单词学习：![[9e98c4ff-f093-4234-b6be-856df6dda70e.mp3]]

新闻播报：![[acf27034-edc6-49c1-9eec-5f5b8d53859f.mp3]]

OpenAI发布多款AI产品，加速“AI同事”落地企业与科研

简报：

• OpenAI发布全新编程大模型GPT-5.3-Codex，该模型在编程及计算机系统操作方面性能显著提升，同时运行效率更高，令牌消耗较上一代减少超50%。
• 推出名为Frontier的企业级平台，旨在帮助企业构建、部署并管理能够执行真实工作的AI智能体（AI同事），已吸引Oracle、Uber等首批合作伙伴。
• 展示了GPT-5与生物技术公司Ginkgo Bioworks的合作成果，AI通过自主设计实验并操控机器人，成功将无细胞蛋白质合成的生产成本降低了约40%。

相关链接：

• ​OpenAI 发布 GPT-5.3-Codex：编程效能跃升，开启“AI 同事”实操时代
• OpenAI 发布 Frontier 平台：打造“AI 同事”生态，加速企业级智能体落地
• OpenAI GPT-5 首秀生物界：AI 自主操控机器人，蛋白质生产成本暴降 40%

OpenAI Launches Multiple AI Products, Accelerating "AI Colleague" Adoption in Enterprises and Research 🚀

Brief:

• OpenAI has released the new large programming model, GPT-5.3-Codex. This model significantly enhances performance in programming and computer system operations, while also boasting higher operational efficiency and reducing token consumption by over 50% compared to its predecessor.
• Introduced Frontier, an enterprise-grade platform designed to help businesses build, deploy, and manage AI agents (AI colleagues) capable of performing real-world tasks. It has already attracted initial partners such as Oracle and Uber.
• Showcased the results of a collaboration between GPT-5 and biotechnology company Ginkgo Bioworks, where AI autonomously designed experiments and controlled robots, successfully reducing the production cost of cell-free protein synthesis by approximately 40%. 🔬💡

Related Links:

• ​OpenAI 发布 GPT-5.3-Codex：编程效能跃升，开启“AI 同事”实操时代
• OpenAI 发布 Frontier 平台：打造“AI 同事”生态，加速企业级智能体落地
• OpenAI GPT-5 首秀生物界：AI 自主操控机器人，蛋白质生产成本暴降 40%

AI指挥机器人做实验，GPT-5将蛋白质生产成本削减四成

简报：

• OpenAI与生物技术公司Ginkgo Bioworks合作，利用GPT-5构建了一个“闭环自动化系统”，由AI直接连接并指挥云端湿实验室的机器人执行操作。
• 在该系统中，GPT-5负责设计实验、分析数据并自主规划下一轮实验，仅用3轮迭代就超越了此前的人类最佳基准。
• 该合作成功将无细胞蛋白质合成（CFPS）的总生产成本降低了约40%，并将试剂成本大幅削减了57%。
• GPT-5通过在高维度参数空间中寻找规律，识别出了一系列人类科学家此前未曾测试过的低成本、高鲁棒性的配方组合。

相关链接：

• OpenAI GPT-5 首秀生物界：AI 自主操控机器人，蛋白质生产成本暴降 40%
• ​OpenAI 发布 GPT-5.3-Codex：编程效能跃升，开启“AI 同事”实操时代
• OpenAI 发布 Frontier 平台：打造“AI 同事”生态，加速企业级智能体落地

GPT-5-Powered AI Directs Robots in Experiments, Slashes Protein Production Costs by 40% 🤖🧪

Brief:

• OpenAI, in collaboration with biotechnology company Ginkgo Bioworks, leveraged GPT-5 to build a "closed-loop automation system" where AI directly connects to and commands robots in a cloud-based wet lab to execute operations.
• Within this system, GPT-5 is responsible for designing experiments, analyzing data, and autonomously planning subsequent experimental rounds, surpassing previous human best benchmarks in just 3 iterations.
• This collaboration successfully reduced the total production cost of cell-free protein synthesis (CFPS) by approximately 40% and significantly slashed reagent costs by 57%. 💰
• By identifying patterns within high-dimensional parameter spaces, GPT-5 pinpointed a series of low-cost, highly robust formula combinations that human scientists had not previously tested.

Related Links:

• OpenAI GPT-5 Debuts in Biotech: AI Autonomously Controls Robots, Protein Production Costs Plummet 40%
• OpenAI Releases GPT-5.3-Codex: Programming Efficiency Leaps, Ushering in the Era of 'AI Colleagues' in Practice
• OpenAI Launches Frontier Platform: Building an 'AI Colleague' Ecosystem, Accelerating Enterprise Agent Deployment

美团开源LongCat新模型：以45亿激活参数挑战巨型模型性能

简报：

• 美团发布新模型LongCat-Flash-Lite，其采用“嵌入扩展”新范式，总参数达685亿，但在推理时仅需激活29亿至45亿参数。
• 通过系统级优化，该模型API在典型负载下可提供500-700 token/s的生成速度，并支持最长256K上下文。
• 在多个基准测试中表现优异，其中代码能力SWE-Bench准确率达54.4%，通用能力MMLU得分85.52，性能可与巨量模型比肩。
• 美团已全面开源该模型权重、技术报告及配套推理引擎SGLang-FluentLLM，并提供每日5000万tokens的免费API试用额度。

相关链接：

• 美团 LongCat-Flash-Lite 震撼发布：45 亿激活参数性能比肩巨量模型
• 融资2.55亿美元
  实验室Fundamental破茧而出，用Nexus模型重塑大模型分析能力
• Claude Opus 4.6限时免费！ZenMux重磅升级，AI模型聚合平台迎来新福利

Meituan Open-Sources New LongCat Model: 4.5 Billion Activated Parameters Rival Giant Model Performance

Brief:

• Meituan has released its new model, LongCat-Flash-Lite, which adopts a novel "embedding expansion" paradigm. While it boasts a total of 68.5 billion parameters, it only requires 2.9 to 4.5 billion parameters to be activated during inference. 🚀
• Through system-level optimizations, the model's API can achieve a generation speed of 500-700 tokens/s under typical loads and supports up to 256K context length.
• It has shown excellent performance across multiple benchmark tests, including a 54.4% accuracy on the SWE-Bench coding capability test and an MMLU score of 85.52 for general capabilities, with performance comparable to giant models. ✨
• Meituan has fully open-sourced the model weights, technical report, and the accompanying inference engine SGLang-FluentLLM, and offers a free API trial quota of 50 million tokens per day. 🎁

Related Links:

• Meituan LongCat-Flash-Lite Shakes the Market: 4.5 Billion Active Parameters Rival Giant Model Performance
• Funded with $255 Million: AI Lab Fundamental Breaks Out, Reshaping Large Model Analysis Capabilities with Nexus Model
• Claude Opus 4.6 Free for a Limited Time! ZenMux Significantly Upgraded, AI Model Aggregation Platform Welcomes New Benefits

特斯拉将积极参与中国辅助驾驶工作，落地时间待定

简报：

• 特斯拉副总裁陶琳于2026年2月6日表示，特斯拉会非常积极地参与中国市场辅助自动驾驶等相关工作。
• 陶琳同时指出，目前尚无法提供该功能在中国市场明确的落地时间。
• 此前，特斯拉CEO马斯克曾表示，FSD（完全自动驾驶能力）将于今年二月在欧洲部署，中国市场会随后跟进。

相关链接：

• 特斯拉陶琳：会积极参与中国市场辅助自动驾驶相关工作
• Waymo 确认完成 160 亿美元融资：估值破千亿，领跑全球无人驾驶赛道

Tesla to Actively Participate in China's Assisted Driving Initiatives, Launch Date Undetermined

Briefing:

• Tesla Vice President Tom Zhu stated on February 6, 2026, that Tesla would very actively participate in assisted autonomous driving and related work in the Chinese market. 🇨🇳
• Zhu also noted that a clear launch date for this feature in the Chinese market cannot be provided at this time. ⏳
• Previously, Tesla CEO Elon Musk had stated that FSD (Full Self-Driving capability) would be deployed in Europe this February, with the Chinese market following thereafter. 🛣️

Related Links:

• Tesla's Tom Zhu: Will Actively Participate in Assisted Autonomous Driving Work in the Chinese Market
• Waymo Confirms dJyH6kDkfC5RlM16.answerText6 Billion Funding Round: Valuation Surpasses dJyH6kDkfC5RlM16.answerText00 Billion, Leading Global Autonomous Driving Race

Waymo完成160亿美元融资，估值突破千亿领跑全球无人驾驶

简报：

• 自动驾驶公司Waymo已正式确认完成新一轮高达160亿美元的融资，公司估值达到1260亿美元。
• Waymo的无人驾驶出租车服务已在美国旧金山湾区、菲尼克斯、洛杉矶及迈阿密等多个城市运营，并与Uber合作进入亚特兰大和奥斯汀市场。
• 自2024年10月C轮融资以来，Waymo估值从450亿美元翻倍增长，吸引了红杉资本、老虎环球基金等投资机构。
• Waymo官方披露，在超过1.27亿英里的全自动驾驶测试中，其严重伤害事故率比人类驾驶员降低了90%。
• Waymo采用“规则驱动”技术方案，通过激光雷达、雷达及摄像头实现高精度感知和决策，与特斯拉的“端到端”神经网络方案不同。
• 公司计划在2026年将服务扩展至达拉斯、拉斯维加斯、华盛顿等十余个美国城市，并筹备进军伦敦与东京，开启国际化商业运营。

相关链接：

• Waymo确认完成160亿美元融资，领跑全球无人驾驶赛道

Waymo Completes dJyH6kDkfC5RlM16.answerText6 Billion Funding Round, Valuation Exceeds dJyH6kDkfC5RlM16.answerText00 Billion, Leading Global Autonomous Driving

Brief:

• Autonomous driving company Waymo has officially confirmed the completion of a new funding round totaling dJyH6kDkfC5RlM16.answerText6 billion, bringing the company's valuation to dJyH6kDkfC5RlM16.answerText26 billion. 🚀
• Waymo's self-driving taxi service is already operating in multiple US cities, including the San Francisco Bay Area, Phoenix, Los Angeles, and Miami, and has partnered with Uber to enter the Atlanta and Austin markets.
• Since its Series C funding round in October 2024, Waymo's valuation has more than doubled from $45 billion, attracting investment firms such as Sequoia Capital and Tiger Global.
• Waymo officially disclosed that in over 127 million miles of fully autonomous driving tests, its serious injury accident rate was 90% lower than that of human drivers. 🚦
• Waymo employs a "rule-driven" technical approach, utilizing LiDAR, radar, and cameras for high-precision perception and decision-making, differentiating it from Tesla's "end-to-end" neural network solution.
• The company plans to expand its services to over a dozen more US cities, including Dallas, Las Vegas, and Washington, by 2026, and is preparing to enter London and Tokyo, commencing international commercial operations. 🌍

Related links:

• Waymo Confirms Completion of dJyH6kDkfC5RlM16.answerText6 Billion Funding Round, Leading the Global Autonomous Driving Race

苹果AI新研究：通过专业反馈，小参数模型Qwen3-Coder实现UI设计超越GPT-5

简报：

• 苹果公司发布AI研究成果，展示通过引入专业设计师的直接反馈进行模型微调，可显著提升生成式AI在界面设计（UI）任务上的表现。
• 经过优化的微调小参数模型Qwen3-Coder，在UI设计的逻辑性与美感方面已成功超越目前领先的GPT-5。
• 研究团队邀请21位资深设计师，收集了1,460条包含批注、手绘草稿和直接修改建议的改进日志，将这些高质量的“视觉反馈”转化为奖励模型以进行AI训练。
• 实验表明，Qwen3-Coder仅通过181个草图反馈样本就实现了显著的性能提升。
• 该研究预示未来设计工具将转向“人机高度协作”模式，AI助手将能承担专业UI设计师角色，缩短开发周期。

相关链接：

• 苹果发布 AI 界面设计新突破：Qwen3-Coder 经微调后 UI 设计能力超越 GPT-5

Apple's New AI Research: Small-Parameter Model Qwen3-Coder Achieves UI Design Beyond GPT-5 Through Expert Feedback

Briefing:

• Apple Inc. has released AI research findings, demonstrating that fine-tuning models by incorporating direct feedback from professional designers can significantly enhance the performance of generative AI in User Interface (UI) design tasks. 🍎
• The optimized and fine-tuned small-parameter model, Qwen3-Coder, has successfully surpassed the currently leading GPT-5 in terms of UI design logic and aesthetics.
• The research team invited 21 senior designers and collected 1,460 improvement logs containing annotations, hand-drawn sketches, and direct modification suggestions. These high-quality "visual feedbacks" were then converted into reward models for AI training. 🎨
• Experiments show that Qwen3-Coder achieved significant performance improvement with just 181 sketch feedback samples.
• This research indicates that future design tools will shift towards a "highly collaborative human-AI" model, where AI assistants will be capable of taking on professional UI designer roles, thereby shortening development cycles. ✨

Related Link:

• Apple Unveils AI Interface Design Breakthrough: Fine-Tuned Qwen3-Coder's UI Design Capabilities Surpass GPT-5