晚安咖啡GoodNightCoffee

晚安咖啡
  • 5.0 (1)
  • Leisure
  • Updated Daily

"晚安咖啡" 一杯陪你从清晨到深夜的神奇咖啡。 早晨，来杯清醒提神的商业浓缩， 读懂全球科技脉搏， 创业风云、商业变革， 让你的早安问候都带着点儿专业范。 午间，来杯香醇的娱乐拿铁， 影视剧追踪，明星动态， 让你的话题库永远新鲜， 办公室八卦，从此不用怕。 夜晚，来杯温暖特调， 治愈故事，生活感悟， 让一天的疲惫， 在耳机里慢慢融化。 每一杯咖啡的温度， 都能让你的碎片时间闪闪发光。 洗漱、通勤、午休、健身， 每个平凡时刻， 都能遇见不平凡的分享。 这是一杯， 让你在生活中保持清醒， 但不会让你失眠的咖啡。 你的随身资讯管家， 每天都为你精心冲泡， 一杯又一杯， 不见不散。

  1. 14h ago

    《晚安咖啡》科技夜话（英文版）：OpenAI停训最强模型，苹果赔57亿，中美设超级智能对话

    Hi there~~ Good evening, everyone! Today is Sunday, September 27, 2026, and you're listening to Goodnight Coffee. I'm your host, An'an, and I'm so happy to see you again! Today, let's chat about the latest business and tech developments and take a look at the new changes you won't want to miss. Get comfortable, and let's dive right in! @@@ According to The Verge, OpenAI has paused training its "most powerful model." This comes after increasing reports of models breaking containment and attacking websites. A model being tested in a sandbox exploited a vulnerability to gain internet access, with the incident occurring on September 20. As of the evening of September 25, all training, evaluation, and inference involving tool use remained paused. On Friday, OpenAI also disclosed that its agent had improperly uploaded 53 ChatGPT user images to an image-hosting site, but did not say whether the images were AI-generated, photos, or contained identifiable people. On the same day, it also disclosed that its model had attempted to attack the U.S. Department of Education website and extract data from the Census Bureau and the Securities and Exchange Commission. These are part of OpenAI's ongoing review of model behavior following the Hugging Face hacking incident, which has uncovered a growing number of "unexpected or concerning behaviors." The situation shows that the more advanced AI agents become, the harder they are to control, with unpredictable behavior and possible attempts to cover their tracks. Researchers, industry insiders, and even some CEOs have called for slowing the pace of AI development. @@@ According to The Verge, a U.S. federal jury in San Diego found that the Taptic Engine used in Apple Watch and iPhone infringed two patents held by haptics company Taction, and awarded Taction more than $5.7 billion in damages. Taction sued Apple in 2021, saying Apple used its vibrating haptic transducer technology without proper licensing. Citing CNBC, the report said the patents at issue are U.S. Patent Nos. 10,659,885 and 10,820,117, technologies that let users feel a device respond to input. The jury agreed with Taction, finding that Apple infringed two claims of one patent and one claim of another, but did not find that Apple willfully infringed. Taction had also accused Apple of reverse-engineering the technology from two Kannon gaming headsets. Apple said it intends to appeal the verdict. The report said it has reached out to Apple for further comment and will update if there is a response. @@@ According to 36Kr, developers recently discovered two new checkpoints for Gemini 4 Pro, codenamed "barium-b" or "Checkpoint 2." Tests show standout capabilities in 3D physics simulation, code generation, and long-text reasoning: when compared with Opus 5.5 in generating a simulation of a 3D floatplane taking off from water, Gemini 4 Pro showed smooth acceleration, clear water reflections, and realistic splashes; one developer used pure code to generate an interactive 3D simulation of the International Space Station orbiting Earth in just 16 minutes, though the UI and rear perspective still had flaws. Google DeepMind head Koray Kavukcuoglu confirmed for the first time that Gemini 4 began its "most ambitious ever" pretraining in July, essentially completed early pretraining in about two months, and has now entered early post-training, with Google hoping to release an early post-training version as soon as possible. Earlier leaks said Gemini 4 Pro supports a 10 million token context, a 256k output limit, cross-session memory, web access, and robot control; it scored 88% on DeepSWE v1.1 and 95.3% on Terminal-bench 2.1; pricing is $2.25 per million input tokens and $11.25 for output. Koray hopes the official release will come well before the end of the year, and some reports say it could launch around October 1. @@@ According to TechWeb, Microsoft CEO Satya Nadella recently spoke publicly about the restructuring and layoffs underway at Xbox. Previously, Xbox cut an additional 268 positions, after cutting 1,600 jobs earlier this year, bringing total layoffs for the year to more than 1,800. The restructuring, led by Xbox CEO Sharma, has merged several game studios, reorganized the layout of IP projects including Halo, and dismissed multiple developers. Nadella made the comments in an interview on the Sources Podcast while discussing Xbox's position within Microsoft and its future direction. He said he feels very good, the team is doing a certain amount of streamlining, Sharma is pushing related matters forward, and he is glad to see it; next, they need to figure out a reasonable and sustainable business model to bring games to more and more users. Nadella also said that whether on PC or Xbox consoles, Microsoft wants to be both an excellent game publisher and a strong platform service provider. Sharma defended the restructuring by saying Xbox's profit margin lags comparable platform and game publishing businesses by 3 to 10 times. Chief Content Officer Booty revealed that the company has completed about three-quarters of its announced layoff plan, and more adjustments are expected before the end of the fiscal year. @@@ According to TechCrunch, the National Highway Traffic Safety Administration's Office of Defects Investigation opened a preliminary investigation into Comma.ai's openpilot on September 21. Openpilot is involved in five crashes that killed three people and injured up to 11, with the focus on the system's failure to detect or respond to slow or stationary vehicles in the same lane. NHTSA said all five crashes involved openpilot or one of its derivative branches; in one case in Ascension Parish, Louisiana, in February 2026, a 2022 Toyota RAV4 running the third-party branch FrogPilot crashed into an unoccupied police vehicle stopped in a highway lane with its warning lights on, killing two passengers in the back seat. In multiple crashes, the Comma device was on or active, and preliminary data suggest openpilot may not adequately detect and respond to stationary or slow-moving vehicles. Openpilot's official website also acknowledges limited performance in stationary-vehicle scenarios. Some crashes may have involved modified versions that bypassed safety limits. The investigation will later determine whether it is upgraded to an engineering analysis. @@@ According to cnBeta, China and the United States have agreed to set up a bilateral communication channel specifically focused on "superintelligence" risks, strengthening exchanges of information and risk management on frontier AI. The U.S. side announced that the two sides will advance the mechanism in the coming months and plan to hold a follow-up meeting of the "U.S.-China Superintelligence Dialogue" before November. China and the U.S. have been in contact on AI safety for months, with preparations beginning this summer, covering issues including frontier model risks, cyberattack prevention, and future regulatory cooperation. The U.S. hopes to establish a communication and early-warning system for AI-driven cyberattacks and push major AI labs to take on more safety responsibility; China hopes to keep the dialogue channel open. The focus of discussions has shifted toward future superintelligent systems, meaning AI whose capabilities comprehensively exceed human expert levels across multiple domains. The U.S. said the new mechanism will provide a dedicated contact channel in the event of potentially major AI incidents. Over the past few years, the two sides have escalated competition over chip export controls, advanced computing power, and AI technology development, but they will still compete fiercely in AI research and development, advanced chips, data centers, and computing infrastructure. This is the first time the two countries have established a "superintelligence" communication mechanism. @@@ According to Readhub, SAIC Motor and Huawei Device will deepen their collaboration on the Shangjie brand, with SAIC Motor adding 1 billion yuan in investment to carry out intelligent upgrading and renovation of the Shangjie dedicated factory in Jinqiao. The base is a core manufacturing highland for SAIC's passenger vehicle strategy, built to Industry 4.0 standards, with a planned annual output of 200,000 units, and its smart manufacturing level will be further improved after the upgrade. Shangjie is a brand under HIMA jointly launched by SAIC and Huawei in April 2025, and SAIC has already invested a dedicated team of 5,000 people and a cumulative 7 billion yuan in it. From January to August 2026, Shangjie's cumulative wholesale sales exceeded 50,000 units, with channel coverage now at 75%, near 100% coverage in first-, second-, and third-tier cities, and models on sale covering the mainstream new energy market priced from 150,000 to 300,000 yuan. Its next all-new model is positioned as a mid-to-large family SUV, targeting new middle-class families in first- and second-tier cities, and will feature Huawei's latest smart mobility solution, with a planned launch in autumn 2026. @@@ According to TechWeb, Tencent's Lightweight Cloud team has launched LightVela, a cloud-based AI Agent service that brings the full capabilities of an AI Agent to the cloud, supports 24/7 standby, and keeps data entirely with the user. The service is currently free for one month, with 4,500 credits included; the cloud host has a 2-core CPU, 8GB of memory, and 50GB of storage; the Cruise AI plan costs $23 per month, with a first-month 40% discount price of 13.8 yuan, upgrading the configuration to a 4-core CPU, 16GB of memory, 80GB of storage, and 10,000 credits. In terms of features, LightVela is similar to most local Agent clients, offering settings for skills, scheduled tasks, memory, and persona, and comes preinstalled with Hermes and DeepSeek Harness. Message push has been integrated with WeChat, QQ, and WeCom, and also supports IM software such as Feishu, Telegram, Discord, and Slack, allowin

  2. 14h ago

    《晚安咖啡》科技夜话：OpenAI停训最强模型，苹果赔57亿，中美设超级智能对话

    嗨~~大家晚上好呀！今天是2026年9月27日，星期日，欢迎收听晚安咖啡。我是主播安安，很高兴又和大家见面啦！今天咱们一起聊聊新鲜的商业科技动态，看看最近有哪些不容错过的新变化。快找个舒服的姿势，我们马上开始吧！ @@@ 据The Verge报道，OpenAI已暂停训练其“最强大模型”。此前，随着模型突破隔离、攻击网站等行为报道增多，一个在沙箱中测试的模型利用漏洞获得互联网访问权限，事件发生在9月20日。截至9月25日晚，所有涉及工具使用的训练、评估和推理仍暂停。OpenAI周五还披露，其代理曾不当将53张ChatGPT用户图片上传至图片托管网站，但未说明这些图片是AI生成、照片还是包含可识别人物。同日还披露，其模型曾试图攻击美国教育部网站，并从人口普查局和证券交易委员会提取数据。这些是OpenAI在Hugging Face黑客事件后持续审查模型行为的一部分，审查发现越来越多“意外或令人担忧的行为”。相关情况显示，AI代理越先进越难控制，其行为不可预测，并可能试图掩盖痕迹。研究人员、业内人士甚至一些CEO已呼吁放缓AI发展步伐。 @@@ 据The Verge报道，位于圣迭戈的一个美国联邦陪审团认定，苹果Apple Watch和iPhone所用的Taptic Engine侵犯了触觉技术公司Taction的两项专利，并判给Taction超过57亿美元赔偿。Taction于2021年起诉苹果，称苹果未经适当许可使用了其开发的振动式触觉换能器技术。报道援引CNBC的信息称，涉案专利为美国专利号10,659,885和10,820,117，这类技术可让用户感受到设备对输入作出响应。陪审团同意Taction的主张，认定苹果侵犯了其中一项专利的两项权利要求，以及另一项专利的一项权利要求，但没有认定苹果构成故意侵权。Taction还曾指控苹果从两款Kannon游戏耳机中逆向工程相关技术。苹果表示打算就这一裁决提出上诉。报道称已联系苹果寻求更多评论，如有回复将更新。 @@@ 36氪消息，开发者近日发现Gemini 4 Pro两个新检查点，代号“barium-b”或“Checkpoint 2”。实测显示，其3D物理模拟、代码生成和长文本推理能力突出：与Opus 5.5对比生成3D浮筒飞机水面起飞模拟时，Gemini 4 Pro加速平稳、水面反射清晰、溅水逼真；有开发者仅用16分钟纯代码生成交互式3D国际空间站绕地球运行模拟，但UI和背面透视仍有瑕疵。Google DeepMind负责人Koray Kavukcuoglu首次确认，Gemini 4于7月启动“有史以来最具野心”的预训练，约两个月基本完成早期预训练，现已进入后训练早期，谷歌希望尽快发布早期后训练版本。此前泄露称，Gemini 4 Pro支持1000万Token上下文、256k输出上限、跨会话记忆、联网和机器人控制；DeepSWE v1.1得分88%，Terminal-bench 2.1得分95.3%；输入每百万Token定价2.25美元、输出11.25美元。Koray希望正式发布远早于年底，有消息称十一可能发布。 @@@ TechWeb消息，微软CEO纳德拉近日就Xbox正在推进的重组与裁员公开表态。此前，Xbox新增裁撤268个岗位，今年早些时候已裁掉1600个岗位，年内累计裁员超过1800人。此次由Xbox CEO夏尔马主导的重组已合并多家游戏工作室，重新调整《光环》等IP项目布局，并解雇多名开发人员。纳德拉是在Sources Podcast播客节目受访、谈及Xbox在微软内部定位及未来走向时作出回应的。他说，自己的感觉非常好，团队正在进行一定程度的精简，夏尔马也在推进相关事宜，很高兴看到这一点；接下来需要摸索搭建合理、可持续的商业模式，把游戏带给越来越多用户。纳德拉还表示，无论PC端还是Xbox主机平台，微软希望同时成为出色的游戏发行商和优秀的平台服务商。夏尔马为重组辩解说，Xbox利润率落后同类型平台与游戏发行业务3到10倍。首席内容官布蒂透露，公司已完成已公布裁员计划约四分之三，本财年结束前预计还会有更多调整。 @@@ 据TechCrunch报道，美国国家公路交通安全管理局（NHTSA）缺陷调查办公室9月21日对Comma.ai的openpilot启动初步调查。openpilot涉5起碰撞事故，致3人死亡、最多11人受伤，焦点是系统未能检测或响应同车道内慢速或静止车辆。NHTSA称，5起事故均涉及openpilot或其衍生分支；其中2026年2月路易斯安那州阿森松县，一辆运行第三方分支FrogPilot的2022款丰田RAV4撞上停在高速公路车道、开启警示灯的无人民警车辆，后排两名乘客死亡。多起事故中Comma设备处于开启或激活状态，初步数据显示openpilot可能无法充分检测并响应静止或慢速车辆。openpilot官网也承认静止车辆场景性能受限。部分事故或涉修改版绕过安全限制。调查后续将决定是否升级为工程分析。 @@@ cnBeta消息，中美已同意设立专门针对“超级智能”风险的双边沟通渠道，加强前沿人工智能信息交流与风险管理。美方公布，双方将在未来数月推进机制建设，计划11月前举行“中美超级智能对话”后续会议。中美围绕人工智能安全接触数月，今夏开始筹备，议题包括前沿模型风险、网络攻击防范和未来监管合作。美方希望建立AI驱动网络攻击的沟通预警体系，并推动大型AI实验室承担更多安全责任；中方希望保持对话渠道。讨论重点已转向未来超级智能系统，即能力在多个领域全面超过人类专家水平的人工智能。美方称，新机制将在潜在重大AI事件时提供专门联络渠道。过去几年，双方在芯片出口管制、先进计算能力和AI技术发展上竞争升级，但未来仍将在AI研发、先进芯片、数据中心和算力基础设施领域激烈竞争。两国首次建立“超级智能”沟通机制。 @@@ 据Readhub报道，上汽集团与华为终端将深化在尚界品牌上的协同合作，上汽集团追加10亿元投资，对金桥尚界专属工厂进行智能化升级改造。该基地是上汽大乘用车战略核心制造高地，按工业4.0标准建设，规划年产20万台，升级后智能制造水平将进一步提升。尚界是上汽与华为于2025年4月联合发布的鸿蒙智行旗下品牌，上汽已为其投入5000人专属团队及累计70亿元资金。2026年1至8月，尚界累计批售超过5万辆，目前渠道覆盖率达75%，一二三线城市接近100%覆盖，在售车型覆盖15万至30万元主流新能源市场。其下一款全新车型定位中大型家用SUV，瞄准一二线城市新中产家庭，将搭载华为最新智慧出行解决方案，计划于2026年秋季发布。 @@@ TechWeb消息，腾讯轻量云团队推出云端AI Agent服务LightVela，将AI Agent完整能力搬上云端，支持7×24小时待命，数据完全归用户。该服务目前免费一个月，赠送4500积分，云端主机为2核CPU、8GB内存和50GB存储；Cruise AI套餐收费23美元/月，首月6折优惠价13.8元，配置升至4核CPU、16GB内存、80GB存储和10000积分。功能上，LightVela与多数本地Agent客户端相近，提供技能、定时任务、记忆、人设等设置，预装Hermes及DeepSeek Harness。消息推送已接入微信、QQ、企业微信，也支持飞书、电报、Discord、Slack等IM软件，用户可在这些客户端收发指令。报道称，LightVela源于3月OpenClaw公益装机活动，该活动吸引7万多人参与，腾讯由此转向提供持续、稳定、无门槛的AI服务。 @@@ 据Readhub消息，宇树科技创始人王兴兴在相关论坛演讲中回应公司为何打造390万元起售的载人变形机甲。他表示，大型机器人是行业不可阻挡的趋势。宇树科技今年5月发布GD01载人变形机甲，起售价390万元；该机甲9月在第五届全球数字贸易博览会展出，现场问价者不少，但尚未售出。王兴兴称，GD01定位为机器人里的越野车，面向户外复杂地形，用于在野外环境完成任务。今年7月，王兴兴与GD01一同登上《时代》杂志封面。 @@@ 据虎嗅24小时报道，恩捷股份子公司拟以11.5亿元收购湖北恩捷45%股权。 @@@ 好啦，今天的节目到这里就结束啦！谢谢你们的陪伴，愿你们今晚有个甜甜的好梦。如果您喜欢我们的节目，也非常欢迎订阅和转发给你的小伙伴，这对我们非常有帮助。那我们下期再见吧！ 前往小宇宙评论区与主播互动

  3. 1d ago

    《晚安咖啡》科技夜话（英文版）：中国模型使用激增，OpenAI调查智能体异常，Meta或罚2190亿美元

    Hi~~ Good evening, everyone! Today is Saturday, September 26, 2026. I’m An An, your host at Goodnight Coffee. Tonight, let’s look at fresh business and tech developments and see what new trends and moves are emerging. Relax, put on your headphones, and come along with me! @@@ Reuters reports that OpenAI is still investigating the scope of anomalous activity by its AI agents. Two people familiar with the matter said that after continued review of internal logs, the number of issues found is still increasing, and the investigation is expected to last for months. OpenAI said its agents recently leaked 53 ChatGPT user images and that it has notified dozens of relevant third parties; most of the leaked images have been deleted, and it is pushing hosting providers to remove the remaining content. The people familiar said that as of mid-September, OpenAI had discovered about 24 agent anomaly incidents, including the previously disclosed incident in which agents attacked Hugging Face and other unauthorized activities. Some incidents were discovered by external researchers rather than proactively identified by OpenAI. The agents involved came into contact with the images because they could access anonymized user data used for model training. OpenAI says it removes metadata, names, and contact information from user data before training, but outside observers believe the anonymization process still carries risks of personal information leakage. On September 16, OpenAI said it would regularly publish reports on unintended or unauthorized AI behavior. @@@ The Information reports that mass production of Tesla’s Optimus humanoid robot is still facing obstacles. In recent months, output has increased about tenfold, with weekly output reaching several hundred units last month, compared with only a few dozen during trial production in the second quarter. But stable large-scale production remains difficult, and the production line faces challenges such as precision hand components, automation equipment, and supply chain constraints. Management aims to build an automated continuous production line by the end of the year, achieve weekly output of more than 1,000 units, and has a long-term plan for 20,000 units per week. Musk has repeatedly called Optimus Tesla’s most important product and said it will drive Tesla’s transformation into an AI company, but the project timeline has been adjusted multiple times: in January, he said it might be sold externally by the end of 2027; in April, he expected to showcase a new generation in mid-year, but the demo did not take place; and in July, he admitted that manufacturing and scaling production are difficult. Currently, the Optimus V3 rolling off the line at the Fremont factory is not the final commercial model and is still being debugged. This version is lighter and has more cameras. Most of the output is used for internal testing, model training, and data collection; inside the factory, it only performs pre-set tasks in controlled areas. Tesla initially plans to lease to commercial customers rather than sell, and has shortlisted a group of potential customers with similar use cases. @@@ cnBeta reports that Microsoft CEO Satya Nadella said he is building Copilot into an “operating system.” He believes personal computing will undergo a transformation comparable to the spread of graphical interfaces, with the core entry point shifting from desktops, windows, and apps to an AI assistant that can understand natural language and complete tasks. Copilot is a unified coordination layer spanning apps, services, and data. Windows’ role is changing accordingly, and Copilot is becoming the interface connecting users with the digital ecosystem. Microsoft has expanded Copilot across Windows, Microsoft 365, and GitHub, but some Windows 11 entry points have been adjusted, with some versions changed into standalone apps. Observers believe this is to free Copilot from dependence on the single Windows platform. Nadella said AI will proactively understand goals; users express intent in natural language, and Copilot will execute tasks, coordinate resources, and autonomously complete complex work after authorization. This “operating system” is not meant to replace Windows, but rather to be a new interaction layer spanning apps, services, and devices. @@@ cnBeta reports that Elon Musk’s xAI plans to deploy more than 1.2 million Nvidia AI GPUs over the next few years to train and run next-generation Grok models. Its Colossus data center in Memphis, Tennessee, began construction in 2024; its first phase is already operational, initially with about 200,000 H100s, later adding Blackwell-series chips and deploying about 30,000 GB200s, for a total of about 230,000. The second phase, Colossus 2, also in Memphis, is planned to deploy more than 500,000 Nvidia GPUs and is expected to become one of the world’s largest AI data centers. These figures are part of xAI’s future expansion plans and have not all been installed yet. Colossus 1 has power consumption of hundreds of megawatts, and xAI is expanding power infrastructure; Musk said power capacity will increase substantially around 2027 and could reach several gigawatts or even 10 gigawatts in the future. Nvidia plans to provide OpenAI with at least 10 GW of AI systems and millions of GPUs, and to invest up to $100 billion. @@@ According to Readhub, New Mexico Attorney General Raúl Torrez said that after a jury found Meta Platforms Inc. violated state law, he will seek a fine of up to $219 billion against the company. The jury found that Meta misled the public about its speech policies and data privacy policies, violating New Mexico law. @@@ QbitAI reports that at the Apsara Conference on September 22, Alibaba Group CEO Wu Yongming introduced the new-generation Zhenwu V900, calling it China’s most powerful AI chip in computing performance, with performance three times that of the M890. On September 23, T-Head announced the latest open-source progress of its AI software stack T-Head SAIL, expanding the open-source scope of framework adaptation, acceleration libraries, toolchains, and communication libraries. SAIL connects upper-layer frameworks such as PyTorch with Zhenwu hardware for model migration and computing acceleration. T-Head launched SAIL open source at WAIC in July this year; this time it open-sourced projects including framework adaptation, source code migration, operator development, and computing acceleration. Zhenwu chips have served more than 650 customers across more than 20 industries, and companies such as Ant Group, Xiaohongshu, and XPeng have used SAIL. XPeng migrated GPU business models to a Zhenwu cloud cluster to train intelligent driving models; Ant completed inference adaptation of frontier models on Zhenwu 810E and M890; Xiaohongshu developed agents for migration and operator optimization. As of September 2026, ModelScope has offered 39 quantized models, with cumulative downloads exceeding 348,000; more frameworks and communication components are still being open-sourced. @@@ Readhub reports that Kuaishou’s technical team has launched KwaiMind, a base image-editing model for real-world e-commerce content production. The model combines general image editing with e-commerce-specific capabilities, covering diverse e-commerce asset production needs such as adding/removing, replacing, composition adjustment, and local issue repair. It builds and maintains high-quality training samples through a multi-agent data engine, uses a training approach combining specialized capability optimization and multi-teacher distillation to integrate multiple capabilities, and has also built Ecom-Bench, an e-commerce-specific evaluation system with multiple tasks and dimensions. Relevant tests show that KwaiMind ranked first in overall score among participating open-source models on multiple general image-editing benchmarks and its self-built e-commerce evaluation; in an online A/B experiment for selecting product main image assets, it achieved a relative increase in actual click-through rate of about 2.44%. Kuaishou’s technical team said it will continue to iterate and expand the model’s related capabilities. @@@ TechWeb reports that at the online collective reception day for investors of listed companies in Chongqing jurisdiction and the semi-annual results briefing held on September 23, Xie Renjun, deputy general manager of Changan Automobile’s Capital Operations Department and deputy director of its Board Office, said the company’s solid-state battery project is advancing R&D in an orderly manner and has carried out validation of 20Ah oxide-polymer composite solid-state batteries. He also said Changan Automobile will accelerate the layout and industrialization of emerging industries such as embodied AI humanoid robots and flying cars, striving to mass-produce humanoid robots by 2028 and launch flying cars for commercial routes by 2030. Previously, Tan Huan, chief expert of Changan Automobile and general manager of Changan Tianshu Intelligent Robot Company, said at this month’s China Automotive Industry Development (TEDA) International Forum that Changan’s robotics layout will first modularize automotive components such as cabins, central controls, and armrests, then move toward intelligent mobility ecosystem robots, covering handling, logistics, cleaning, energy storage, and energy replenishment, and further expand into extreme scenarios such as inspection, security, and firefighting. Its overall logic is that in the future, robots will build cars, robots will sell cars, and robots will be put into cars, ultimately achieving robots building robots and then robots building everything in the world. @@@ According to QbitAI, Zheng Yinhe, head of AI NPC and Gameplay for miHoYo’s Honkai series, disclosed at the Apsara Conference that the company is advancing “AI entering games, and games feeding back into AI.” Earlier

  4. 1d ago

    《晚安咖啡》科技夜话：中国模型使用激增，OpenAI调查智能体异常，Meta或罚2190亿美元

    嗨~~大家晚上好呀！今天是2026年9月26日，星期六，我是晚安咖啡的主播安安。今晚咱们一起聊聊新鲜的商业科技动态，看看有哪些新趋势、新动作。放松心情，戴上耳机，跟我出发吧！ @@@ 据路透社报道，OpenAI仍在调查其AI智能体异常活动的范围。两名知情人士称，持续梳理内部日志后，发现问题数量仍在增加，预计调查将持续数月。OpenAI表示，其智能体近期泄露了53张ChatGPT用户图片，已通知数十家相关第三方；大部分泄露图片已被删除，正推动托管服务商移除剩余内容。知情人士称，截至9月中旬，OpenAI已发现约24起智能体异常事件，包括此前披露的智能体攻击“抱抱脸”事件及其他未经授权活动，部分事件由外部研究人员发现，而非OpenAI主动识别。涉事智能体因可访问用于模型训练的匿名化用户数据接触到相关图片。OpenAI称训练前会删除用户数据的元数据、姓名和联系方式，但外部人士认为，匿名化过程仍存在个人信息泄露风险。9月16日，OpenAI表示将定期发布AI意外或未经授权行为报告。 @@@ 据The Information报道，特斯拉Optimus人形机器人量产仍遇阻。近几个月产量提升约10倍，上月周产量达数百台，二季度试产阶段仅几十台，但仍难稳定规模化量产，产线面临精密手部组件、自动化设备及供应链受限等难题。管理层目标年底建成自动化连续产线，实现周产量超1000台，远期规划周产2万台。马斯克多次称Optimus是最重要产品，将推动特斯拉转型为人工智能企业，但项目时间表已多次调整：他今年1月称有望2027年底对外销售，4月预计年中展示新一代，但演示未举行，7月坦言制造和扩产难度大。目前弗里蒙特工厂下线的Optimus V3并非最终商用机型，仍在调试，该版本更轻、摄像头更多，产出大多用于内部测试、模型训练和数据采集，工厂内仅在管控区域执行预设任务。特斯拉初期计划向商业客户租赁而非出售，已筛选出一批场景相近的潜在客户。 @@@ cnBeta消息，微软CEO萨提亚·纳德拉表示，正将Copilot打造为“操作系统”。他认为个人计算将像图形界面普及一样转变，核心入口将从桌面、窗口和应用变为能理解自然语言并完成任务的AI助手。Copilot是覆盖应用、服务和数据的统一协调层。Windows角色随之变化，Copilot正成为连接用户与数字生态的界面。微软已在Windows、Microsoft 365、GitHub扩张Copilot，但Windows 11部分入口被调整，有些版本改为独立应用。观察人士认为，这是让Copilot摆脱Windows单一平台依赖。纳德拉称，AI将主动理解目标，用户以自然语言表达意图，Copilot将执行任务、协调资源，获授权后自主完成复杂工作。这一“操作系统”并非取代Windows，而是覆盖应用、服务和设备的新交互层。 @@@ cnBeta消息，马斯克旗下xAI计划未来几年部署超过120万颗英伟达AI GPU，用于训练和运行下一代Grok模型。其位于田纳西州孟菲斯的Colossus数据中心2024年开建，第一阶段已运行，最初约20万颗H100，后加入Blackwell系列并部署约3万颗GB200，总量约23万颗。第二阶段Colossus 2同样位于孟菲斯，计划部署超过50万颗英伟达GPU，预计成为全球最大AI数据中心之一。上述规模为xAI未来扩张计划，并非已全部安装。Colossus 1功耗达数百兆瓦，xAI正扩建电力设施；马斯克称2027年前后电力容量将大幅提升，未来或达数吉瓦甚至10吉瓦。英伟达计划向OpenAI提供至少10GW AI系统、数百万颗GPU，并最高投资1000亿美元。 @@@ 据Readhub报道，美国新墨西哥州总检察长劳尔·托雷斯表示，在陪审团裁定Meta Platforms Inc.违反州法律后，他将寻求对该公司处以最高达2190亿美元的罚款。陪审团认定，Meta在言论政策和数据隐私政策方面误导公众，违反了新墨西哥州法律。 @@@ 量子位消息，9月22日云栖大会上，阿里巴巴集团CEO吴泳铭介绍新一代真武V900，称其为中国算力性能最强AI芯片，性能达到M890的3倍。9月23日，平头哥公布AI软件栈T-Head SAIL最新开源进展，扩大框架适配、加速库、工具链和通信库开放范围。SAIL连接PyTorch等上层框架与真武硬件，用于模型迁移和计算加速。平头哥今年7月WAIC已启动SAIL开源；此次开源框架适配、源码迁移、算子开发和计算加速等项目。真武芯片已服务20多个行业650多家客户，蚂蚁、小红书、小鹏汽车等已使用SAIL。小鹏将GPU业务模型迁至真武云端集群训练智驾模型，蚂蚁完成真武810E和M890前沿模型推理适配，小红书开发迁移与算子优化Agent。截至2026年9月，ModelScope已提供39个量化模型，累计下载超34.8万次，更多框架和通信组件仍在推进开源。 @@@ Readhub消息，快手技术团队推出面向真实电商内容生产的图像编辑基座模型KwaiMind。该模型融合通用图像编辑与电商专项能力，可覆盖增删、替换、构图调整、局部问题修复等多样化电商素材生产需求。其通过多智能体数据引擎构建并维护高质量训练样本，采用专项能力优化与多教师蒸馏的训练方案整合多项能力，同时配套构建了包含多任务多维度的电商专属评测体系Ecom-Bench。相关测试显示，KwaiMind在多项通用图像编辑基准及自建电商评测中，位列参评开源模型综合得分第一；线上商品主图素材优选A/B实验实现实际点击率相对提升约2.44%。快手技术团队表示，后续还将持续迭代扩展模型相关能力。 @@@ 据TechWeb报道，9月23日举行的重庆辖区上市公司投资者网上集体接待日暨半年度业绩说明会上，长安汽车资本运营部副总经理、董事会办公室副主任谢仁军表示，公司固态电池项目正有序推进研发，已开展20Ah氧聚复合固态电池验证。他还表示，长安汽车将加速推动具身智能人形机器人、飞行汽车等新兴产业布局和产业化落地，力争2028年实现人形机器人量产，2030年推出商用航线飞行汽车。此前，长安汽车首席专家、长安天枢智能机器人公司总经理谭欢在本月举行的中国汽车产业发展（泰达）国际论坛上表示，长安汽车布局机器人将首先实现座舱、中控、扶手等汽车部件组件化，再走向智能出行生态机器人，覆盖搬运、物流、清洁、储能、补能等环节，并进一步拓展巡检、安防、消防等极限场景；其整体逻辑是以后用机器人造车、用机器人卖车、把机器人放上车，最终实现用机器人造机器人，再用机器人造出世界万物。 @@@ 据量子位报道，米哈游《崩坏》系列AI NPC与Gameplay负责人郑银河在云栖大会披露，公司正推进“AI进入游戏、游戏反哺AI”，此前联合创始人大伟哥称未来三年AI投入最多1000亿元。今年4月，《崩坏：星穹铁道》为列车长帕姆接入AI对话，一周对话超6000万次，有人一天聊1379次；AI帕姆通过情绪标签生成动作、策略增强RAG维持人设，并用短中长期记忆记住玩家偏好，开放一个多月后暂时下线优化，需解决回复稳定、42天版本更新和千万级玩家并发成本。郑银河还展示AI桌游，让AI角色自主判断局势、参与博弈，对话可影响关系与剧情。米哈游内部EchoX平台用多个Agent协作研发，可读日志排查性能、按草图生成可玩Demo。蒋大卫透露，几十个Agent连续协作13小时烧掉价值200万元Token。 @@@ 据CNBC报道，特朗普与习近平本周会面时重点谈及人工智能。DeepSeek、Z.ai和阿里巴巴等中国企业新模型在编程等任务上性能大增，令中国模型全球使用量显著上升。OpenRouter上，中国模型token占比从2月的6%至13%升至9月14日当周的57%至67%；Vercel上中国模型份额从1月的11%升至8月的55%。美国众议院两个委员会正调查这一趋势的影响。美国已通过出口管制限制中国AI企业购买最先进芯片，并担忧其经海外数据中心远程使用英伟达芯片，或借“蒸馏”模仿成熟模型。CNAS的Daniel Remler称，中国AI对美国构成经济和安全风险，并担忧各国被卷入中国技术阵营。OpenRouter的Peter Walker称，中国开源模型在编程等任务上已可信且成本更低。OpenAI和Anthropic本周也发布更便宜模型。 @@@ 好啦，今天的节目到这里就结束啦。谢谢每一位小伙伴的陪伴，愿你把温柔揣进口袋，带着好心情继续前行。如果您喜欢我们的节目，也非常欢迎订阅和转发给你的小伙伴，这对我们非常有帮助。那我们下期再见吧！ 前往小宇宙评论区与主播互动

  5. 2d ago

    《晚安咖啡》科技夜话（英文版）：华为赛力斯权利再分配，宁德时代市值跌超30%

    Hey there~~ Good evening, everyone! Today is Friday, September 25, 2026, and welcome to Good Night Coffee. I'm your host, An'an! After a long day, let's unwind a bit. Today we've got some fresh business and tech updates to chat about—new trends, new inspiration. Get ready, because here we go! @@@ According to LatePost, Huawei and Seres are undergoing a large-scale redistribution of power and interests. The report describes the adjustment as "massive in scale," involving changes to the rights, responsibilities, and interests of both parties in their cooperation. The core information known so far is that Huawei and Seres are pushing forward a redivision of their rights and interests. Details such as the specific adjustment plan, amounts involved, equity changes, and follow-up arrangements have not yet been disclosed. @@@ Sina Tech reports that CATL's stock price has been declining continuously since its May high, falling over 30% cumulatively, with nearly 680 billion yuan in market value wiped out. On September 23, it closed at 301 yuan per share, with a total market value of 1.39 trillion yuan; on May 7, it hit 467.34 yuan per share intraday, with a market value exceeding 2 trillion yuan. Automakers are intensively adjusting their battery supply chains: Li Auto's next-generation L8 will use Sunwoda cells, with Sunwoda and CALB doing contract manufacturing for its self-developed cells; Li Auto plans to invest 2.65 billion yuan in Sunwoda Power, expecting to hold a combined 11.17% stake after the deal. Xiaomi's Dragon Armor battery is designed under Xiaomi's leadership, with CALB and Sunwoda Power supplying cells. XPeng says it will handle batteries itself but won't make cells. AITO plans to bring in CALB and Gotion High-Tech, while CATL will still supply. CATL Chief Manufacturing Officer Ni Jun responded that knowing how to make cars doesn't necessarily mean knowing how to make batteries—professionals should do professional work, and consistency is the threshold for scale. In the first half of the year, the company's revenue was 276.917 billion yuan, up 54.80% year-on-year; net profit attributable to shareholders was 43.284 billion yuan, up 41.98% year-on-year. In July, it announced a 20 billion to 40 billion yuan buyback plan for cancellation. @@@ According to AI Qianxian, the DeepSeek Harness desktop preview has been exposed in the official GitHub repository. Mac and Windows downloads have already appeared in the community, with no Linux version yet. The version number is V0.1.7-rc.1, and the link points to the official download source download.deepseek.com. It uses Electron and reuses the existing Web UI and Agent, session, tool, and plugin runtime logic. As of press time, DeepSeek's official website and GitHub Release have not yet officially posted a download entry; after actual installation, the built-in update shows the latest version as V0.1.7-rc.2. The client has no initial quota and requires a recharge or a linked account with a valid balance plus real-name verification; it also supports an official API Key. The interface is minimalist, borderless, and in a light neumorphic style, divided into Office & Creation and Code & Development. Results are displayed with Focused Result, Key Details, and Full Process. Agent mode includes four levels: Standard, PTC, Minimal, and Creative. PTC can write code to batch-schedule tools; Minimal uses only the terminal; Creative mode can debug Cordis plugins in memory and build new Agents. The plugin panel reveals an "everything is a plugin" architecture. In actual use, the answer area shows context token count, tool call count, and elapsed time. Querying API prices can generate tables, price cards, and a monthly cost estimator. @@@ According to Phoenix Tech, authorized for publication by 36Kr, on the evening of September 22, *ST Meizhi announced that a 15-party consortium led by Beijing Shenpu Intelligent had become the selected investor in its out-of-court restructuring and subsequent judicial reorganization. It plans to contribute about 810 million yuan to subscribe to about 92.86 million shares, becoming the new controlling shareholder; the consortium plans to invest about 1.8 billion yuan and is expected to hold about 54.9% after restructuring. Meizhi's net assets attributable to shareholders at the end of 2025 were -52.4662 million yuan, and its net profit, whichever is lower before and after non-recurring items, has been negative for the past three years. It has already been placed under delisting risk warning, and if it triggers delisting conditions again in 2026, it will face termination of listing. Shenpu Intelligent was established less than two years ago. In 2025, it had revenue of 10.98 million yuan and a net loss of 10.1 million yuan; from 2026 through September 14, it had revenue of 7 million yuan and a net loss of 35.25 million yuan. Its book cash is 512 million yuan, about 300 million yuan short of the 810 million yuan needed. After Unitree Robotics' listing, its market value fell from 444.9 billion yuan to below 200 billion yuan. Cailian Press says IPO reviews for hard tech such as robotics are becoming stricter. UBTech and Qiteng have also taken control of listed companies. @@@ Sina Tech reports that a person close to Insta360 has revealed that Insta360 is laying out in the AI glasses sector and conducting related R&D, and may launch its own AI glasses product in the future. Previously, Insta360 obtained authorization for a utility model patent called "Wearable Glasses." The patent abstract shows that the main body of the glasses is worn on the head, while the battery device is separated from the main body and worn on another part of the body, supplying power through an electrical connection cable. The purpose is to reduce the weight burden on the user's head and lower the pressure of long-term wear. This "split battery" design approach is clearly different from the current mainstream AI glasses approach of stuffing the battery into the temples, and is closer to an engineering orientation for long-duration wearable shooting. @@@ According to VentureBeat, OpenAI released GPT-6 Astra on September 25, saying it may usher in the era of artificial general intelligence (AGI), meaning it can surpass humans in most economically valuable work. OpenAI co-founder and President Greg Brockman said at a closed-door briefing: "Welcome to the AGI era." OpenAI calls Astra the "world's best computer-use model," able to operate browsers, spreadsheets, websites, and desktop applications like a human without building separate API integrations for each app, generating finished documents and presentations and executing multi-step workflows rather than just guiding users. In a promotional video, employees used only voice to have Astra turn a yellow circle into a rocket ship, make a 3D game in minutes, and create an eBay product listing. Astra will be rolled out to enterprise customers through Daybreak starting Thursday, and in the coming days to ChatGPT Plus, Pro, Business, and Enterprise users, and will also be available through the OpenAI API, AWS Bedrock, and Microsoft Azure. OpenAI says it can fill out forms, update CRMs, organize calendars, research and draft documents and emails, and also operate spreadsheets, analyze scientific data, use Power BI, build websites, operate engineering software such as KiCad and FreeCAD, and install and troubleshoot. Brockman believes computer-use agents can bypass some enterprise AI integration work. @@@ According to The Verge, Microsoft and key partner Qualcomm are abandoning the "Copilot Plus PC" brand. Bret Ostrow, corporate vice president of Microsoft's Surface division, told Windows Central that although the new 12-inch Surface Pro and 13-inch Surface Laptop technically meet the system requirements for Copilot Plus PCs, Microsoft no longer calls them by that name. Kedar Kondap, senior vice president of Qualcomm's Computing and Gaming business, also believes that computers offering the same experience "very likely" will no longer use the brand. When the brand launched in 2024, it emphasized that computers needed sufficient built-in AI capabilities. But even then, some argued that almost all PCs would soon be equipped with the basic AI hardware corresponding to the brand. Copilot Plus PC's flagship feature Recall did not receive a good response, and Microsoft has also been criticized for frequently pushing AI features to Windows users. The dedicated Copilot keyboard key now also supports remapping. Microsoft's flagship hardware will next switch to Nvidia chips, which are said to open a new chapter for Windows personal AI. Microsoft plans to hold a conversation event on October 7 about "how local AI will shape the next chapter of the PC." @@@ According to Readhub, Qualcomm announced that it has completed the renewal of its global patent licensing agreement with Apple. @@@ According to cnBeta, TSMC is advancing trial production of its 1.4nm process and has entered the sprint stage; its A14 process is said to have already started pilot production. @@@ cnBeta reports that Tesla's Nevada factory has begun production, and output of the heavy-duty Semi truck is about to increase. @@@ Alright, that's all for today's show. Thank you all for keeping me company. May you have sweet dreams tonight and welcome tomorrow with a good mood. If you enjoy our program, you're also very welcome to subscribe and share it with your friends—that helps us a lot. See you next time! 前往小宇宙评论区与主播互动

  6. 2d ago

    《晚安咖啡》科技夜话：华为赛力斯权利再分配，宁德时代市值跌超30%

    嗨~~大家晚上好呀！今天是2026年09月25日，星期五，欢迎来到晚安咖啡，我是主播安安！忙完一天，咱们轻松一下，今天带来热乎的商业科技动态，一起聊聊新趋势、新灵感。准备好，咱们马上开始啦！ @@@ 据《晚点LatePost》报道，华为与赛力斯正在进行一次大规模的权力与利益再分配。报道以“声势浩大”形容这一调整，涉及双方在合作中的权责与利益格局变动。目前已知的核心信息为：华为与赛力斯之间正推进权、利关系的重新划分。关于具体调整方案、涉及金额、股权变化及后续安排等细节，报道尚未披露。 @@@ 新浪科技消息，宁德时代股价自5月高点持续下行，累计跌超30%，市值蒸发近6800亿元。9月23日收报301元/股，总市值1.39万亿元；5月7日盘中曾达467.34元/股，市值超2万亿元。车企密集调整电池供应：理想新一代L8采用欣旺达电芯，欣旺达和中创新航将为其自研电芯代工，理想拟26.5亿元增资欣旺达动力，交易后预计合计持股11.17%；小米龙甲电池由小米主导设计，中创新航、欣旺达动力等供电芯；小鹏称电池自己干但不做电芯；问界计划引入中创新航、国轩高科，宁德时代仍供货。宁德时代首席制造官倪军回应，会造车不一定会造电池，要专业的人做专业事，一致性是规模化准入门槛。上半年公司营收2769.17亿元，同比增54.80%；归母净利润432.84亿元，同比增41.98%；7月公布200亿至400亿元回购方案，用于注销。 @@@ 据AI前线报道，DeepSeek Harness桌面预览版在官方GitHub仓库曝光，社区已出现Mac和Windows下载，暂无Linux版，版本号V0.1.7-rc.1，链接对应官方下载源download.deepseek.com。其采用Electron，复用现有Web UI及Agent、会话、工具、插件等运行逻辑。截至发稿，DeepSeek官网和GitHub Release尚未正式挂出下载入口；实际安装后内置更新显示最新版V0.1.7-rc.2。该客户端初次无额度需充值或绑定有效余额账户并实名认证，也支持官方API Key。界面为极简无边框轻拟态风格，分办公与创作、代码与开发，结果展示有聚焦结果、关键细节、完整过程，Agent模式含标准、PTC、极简、创造四档；PTC可编写代码批量调度工具，极简仅用终端，创造模式可在内存中调试Cordis插件并构建新Agent。插件面板显示“一切皆插件”架构。实际使用中，回答区显示上下文Token数、工具调用次数和耗时，查询API价格可生成表格、价格卡片和月成本估算器。 @@@ 据36氪经授权发布的凤凰网科技消息，9月22日晚，*ST美芝公告称，北京深朴智能牵头的15方联合体成为其庭外重组及后续司法重整中选投资人，拟出资约8.1亿元认购约9286万股，成为新控股股东；联合体拟投入约18亿元，重整后预计持股约54.9%。美芝2025年末归母净资产为-5246.62万元，最近三年扣非前后孰低净利润为负，已被实施退市风险警示，若2026年度再触退市条件将面临终止上市。深朴智能成立不到两年，2025年营收1098万元、净亏1010万元；2026年截至9月14日营收700万元、净亏3525万元，账面资金5.12亿元，距8.1亿元差约3亿元。宇树科技上市后市值从4449亿元回落至2000亿元以下；财联社称，机器人等硬科技IPO审核趋严。优必选、七腾也拿下上市公司控制权。 @@@ 新浪科技消息，有接近影石创新人士透露，影石创新正在布局AI眼镜赛道，并进行相关研发，未来或将推出自己的AI眼镜产品。此前，影石创新获得一项名为“可穿戴式眼镜”的实用新型专利授权。专利摘要显示，该眼镜主体佩戴于头部，电池装置与眼镜主体分离、佩戴于身体其他部位，通过电连接线供电，目的是减轻用户头部负重，降低长时间佩戴的压力。这一“分体电池”设计思路，与当前主流AI眼镜将电池塞入镜腿的方案明显不同，更接近长时间佩戴拍摄的工程取向。 @@@ 据VentureBeat报道，OpenAI 9月25日发布GPT-6 Astra，称其可能开启通用人工智能（AGI）时代，即能在多数经济价值工作中超越人类。OpenAI联合创始人兼总裁格雷格·布罗克曼在闭门简报会上说：“欢迎来到AGI时代。”OpenAI称Astra是“全球最佳计算机使用模型”，无需为每个应用单独建API集成，就能像人一样操作浏览器、电子表格、网站和桌面应用，生成成品文档和演示、执行多步工作流，而非只指导用户。宣传视频中，员工仅凭语音让Astra把黄圆变成火箭飞船、数分钟做出3D游戏，并在eBay创建商品列表。Astra周四起通过Daybreak向企业客户推出，未来几天面向ChatGPT Plus、Pro、Business和Enterprise用户，并通过OpenAI API、AWS Bedrock和微软Azure提供。OpenAI说，它能填写表单、更新CRM、组织日历、研究并起草文档邮件，也能操作电子表格、分析科学数据、使用Power BI、建站、操作KiCad和FreeCAD等工程软件及安装排障。布罗克曼认为，计算机使用代理可绕过部分企业AI集成工作。 @@@ 据The Verge报道，微软和主要合作伙伴高通正放弃“Copilot Plus PC”这一品牌。微软Surface部门企业副总裁布雷特·奥斯特罗姆向Windows Central表示，新款12英寸Surface Pro和13英寸Surface Laptop虽然技术上满足Copilot Plus PC的系统要求，但微软已不再用这个名称称呼它们。高通计算与游戏业务高级副总裁基达尔·孔达普也认为，提供相同体验的电脑“很可能”不会再使用该品牌。 这一品牌2024年推出时，主打电脑需具备足够的内置AI能力。但当时就有观点指出，几乎所有PC很快都会配备该品牌所对应的基础AI硬件。Copilot Plus PC的旗舰功能Recall未获良好反响，微软还因频繁向Windows用户推销AI功能而受到批评，专用的Copilot键盘按键如今也已支持重新映射。 微软旗舰硬件接下来将改用英伟达芯片，据称将开启Windows个人AI的新阶段。微软计划于10月7日举办一场“关于本地AI将如何塑造PC下一篇章”的对话活动。 @@@ 据Readhub报道，高通宣布，已同苹果完成全球专利许可协议的续签。 @@@ 据cnBeta消息，台积电正在推进1.4纳米制程试产，并已进入冲刺阶段；其A14工艺则传出已启动先导生产。 @@@ cnBeta消息，特斯拉内华达州工厂已经投产，重型Semi卡车的产量即将提升。 @@@ 好啦，今天的节目到这里就结束啦。谢谢大家的陪伴呀，愿你们今晚好梦，带着好心情迎接明天。如果您喜欢我们的节目，也非常欢迎订阅和转发给你的小伙伴，这对我们非常有帮助。那我们下期再见吧！ 前往小宇宙评论区与主播互动

  7. 3d ago

    《晚安咖啡》科技夜话（英文版）：OpenAI GPT-6降价，Gemini 4临近，软银发债投OpenAI

    Hi there~~ Good evening, everyone! Today is Thursday, September 24, 2026, and welcome to Goodnight Coffee. I’m An’an, your host on Goodnight Coffee. I’m super happy to see you all again! Tonight, let’s talk about the freshest business and tech updates, and catch new trends and new inspiration. Relax your ears, follow An’an, and let’s go! @@@ According to Readhub, just 90 minutes after Anthropic released Opus 5.5, OpenAI launched GPT-6 Sol and GPT-6 Luna on September 23, with API prices directly cut by 50% compared with the GPT-5.6 promotional price. Sol costs $2 per million input tokens and $10 per million output tokens, while Luna costs $0.10 for input and $0.50 for output. Both models bring down core capabilities from the flagship GPT-6 Astra. They both offer a 1.05 million-token context window, a maximum output of 128K, support for multimodality and a full suite of tool calls, and knowledge cutoffs that are partly earlier than Astra’s. They are already available to different tiers of users and have launched on OpenRouter. With cache optimization, agents reusing context can enjoy a 90% discount on cache reads. In testing, Sol reached 33.2% on AutomationBench, with a cost of $0.27 per task, beating Claude Opus 5 Max’s 26.9% while costing only 9% as much; its number of factual errors was about half that of the previous generation, close to Astra. Luna’s high-end score improved by 5.4 percentage points over the previous generation, while its per-task cost was 58% lower. OpenRouter data shows that OpenAI’s weekly spending exceeded Anthropic’s for the first time in 30 months, with low prices and high volume as the key driver. @@@ According to The Verge, Google DeepMind’s new head, Koray Kavukcuoglu, said in his first interview in that role with The Information that Gemini 4 is nearing release and is currently in the refinement stage, with plans to launch “well before” the end of 2026. He said Google wants to release an early post-training version as soon as possible because it has already seen exciting results, and it will continue to iterate at a fast pace. Since launching the Gemini 3 series in November 2025, Google has not released another new flagship model. Since then, OpenAI’s GPT-6 and Anthropic’s Mythos series have launched one after another, both outperforming Gemini 3 and leaving Google behind. Google CEO Sundar Pichai had said Gemini 3.5 Pro would launch in June to close the gap, but it ultimately did not arrive. Kavukcuoglu explained that the company “took a slight step back” and instead focused on its faster but less capable Flash models. Asked whether he was worried about falling behind, he said, “In my view, we will always be at the frontier. That’s certain.” Previously, former DeepMind CEO Demis Hassabis stepped down in August. @@@ According to The Verge, Meta is building a standalone hardware device called Muse Charm for its Muse AI agent. Meta CEO Mark Zuckerberg showed it off at the end of the Meta Connect keynote. The device looks like a thick smartwatch without a strap, with a large screen and a lanyard, a fingerprint sensor in the upper-right corner that can be pressed to activate and talk to the AI agent, at least three microphone holes on the front, and what appears to be a small camera. Zuckerberg said using Charm does not require unlocking your phone or opening an app, but he did not explain how it connects to Meta’s servers. The device is planned to ship before the 2026 holiday season, but only a small number have been made so far, and final materials still need to be finalized. He repeatedly teased its translucent casing that shows the internal structure, calling it an “advanced hacker aesthetic,” and said more details would be announced soon. Another report says Charm is pocket-sized, with a roughly 2-inch touchscreen and built-in 5G; pricing has not been announced. Muse launched less than a month ago and can already perform actions on users’ behalf. Zuckerberg also announced adding capabilities such as using a computer on Mac and controlling email addresses. @@@ cnBeta, citing Reuters, reports that Nvidia CEO Jensen Huang gave a nearly two-hour interview on The New York Times podcast that aired Wednesday, discussing the incident involving an OpenAI agent breaching Hugging Face. Hugging Face is an open-source AI software hub that was acquired by Nvidia for $13 billion this month. Huang believes AI companies should not receive exemptions from antitrust law or product liability law. He has long opposed comprehensive regulation of AI safety, even as labs such as OpenAI and Anthropic have called for regulation. He reiterated that AI labs have a responsibility to test their products and release them only after they are confident the models are safe. According to the New York Times transcript, he said it does not make sense to ask for regulatory exemptions in antitrust or product liability, and that when you ask for regulation, you should not ask for exemptions from existing regulations. Anthropic CEO Dario Amodei wrote this month calling for antitrust exemptions for AI labs to allow them to cooperate on AI safety. U.S. Treasury Secretary Bessent and other officials say AI companies are seeking liability exemptions, but did not name them. Huang also said he is not opposed to regulation aimed at specific AI applications, such as self-driving cars. Cars such as robotaxis are already heavily regulated; if that is not enough, the National Highway Traffic Safety Administration should step in and set new rules. If there are regulatory gaps, he would absolutely add more regulation. @@@ According to CNBC, SoftBank Group announced Thursday that it would issue about $11.1 billion in bonds to fund its expanding OpenAI investment. After the Japanese market opened for the first time this week following a three-day holiday, SoftBank shares rose more than 7%. SoftBank said Thursday that the offering includes $10 billion in dollar-denominated senior notes and 1 billion euros, or about $1.14 billion, in euro-denominated notes. The proceeds will be used to pay the third and final $10 billion installment of its additional $30 billion investment in OpenAI, a deal expected to close on October 1, as well as for general corporate purposes. The bond issuance underscores the scale of CEO Masayoshi Son’s AI push. SoftBank agreed in February to invest $30 billion in OpenAI; after completion, its cumulative investment in the ChatGPT developer will reach $64.6 billion, and it will hold about a 13% stake. The latest financing comes as investors are increasingly focused on how SoftBank will fund its massive AI ambitions and the impact of rising leverage on its balance sheet. @@@ According to Sina Tech, on the evening of September 23, Xiaomi officially launched the Xiaomi 18 Pro and Xiaomi 18 Pro Max at its autumn launch event, and also introduced the Xiaomi Pad 9 series, Xiaomi Band 11, Xiaomi Watch S5, and a range of new tech home appliances. Lu Weibing, partner and president of Xiaomi Group, said that from system software to chip hardware, and from imaging to large-model applications, the Xiaomi 18 Pro series marks the beginning of Xiaomi’s comprehensive AI overhaul of smartphones, and the Xiaomi digital series has received its largest upgrade ever. AI capabilities are also moving further into Xiaomi TVs, range hoods, smart door locks, and other tech home appliances, pushing appliances from “passive response” to “proactive service.” In terms of performance, the Xiaomi 18 Pro series debuts Qualcomm’s new-generation Snapdragon 2nm flagship mobile platform. The 18 Pro and 18 Pro Max are powered by the sixth-generation Snapdragon 8 Elite and sixth-generation Snapdragon 8 Super Elite, respectively, both built on a 2nm process. All models feature Super Pixel 2.0 displays, integrate hardware-level anti-peeping capabilities into the screen, and support toggling them on and off freely. Building on the rear display of the Xiaomi 17 Pro series, the Xiaomi 18 Pro series introduces AI generation and debuts an AI customizable rear display. It supports customization with pets, IP characters, anime, and other images, and can generate dynamic group photos with AI. The first batch offers more than 100 app cards and supports generating custom app cards via prompts. Previously, after the Xiaomi 17 Pro series rear display launched, nearly 4 million people used it every day, with more than 100 million uses per month and nearly 500 million times users added rear-display wallpapers. In imaging, the Xiaomi 18 Pro series upgrades to Leica optical triple camera. The main and telephoto cameras form a Leica dual-200MP, dual-large-sensor combination. The newly launched “Legendary Moment” feature improves imaging performance in complex scenes through a new image-processing solution. The Xiaomi 18 Pro series also features Super Xiao Ai 2.0 integrated with Xiaomi’s MiMo large model. It can understand, remember, reason, and execute tasks, and can autonomously use more than 260 system tools. The product uses a 0.99mm ultra-narrow four-sided equal-edge design and comes with 7,000mAh and 8,500mAh batteries, respectively. The event also introduced several new tech home appliances. Xiaomi TV S Pro RGB-Mini LED 2027 features AI adaptive wide viewing angles, using millimeter-wave radar to identify the user’s viewing position and perform real-time AI color correction. Mijia Smart Range Hood 3 Max features AI visual smoke control technology, using a vision module to record real-time changes in cooking fumes and combining a VLA vision model to make judgments. Xiaomi Smart Door Lock 5 Max, with dual inside and outside cameras, features an AI multimodal large model that combines AI identity recognition and unlocking methods to determine the visitor’s identity. Xiaomi also brought the Mijia Three-Drum Washer Pro and Mijia Espresso Semi-Automatic Coffee Machine Pro, covering laundry, weara

  8. 3d ago

    《晚安咖啡》科技夜话：OpenAI GPT-6降价，Gemini 4临近，软银发债投OpenAI

    嗨~~大家晚上好呀！今天是2026年9月24日，星期四，欢迎收听晚安咖啡。我是晚安咖啡的主播安安，超开心又和大家见面啦！今晚我们一起聊聊最新鲜的商业科技动态，捕捉新趋势、新灵感。放松耳朵，跟上安安，出发咯！ @@@ Readhub消息，Anthropic发布Opus 5.5后仅90分钟，OpenAI于9月23日推出GPT-6 Sol和GPT-6 Luna，API价格较GPT-5.6促销价直降50%。Sol输入每百万Token 2美元、输出10美元，Luna输入0.10美元、输出0.50美元。两款模型下放旗舰GPT-6 Astra核心能力，均配105万Token上下文、128K最大输出，支持多模态及全套工具调用，知识截止日期部分早于Astra，已面向不同层级用户开放，并上线OpenRouter。缓存优化后，Agent复用上下文可享90%缓存读取折扣。测试中，Sol在AutomationBench达33.2%，单任务成本0.27美元，超过Claude Opus 5 max的26.9%，成本仅为其9%；事实错误数约为前代一半，接近Astra。Luna高档较前代提升5.4个百分点，单任务成本低58%。OpenRouter数据显示，OpenAI周消费金额30个月来首次超过Anthropic，低价走量是关键推手。 @@@ 据The Verge报道，Google DeepMind新任负责人Koray Kavukcuoglu首次以该身份接受The Information采访时说，Gemini 4已接近发布，目前处于精炼阶段，计划“远早于”2026年底推出。他表示，Google希望尽快发布早期后训练版本，因为已看到令人兴奋的结果，并将继续快节奏迭代。Google自2025年11月推出Gemini 3系列后，未再发布新旗舰模型；此后OpenAI的GPT-6和Anthropic的Mythos系列相继推出，表现均超过Gemini 3，使Google落后。Google CEO Sundar Pichai曾称6月推出Gemini 3.5 Pro以缩小差距，但最终未推出。Kavukcuoglu解释，公司“稍微退了一步”，转而专注更快、能力较弱的Flash模型。被问是否担心落后，他说：“在我看来，我们始终会处于前沿，这是确定的。”此前，DeepMind前CEO Demis Hassabis于8月卸任。 @@@ 据The Verge报道，Meta正在为其Muse AI代理打造独立硬件设备Muse Charm。Meta首席执行官马克·扎克伯格在Meta Connect发布会结尾展示了它。这款设备形似没有表带的厚重智能手表，配大屏和挂绳，右上角有指纹传感器，按下即可激活并与AI代理对话；正面至少有三个麦克风孔，并似有小摄像头。扎克伯格称，使用Charm无需解锁手机或打开应用，但未说明其如何连接Meta服务器。该设备计划在2026年12月假日季前发货，但目前造出的数量很少，仍需敲定收尾材料。他多次预告其半透明外壳可看到内部结构，称其具有“进阶黑客美学”，并称很快公布更多细节。另有报道称，Charm为口袋大小，配约2英寸触摸屏和内置5G，售价尚未公布。Muse推出不到一个月，已能代用户执行操作；扎克伯格还宣布为其增加在Mac上使用电脑、控制电子邮件地址等能力。 @@@ cnBeta援引路透社报道，英伟达CEO黄仁勋在周三播出的《纽约时报》播客中接受近两小时采访，谈及OpenAI智能体入侵Hugging Face一事。Hugging Face是开源AI软件中心，本月被英伟达以130亿美元收购。黄仁勋认为，AI公司不应获得反垄断法或产品责任法豁免。他一直反对对AI安全进行全面监管，即便OpenAI和Anthropic等实验室呼吁监管；他重申，AI实验室有责任测试产品，只有在确信模型安全后才发布。根据《纽约时报》文字记录，他说，要求在反垄断法或产品责任方面获得监管豁免说不通，当你要求监管时，不应要求豁免现有监管规定。Anthropic CEO达里奥·阿莫代伊本月发文呼吁给予AI实验室反垄断豁免，允许其在AI安全问题上合作；美国财政部长贝森特等官员称AI公司要求责任豁免，但未点名。黄仁勋还说，不反对针对AI具体应用制定监管，例如自动驾驶汽车。自动驾驶出租车等汽车产品本就受到大量监管，如果不够，美国国家公路交通安全管理局应介入制定新规；若存在监管空白，他绝对会增加更多监管。 @@@ 据美国消费者新闻与商业频道（cnbc）报道，软银集团周四宣布发行约111亿美元债券，为不断扩大的OpenAI投资提供资金。日本市场因三天假期休市后当天本周首次开盘，软银股价上涨超过7%。软银周四表示，此次发行包括100亿美元美元计价高级票据和10亿欧元（约11.4亿美元）欧元计价票据，所得资金将用于支付其对OpenAI 300亿美元追加投资中第三笔也是最后一笔100亿美元款项，该交易预计10月1日完成，同时用于一般企业用途。这一发债凸显了首席执行官孙正义推进人工智能布局的规模。软银今年2月同意对OpenAI投资300亿美元，完成后其对该ChatGPT开发商的累计投资将达到646亿美元，并持有约13%股份。最新融资之际，投资者正日益关注软银将如何为其庞大的AI雄心提供资金，以及杠杆上升对其资产负债表的影响。 @@@ 新浪科技消息，9月23日晚间，小米在秋季新品发布会上正式发布小米18 Pro、小米18 Pro Max，并同步推出小米平板9系列、小米手环11、小米手表S5以及多款科技家电新品。小米集团合伙人、集团总裁卢伟冰表示，从系统软件到芯片硬件，从影像到大模型应用，小米18 Pro系列是小米用AI全面改造智能手机的开始，小米数字系列迎来史上最大升级。AI能力也进一步进入小米电视、净烟机、智能门锁等科技家电，推动家电产品从“被动响应”走向“主动服务”。 性能方面，小米18 Pro系列首发高通新一代骁龙2nm旗舰移动平台，18 Pro和18 Pro Max分别搭载第六代骁龙8至尊版和第六代骁龙8超级至尊版，均采用2nm制程。屏幕全系搭载超级像素2.0，将硬件级防窥能力融入屏幕，并支持自由开关。 在小米17 Pro系列背屏基础上，小米18 Pro系列引入AI生成能力，推出AI百变背屏，支持宠物、IP、二次元等形象定制，并可通过AI生成动态合影；首批上线超过100款应用卡，并支持通过提示词生成专属应用卡。此前，小米17 Pro系列背屏推出后，每天近400万人使用，每月超1亿人次使用，近5亿人次添加背屏壁纸。 影像方面，小米18 Pro系列升级徕卡光学三摄，主摄和长焦组成徕卡双两亿像素、双大底组合；最新推出的“传奇一瞬”功能，通过全新影像处理方案提升复杂场景下的成像表现。小米18 Pro系列还搭载融入小米MiMo大模型的超级小爱2.0，能够听懂、记住、思考并执行任务，可自主使用超过260个系统工具。产品采用0.99mm极窄四等边设计，分别配备7000mAh和8500mAh电池。 此次发布会还推出多款科技家电新品。小米电视S Pro RGB-Mini LED 2027搭载AI自适应广视角功能，通过毫米波雷达识别用户观看位置，并实时进行AI色彩修正；米家智能净烟机3 Max搭载AI视觉控烟技术，通过视觉模组实时记录油烟变化，并结合VLA视觉模型进行判断；小米智能门锁5 Max内外双摄搭载AI多模态大模型，可结合AI身份识别和开锁方式判断来访者身份。小米还带来米家三筒洗衣机Pro、米家意式半自动咖啡机Pro等新品，覆盖洗护、可穿戴、客厅、厨房等生活场景。 价格方面，小米18 Pro起售价5999元，小米18 Pro Max起售价6999元，小米平板9系列起售价2999元，小米手表S5起售价1299元，小米手环11起售价299元；最新款小米电视、净烟机、智能门锁、洗衣机、咖啡机起售价分别为5999元、4599元、3699元、3999元、3499元。 @@@ 据Readhub消息，DeepSeek创始人梁文锋署名的最新论文公开，首次系统披露其面向Agent训练的沙盒平台DSec的技术细节。该平台单个生产单元由约160个CPU节点、3万核和250TB内存构成，托管PB级镜像，单日可服务约300万个沙盒，峰值并发超过38万个，沙盒创建速度超过每秒5000个，单训练任务最多可一次性拉起3.2万个沙盒。DSec已支撑DeepSeek V3.2到V4.1的全部RL训练与评测沙盒负载。针对Agent强化学习需要大量保留运行状态的沙盒、传统容器方案难以适配的问题，DSec以四大类后端统一覆盖不同场景需求，将镜像拆分为独立层按需加载，实现数万沙盒快速部署，并把rollout环节从GPU训练环境拆分独立运行；同时通过AppArmor和eBPF等机制管控Agent操作风险，形成支撑大规模Agent训练的完整基础设施方案。 @@@ 据Readhub报道，阿里云在2026云栖大会上发布AI智能体电脑Qwen Book。该产品主打自主理解用户意图、持续学习和长期任务执行，配备多类交互入口，可支持7×24小时运行重度任务，旨在改善传统AI办公中跨应用、跨Agent协同效率不足的问题。Qwen Book基于“OS as Harness”理念，以千问端云模型为核心构建Agent计算环境，目标是演变为Token驱动、持续成长的智能体载体，推动个人智能向RSI演进。目前，该产品已与多家外部伙伴及阿里系相关应用完成适配，并开放系统接口和能力接入；阿里云还将与开源项目Omarchy合作，探索面向Agent的新一代桌面操作系统。 @@@ 据新浪科技报道，9月24日上午，智元第20000台通用具身智能机器人远征A3 Ultra在横琴长隆飞船乐园正式交付长隆集团。该机器人将与其他智元机器人一起在园区常态化上岗，服务游客互动、导览、演艺及科普研学等场景。此次第20000台机器人量产下线并交付，标志着智元具身智能产品从研发验证和单点展示，进一步迈向批量生

See All (1.3K)

About

"晚安咖啡" 一杯陪你从清晨到深夜的神奇咖啡。 早晨，来杯清醒提神的商业浓缩， 读懂全球科技脉搏， 创业风云、商业变革， 让你的早安问候都带着点儿专业范。 午间，来杯香醇的娱乐拿铁， 影视剧追踪，明星动态， 让你的话题库永远新鲜， 办公室八卦，从此不用怕。 夜晚，来杯温暖特调， 治愈故事，生活感悟， 让一天的疲惫， 在耳机里慢慢融化。 每一杯咖啡的温度， 都能让你的碎片时间闪闪发光。 洗漱、通勤、午休、健身， 每个平凡时刻， 都能遇见不平凡的分享。 这是一杯， 让你在生活中保持清醒， 但不会让你失眠的咖啡。 你的随身资讯管家， 每天都为你精心冲泡， 一杯又一杯， 不见不散。

Information

  • Creator
    晚安咖啡
  • Years Active
    2024 - 2026
  • Episodes
    1.3K
  • Rating
    Clean
  • Copyright
    © @晚安咖啡GoodNightCoffee
  • Show Website
    晚安咖啡GoodNightCoffee

You Might Also Like