础设施工程材料全国重点实验室与阿里云联合研发，是国内首个混凝土材料科学大模型。
• 模型以“人工智能 + 混凝土材料”的深度融合为核心，旨在为全球基础设施建设的数字化与低碳化转型提供中国方案。
• 该成果已在南京北站建设项目中首次应用，标志着我国工程材料智能化取得显著进步。

相关链接：

• 东南大学发布全球首个混凝土科学大模型

AI Empowers Traditional Building Materials: Southeast University Releases Concrete Science Large Model

Briefing:

• Southeast University unveiled "Tong Zhen Tong Zhi" (砼真砼知), the world's first large model for concrete material science, on February 1, 2026. 🚀
• Developed jointly by Southeast University, the National Key Laboratory for Major Infrastructure Engineering Materials, and Alibaba Cloud, it stands as China's first large model in concrete material science.
• The model is centered on the deep integration of "Artificial Intelligence + Concrete Materials," aiming to offer a Chinese solution for the digital and low-carbon transformation of global infrastructure construction. 💡
• This achievement has already been successfully applied in the Nanjing North Railway Station construction project, marking significant progress in the intelligentization of engineering materials in China.

Related Links:

• Southeast University Releases World's First Large Model for Concrete Science 🏗️