"GTA 6" launches on PS5 and Xbox today, with a media score of 99
Google releases Gemini 4 Neon, focusing on complex tasks and lower single cost
The prices of many DJI cameras increased today, and the Pocket 4P Vlog set increased by 560 yuan
NVIDIA employee meeting: The board of directors will not let Huang Renxun return, and the interim CEO promises a salary increase
Anthropic officially sets IPO price: $83 per share, valuing approximately $1.5 trillion
Intel lays off more global IT staff as it continues to streamline its organization
After Doubao, Qianwen Office also connected to third-party models such as Kimi and DeepSeek.
Three months after Doubao was launched, DAU exceeded 8 million, and enterprise ARR exceeded 2 billion.
WorkBuddy’s daily active users exceed 10 million, Tencent’s cloud revenue increased by 35% in the third quarter
Nadella apologizes for Microsoft's global service outage: There is still much work to be done
Dong Mingzhu: Gree has been developing robots for more than ten years, and all production line robots are completed independently.
Zhijie obtained independent production qualification and Chery Group added a new member company
Microsoft executive: NVIDIA personnel turmoil may prompt customers to look for alternative chips
Apple announces App Store Awards finalists
Google Play will force apps to adapt to foldable screens starting next year
AMD executive: AI cannot replace supercomputers that require high-precision calculations
"Sonic Racing Cross World" launches McLaren linkage, adding new racers and cars for free
Medical AI evaluation platform InclusionMed is launched, and 22 models are compared in the first phase
Tencent open sourced three versions of Hy4 flash, supporting millions of Token contexts
Jimeng Seedance 2.5 supports native 4K, and the maximum single generation time is 30 seconds.
Keling Image 4.0 is online, with a maximum of 2,000 words for a single picture and a maximum of 10 pictures for a group of pictures.
UFIDA releases YonWork enterprise intelligence, recruitment and talent allocation can be connected to the execution process
DeepSeek model calls no longer charge separate search and crawl fees, and independent APIs are still billed on a pay-per-use basis.
Disney and AWS create "Star Wars" robots that can sense the movements of guests and companions
The father of "The Sims" announces MyMems, turning personal memories into explorable game worlds
RTX Spark mini desktop launches with up to 128GB of unified memory
Starbucks launches "Star Baked" in more than 100 stores, with the first batch of 11 food products available online
Group controlled more than 3,000 mobile phones to rob LABUBU, two scalpers were jailed
G-Dragon’s agency will launch a robot girl group and hold a concert in Macau in December
Mayday Concert Movie Officially Announced on Apple Vision Pro Platform
Zheng Yuanjie uses AI to animate fairy tales and releases them for free with authorized accounts
"GTA 6" launches on PS5 and Xbox today, with a media score of 99
The first media reviews of Grand Theft Auto 6 (GTA 6) were released on November 18. According to Variety's report that day, the game received a score of 99 on Metacritic, tying the 1998 N64 version of "The Legend of Zelda: Ocarina of Time" for the highest score in platform history. Rockstar’s official website confirmed that the game will be released on PS5 and Xbox Series X|S on November 19th.
After nearly 80 hours of experience, The Verge’s review affirmed the relationship between the dual protagonists of Jason and Lucia, as well as the narrative combined with the open-world gameplay. The story is set in Leonida, which is based on Florida. Players can play two people and gradually build their relationship; in addition to the main mission, they can also explore areas such as Vice City, beaches, and swamps through different activities.
The review used the launch version of the PS5 and said it ran well overall, but also noted that the combat controls were still clunky, the weapon wheel was easy to misselect, and there were a few physics and item display errors. After completing the main story and playing for over 70 hours, the reviewer still made less than 60% of the overall progress.
Related reading:"GTA VI" launch experience: The best game of the past decade, and may be the best game of the next decade
Google releases Gemini 4 Neon, focusing on complex tasks and lower single cost
Google released Gemini 4 Neon on November 18, which is aimed at software development, knowledge work and scientific research. It is now open to Gemini API, Gemini for business, Antigravity and Gemini applications. Among them, the Gemini application and Antigravity are available to Google AI Pro and Ultra subscribers.
Neon supports 1 million Token contexts and can comprehensively handle spreadsheets, PDFs, charts, audio and video. The API charges $3 per million input tokens, $15 for outputs, and $0.15 for cached inputs. Google said that in the default high thinking gear, Neon maintains a high capability in Artificial Analysis Intelligence Index while maintaining a single task cost that is 32% lower than Carbon.
The model can complete multi-step tasks independently, and can also schedule Hydrogen sub-agents in Antigravity. According to tests published by Google, Neon scored 61 in the Artificial Analysis Intelligence Index at the xhigh level, ranking third, and scored 46.2% in the FrontierCode 2.0 test at the default high level.
The prices of many DJI cameras increased today, and the Pocket 4P Vlog set increased by 560 yuan
Starting from 0:00 on November 18, DJI will increase the recommended retail price of some handheld imaging products, including Osmo Pocket 4P, Osmo Pocket 4 and Osmo Nano. The current price and price adjustment details of the official mall are displayed:
- The Pocket 4P standard package has increased from 3,799 yuan to 4,099 yuan, and the Vlog package has increased from 4,299 yuan to 4,859 yuan, with the latter increasing by 560 yuan.
- The Pocket 4 standard set has increased from 2,999 yuan to 3,199 yuan; the all-round set has increased from 3,799 yuan to 4,299 yuan, an increase of 500 yuan; the Meipai set and Changpai set are currently priced at 4,598 yuan and 3,638 yuan respectively.
- The Osmo Nano 128GB standard package has increased from 2,298 yuan to 2,498 yuan; the 64GB standard package remains at 1,998 yuan.
DJI previously stated in its price adjustment explanation on November 13 that the continued sharp increase in the cost of many key components such as global storage was the main reason for this adjustment.
NVIDIA employee meeting: The board of directors will not let Huang Renxun return, and the interim CEO promises a salary increase
According to "Wired" and CNBC reports, Nvidia held its first all-hands meeting after Huang Renxun's departure. Interim CEO Colette Kress and new chairman Harvey Jones were present, and the board again said it would not reinstate Huang as CEO.
Kress said at the meeting that the company will increase employee compensation and increase equity incentives. She said the most common of the roughly 700 questions she received before the meeting was about pay, followed by calls for more transparency from the board; she hopes employees won't have to worry about their pay.
Jones said the board accepted Huang's resignation due to his personal behavior and "lack of candor" and would consider internal and external candidates when searching for a successor, as well as accept employee recommendations. He listed criteria including technical and engineering experience, strategic thinking, cultural fit and a track record of success, saying he wouldn't lower the requirements for speed.
Anthropic officially sets IPO price: $83 per share, valuing approximately $1.5 trillion
Anthropic officially set an initial public offering (IPO) price of $83 per share on November 18, local time in the United States, which is higher than the previously announced guidance range of $70 to $78. The company plans to begin trading on Nasdaq under the stock symbol "ANT" on November 19, with a fully diluted valuation of approximately US$1.5 trillion.
Anthropic is increasing the number of shares it plans to offer to the public from 825 million shares to 980 million shares, and expects to raise about $80 billion through the IPO, Fortune reported. If the underwriters exercise their over-allotment option, the amount raised could be up to approximately $92 billion. At the same time, seven companies including Google, Nvidia, AMD, Broadcom, Micron, Microsoft and TSMC will also participate in private placements totaling US$26 billion at the IPO price.
The listing will also bring considerable paper wealth to Anthropic employees. Previously, SaaStr founder Jason Lemkin estimated that the net book value of employee options, deducting vesting costs and adding unvested restricted stock units, would be about $216 billion at the upper end of the original offering price of $78 per share. As the final issuance price increases, the valuation of this part of the equity will also increase accordingly, but the actual income depends on the number of employee shares held, vesting and exercise conditions.
Anthropic's Claude business is also expanding rapidly. According to Reuters, the company's annualized revenue has reached about $81 billion, driven by products such as Claude Code. However, AI model training and computing infrastructure still bring high costs, and the company recorded an operating loss of approximately US$567 million in the third quarter.
Intel lays off more global IT staff as it continues to streamline its organization
According to Business Insider, Intel is laying off employees again this week, and multiple global IT team employees expressed their impact on social platforms. An Intel spokesperson responded that the company will continue to evaluate staffing to match business priorities with resources.
This round of layoffs continues CEO Chen Liwu’s restructuring direction. He has previously proposed reducing layers of management, lowering expenses and shifting resources away from lower-priority projects. Intel has made several smaller rounds of layoffs this year.
Intel had 81,100 employees as of the end of September this year, down from 85,100 at the end of 2025, according to company documents cited in the report.
After Doubao, Qianwen Office also connected to third-party models such as Kimi and DeepSeek.
Following the work of Doubao, Qianwen Office also provides users with third-party model options. The client interface displayed in the product notes shows that in addition to using the Qianwen model, users can also specify models such as DeepSeek, GLM, Kimi, MiniMax and Step in the task.
The list includes DeepSeek-V4-Pro, GLM-5.3, GLM-5.3-Flash, Kimi-K3.1-Pro, MiniMax-M3.1-Flash and Step-5-Preview, with different resource consumption levels marked. Users can switch models according to task requirements and do not always have to use the platform's self-developed models.
According to the actual measurement of this article, Qianwen Office currently provides a preset third-party model list, but does not yet support users to configure custom models; WorkBuddy has provided custom model access. There are still differences between the two products in how model selection is opened.
Three months after Doubao was launched, DAU exceeded 8 million, and enterprise ARR exceeded 2 billion.
ByteDance’s AI office product Doubao released the latest data. On November 18, Doubao officially announced that only three months after Doubao was launched, the number of daily active users (DAU) has exceeded 8 million, and the annual recurring revenue (ARR) of enterprise customers has exceeded 2 billion yuan.
Doubao Job was officially released on August 25 this year. It is an AI product and brand created by Doubao for productivity scenarios.
WorkBuddy’s daily active users exceed 10 million, Tencent’s cloud revenue increased by 35% in the third quarter
Tencent released its third quarter results for 2026 on November 18, with revenue of 207.3 billion yuan, a year-on-year increase of 8%. WorkBuddy has more than 10 million daily active users, and the average daily user consumption exceeds 100 trillion Tokens; Tencent said that it has become the largest AI application in China measured by users' daily large language model Token consumption.
AI business growth and investment are accelerating at the same time:
- Cloud service revenue increased by 35% year-on-year, mainly driven by WorkBuddy, model-as-a-service and GPU rental demand; financial technology and enterprise service revenue was 63.7 billion yuan, a year-on-year increase of 9%.
- Under the IFRS standard, net profit attributable to the parent company was 63.6 billion yuan, a year-on-year increase of 0.8%; non-IFRS net profit attributable to the parent company was 66.5 billion yuan, a year-on-year decrease of 6%. Excluding the contribution of new AI products, non-IFRS operating profit was 83.3 billion yuan, a year-on-year increase of 8%.
- R&D investment was approximately 32.36 billion yuan, a year-on-year increase of 42%; capital expenditure payments reached 65.1 billion yuan, a year-on-year increase of 225%. Affected by prepayments for computing power purchases, free cash flow for the quarter was negative 69.1 billion yuan; excluding these prepayments, it was positive 16.9 billion yuan.
The financial report also disclosed that in the third quarter, Tencent’s local game revenue increased by 11% year-on-year, international game revenue decreased by 9% year-on-year, and marketing service revenue increased by 19% year-on-year. Ma Huateng said that the company has allocated more computing power to model-as-a-service and GPU leasing, driving cloud service revenue growth to further increase. WeChat AI assistant "Xiaowei" is also expanding the scope of gray testing and deepening its integration with mini programs and stores.
According to the Science and Technology Innovation Board Daily's report on the performance meeting that night, Tencent executives expect that as the AI infrastructure built in 2026 and 2027 is reused in subsequent years, the operating pressure caused by related investments will gradually ease in 2028. Executives also expect that the reduction in purchase prices for similar hardware and the improvement in model training and inference efficiency will jointly improve the input-output efficiency of the AI business.
Nadella apologizes for Microsoft's global service outage: There is still much work to be done
On November 17, local time, Microsoft CEO Satya Nadella opened the Ignite conference in response to last week's global service outage. He said that digital infrastructure is crucial to global operations, and Microsoft takes the trust of customers seriously; to provide resilience that customers and the world can trust, the company "still has a lot of work to do" and promised to learn lessons and continue to improve.
According to a report by The Paper reprinted by China Business News, the previous incident affected Microsoft 365, Outlook, Teams and other services, with a direct impact window lasting 23 hours and 50 minutes. In its preliminary report, Microsoft attributed the cause to a combination of factors such as software defects, increased load caused by Teams traffic switching, and the temporary unavailability of some network gateway capacity.
Microsoft said it has rolled back code, increased infrastructure capacity, and strengthened monitoring and traffic distribution protection measures, and plans to release a final incident report on November 20. Nadella also thanked the staff who ensure business continuity across organizations.
Dong Mingzhu: Gree has been developing robots for more than ten years, and all production line robots are completed independently.
According to Jiemian News citing Kankan News, Dong Mingzhu said at the APEC Business Leaders Summit held on November 17 that Gree has been developing robots for more than ten years. Currently, all the robots on Gree's electrical appliance automation production lines are completed by Gree independently.
Talking about the core components such as reducers, drives, and servo motors required for embodied intelligent robots, Dong Mingzhu took the initiative to look for cooperation opportunities at the scene and said to the guests: "I believe we will be a good partner for you. Whatever components you want, we will provide them."
Zhijie obtained independent production qualification and Chery Group added a new member company
According to the Cailian News Agency, an announcement from the Ministry of Industry and Information Technology showed that Chery Group was approved to add Anhui Zhijie New Energy Vehicle Co., Ltd. as a member company, and Zhijie obtained the independent production qualification for road motor vehicles.
The production address listed in the announcement is No. 8 Wanjiang Avenue, Economic and Technological Development Zone, Wuhu City, Anhui Province, and the registered address is in the same development zone. The access to road motor vehicle manufacturing enterprises is an important qualification for enterprises to carry out vehicle production and operation activities in accordance with the law.
In the same announcement, Dongfeng Group also received approval to add Lantu Automotive Technology Co., Ltd. as a member company. Lantu issued an announcement on the evening of November 17, disclosing that it had obtained independent production qualifications.
Microsoft executive: NVIDIA personnel turmoil may prompt customers to look for alternative chips
According to Bloomberg, Microsoft's commercial business CEO Judson Altoff said at the Ignite conference that the Nvidia personnel crisis caused by Huang Renxun's departure may push customers to consider other AI chips.
He said that before Huang Renxun resigned, customers had already expressed their desire to have more choices, and the recent news made them even more aware of this. He has been emphasizing over the past year that diversified models and chips will play an important role, and mentioned self-research projects of large cloud service providers and alternative products launched by AMD.
Altov also said that most customers are still waiting for the Nvidia crisis to subside. He believes that Nvidia has some corporate governance issues that need to be dealt with and will resolve these issues. Microsoft will continue to cooperate with Nvidia in the future.
Apple announces App Store Awards finalists
Apple announced 45 global nominations for the 2026 App Store Awards, covering iPhone, iPad, Mac, Apple Vision Pro, Apple Watch, Apple TV, Apple Arcade, and cultural impact categories. The winners will be announced in the coming weeks.
In the global list, the three nominees for iPhone App of the Year are the AI reading app ElevenReader, the retro camera app (Not Boring) Camera, and the flight travel assistant Raylie. Mac App of the Year nominations went to clipboard tool Paste, task management tool Things 3 and writing app Ulysses.
In terms of games, "Diver Dave", BALL x PIT and Gambonanza were shortlisted for the iPhone Game of the Year; "Arknights: Endland" and "Sixteen Voices of Yanyun" were shortlisted for the iPad Game of the Year. Works such as "The Little Bookshop", and Roger, and "Consume Me" were nominated for cultural impact games.
Google Play will force apps to adapt to foldable screens starting next year
Google announced new quality requirements for Google Play adaptive apps: Starting from April 2027, apps must meet the minimum standards for folding screen adaptation, otherwise they will not be released or updated in the store, and their exposure may also be reduced.
Minimum standards include filling the available display area, avoiding black borders or compatibility modes; retaining scroll position, text input and media playback status when rotating, folding, expanding and window size changes, and supporting multiple windows. Developers must also provide 3 foldable screen screenshots in the product listing, covering different internal and external screen ratios; these screenshots can be uploaded now.
The game's base adaptation period is August 2027, but store screenshot requirements also start in April. Google plans to extend the requirements to large-screen devices such as tablets and Googlebooks in subsequent phases.
AMD executive: AI cannot replace supercomputers that require high-precision calculations
Scott Tease, corporate vice president of AMD High Performance Computing and Sovereign AI, said in an interview with Tom’s Hardware that most high-performance computing research institutions have realized that AI will not replace traditional HPC; scientific computing that relies on double-precision FP64 will exist for a long time.
He believes that AI has become an important part of scientific discovery, and the way it is deployed will affect the way it works. For example, multiple AI agents can be allowed to explore different research paths, leaving complex simulations to traditional high-precision computing systems. Clusters of CPUs or GPUs that are isolated from each other and lack interconnections have difficulty keeping up with this way of working.
The direction proposed by Tease is to connect AI, HPC and quantum computing into a unified computing system to determine when each task is suitable for AI, GPU acceleration or quantum environment. AMD is working with IBM and Oak Ridge National Laboratory to connect classical computing, Umbra HPC systems and IBM's 129-qubit Quantum Nighthawk into the same research environment to explore the software, architecture and workflow of hybrid systems.
"Sonic Racing Cross World" launches McLaren linkage, adding new racers and cars for free
Sonic Racing: Crossover World launches a free update on November 18, adding Sonic (McLaren Racing Car) wearing a McLaren racing suit and the original McLaren Racing Car. Players can obtain it after updating the game to the latest version.
The linked "McLaren Racing Event" will start at 8:00 on November 20, Beijing time, and last until 7:59 on November 30. Players will be divided into three teams to participate in online racing, competing for total points ranking according to special rules of the event, including competing for the number of "touch sprints".
Event points can be used to obtain special edition in-game items to decorate the car with stickers, horns, etc. Sega also released an introductory trailer for the new racer and car.
Medical AI evaluation platform InclusionMed is launched, and 22 models are compared in the first phase
InclusionMed, the medical and health AI evaluation and analysis platform, is launched, and the evaluation map InclusionMed Atlas and the first evaluation index InclusionMed Index are launched at the same time. The project is jointly constructed by researchers from 29 institutions including Stanford University, Harvard University, Peking Union Medical College, and Ant Group.
The platform uses two dimensions to sort out the evaluation: on one side are three types of users: patients and the public, medical professionals, and medical managers; on the other side are three types of abilities: knowledge, reasoning, and task execution. They intersect to form a nine-square grid, showing the evaluation coverage of different medical needs.
The first phase of the evaluation focused on clinical diagnosis and treatment and medical management, including five benchmarks and 1,034 tasks, comparing 22 models from 11 manufacturers. The four public benchmarks and the team's self-built InclusionMed-DxFiles form an index; the latter is based on difficult cases in medical journals and examines the model's ability to make clinical judgments based on case clues. The project team also plans to expand tasks such as doctor-patient joint decision-making and diagnosis and treatment scheduling under resource constraints.
Tencent open sourced three versions of Hy4 flash, supporting millions of Token contexts
Tencent Hunyuan open source Hy4 flash, Hy4 flash-FP8 and Hy4 flash Base, developers can obtain the weight through Hugging Face, ModelScope and GitCode. According to the official warehouse, the model natively supports text, image input and text output, and the context window has a maximum of 1 million Tokens.
The total parameters listed in the Hy4 flash official warehouse are 220 billion, including 64 billion Engram parameters, and each Token activates approximately 12 billion parameters. The FP8 version is used to reduce the graphics memory requirements of deployment; Base is a pre-trained model without instruction fine-tuning, suitable for continued training and customization.
Tencent positions it as a low-cost model for daily tasks such as deep search, document processing, slideshow production, simple front-end and computer operations. In the internal WorkIndex review, it scored 42.1, higher than the Hy4 preview's 39.7, and the average cost per task was about one-third of the latter. WorkIndex covers eight types of tasks, including slides, documents, forms, audio and video processing, and visual understanding.
Jimeng Seedance 2.5 supports native 4K, and the maximum single generation time is 30 seconds.
Dream Seedance 2.5 launches a 4K version, upgrading capabilities such as single-shot video generation up to 30 seconds, full-modal reference, video editing and extension to 4K specifications. The 4K mode supports the 10bit BT.709 color format and provides six formats including landscape, portrait, square and ultra-wide.
According to Ai Faner's actual experience, creators can first use 480P in sample mode to determine the lens, composition and rhythm, and then select "Ultra HD 4K" in "Generate Feature Film" to reduce the cost of trial and error in high-standard production. Existing 4K videos can also be entered into the first frame reference, editing and extension process.
During the test, Aifaner used the real-shot 4K video of smart glasses, extended it to 6 seconds, modified the background lighting effects and soundtrack, and added guide lines and text boxes to the camera at the 2nd second. Jimeng has recently integrated scripts, assets, storyboards, video generation and post-editing into the short drama studio for use by teams that continue to produce content.
Related reading:Jimeng Seedance 2.5 major upgrade, AI video generation finally has its own 4K production line
可灵 Image 4.0 上线,单图最多生成 2000 字、组图最多 10 张
可灵 AI 于 11 月 17 日晚正式上线图像创作模型 Kling Image 4.0。新模型支持原生 8K 输出,并提升画面真实感、文字生成、指令理解与一致性。
文字与设计能力方面,Kling Image 4.0 可在单张图片内生成最多 2000 字;用户用一次描述,可以生成最多 10 张连贯图片。它也支持电商详情页和网页设计长图,一次呈现完整页面视觉。
用友发布 YonWork 企业智能体,招聘与人才配置可接入执行流程
据 36 氪报道,用友 HR SaaS 发布三项 AI 产品,分别覆盖业务规则关联、任务执行与智能体治理。
其中,YonOnto 关联企业经营、人力数据和管理规则,为 AI 分析提供业务依据;YonWork 围绕经营目标组织多个智能体协同,将招聘、人才配置等任务从分析建议推进到执行,由人在关键节点决策;YonAIG 则负责智能体全生命周期管理、安全治理、过程追溯和效果评估。
用友官方资料进一步说明,YonWork 与 YonOnto、BIP 的分工分别是编排任务、提供业务语义与规则,以及承接审批和业务操作。哪些动作自动执行、哪些环节必须由人审批,由企业根据授权和业务要求设定。
DeepSeek 模型调用不再单收搜索与抓取费,独立 API 仍按次计费
DeepSeek 更新官方价格文档,自北京时间 11 月 18 日 12:00 起,调用模型时触发的服务端搜索和网页抓取不再单独收费,仅按模型 Token 用量计费。
适用入口包括 DeepSeek 联网工具、Anthropic 兼容 API 的 web_search 工具,以及 Responses API 的 web_search 工具。开发者通过这些入口让模型检索网页时,费用计入模型的输入和输出 Token 消耗。
独立调用 Search API 和 Fetch API 仍按请求次数计费。前者根据查询返回搜索结果,后者抓取指定网页;此次调整针对的是模型调用过程中触发的服务端工具。
迪士尼与 AWS 制作「星战」机器人,可感知游客和同伴的动作
据《好莱坞报道》报道,迪士尼与 AWS 公布三台来自《星球大战》世界的安泽兰人机器人,名为 Vilo Vele、Kado Bre’a 和 Jibu Bre’a,将于本月底在拉斯维加斯的 AWS re:Invent 大会上首次公开亮相。
机器人由迪士尼幻想工程团队设计,各自装有摄像头和麦克风。AWS 技术帮助它们同时感知面前的游客、工作坊中的活动,以及另外两个角色正在做什么,再据此选择动作和外星语言台词。
AWS 表示,台词与动作均由幻想工程团队编写,AI 在这套表演范围内引导角色反应。团队还使用编程工具 Kiro 加快实现。迪士尼希望借此让角色根据不同游客作出个性化互动;目前公布的展示地点为 re:Invent。
《模拟人生》之父公布 MyMems,把个人记忆变成可探索的游戏世界
《模拟城市》《模拟人生》创作者威尔·赖特正在制作一款以个人记忆为素材的手机游戏 MyMems,计划于 2027 年推出。据 The Verge 对开发团队的采访,玩家可以对着手机讲述一段往事,再将生成的记忆片段放进游戏地图。
游戏由 Gallium Studios 开发,允许玩家加入代表亲友的 AI 角色。随着记忆不断累积,世界逐渐形成一张与个人经历有关的地图;玩家可将世界设为私密,也可与他人分享。团队将它描述为带有深入社会模拟的温馨世界建造游戏。
赖特说,《模拟人生》的玩家常会在游戏里重建自己的家庭,MyMems 则进一步把整个游戏围绕玩家本人展开。他希望它像翻阅家庭剪贴簿:随着页数增加,人物、事件和物品不断涌现,组成属于自己的生活场景。
RTX Spark 迷你台式机上市,最高配备 128GB 统一内存
英伟达宣布,华硕、戴尔与微星的 RTX Spark 紧凑型台式机已上市,宏碁、技嘉、惠普和微软的相关机型将陆续推出。产品面向本地 AI、模型微调与创作工作,也可作为持续运行个人智能体的独立 Windows 电脑。
这些设备配备 20 核 Grace CPU、拥有 6144 个 CUDA 核心的 Blackwell RTX GPU,最高搭载 128GB 统一内存,并提供万兆以太网。据 The Verge 报道,华硕 ProArt GR1X 的 64GB 内存、1TB 存储版本起价为 4499 美元。
软件侧支持原生 CUDA,以及通过 WSL 使用 Linux 开发工具。英伟达同时扩展 Adobe 软件的 Windows on Arm 原生支持,并在 Blender 5.3 中加入 DLSS 4.5 光线重建;后者需安装最新的 Game Ready 或 Studio 驱动。
星巴克百余家门店推出「星现烤」,首批上线 11 款食品
星巴克中国披露,「星现烤」系列已在全国 30 余个城市、100 余家门店上市,首批共 11 款食品,覆盖早餐和下午茶。
这套流程把烘焙环节搬进门店:面团在工厂制作后,以冷冻原料形态经冷链送到门店,再由店员完成烘烤、冷却、上架和当日销售。门店配置专业烤箱、冷冻储存设备和专用食品柜,店员接受相应操作培训。
首批产品包括香肠酥皮卷、草莓酱可颂、榛果可可酱可颂、手撕贝果和金扇蝴蝶酥等。两款注酱可颂在现烤后由店员现场注酱、撒粉。公司称,自去年 9 月起已在近 10 家门店测试这一模式,部分试点门店的食品销售几乎翻倍。
群控 3000 多部手机抢 LABUBU,两名黄牛获刑
据《检察日报》报道,上海市静安区检察院办理了一起利用外挂批量抢购 LABUBU、星星人等潮玩的案件。涉案人员最多可通过云端服务器,群控 3000 多部手机同时下单,2025 年 6 月至 8 月累计成功抢购 2.8 万余单。
吴某负责开发外挂,租借云端服务器、购买算法,并借助 AI 工具编写前端代码;夏某负责制作教程、招揽「抢手」和组建抢购群。抢手提前登录泡泡玛特微信小程序后,程序便可定时自动下单,单部手机约每 2 秒发送一次购买请求。
检察机关通过代码分析和测试环境中的真机运行,确认涉案程序绕过了平台风控校验,能够自动提交并创建有效订单。
11 月 5 日,法院以非法获取计算机信息系统数据罪,判处吴某有期徒刑三年、缓刑四年,并处罚金 5 万元;判处夏某有期徒刑三年、缓刑三年,并处罚金 3 万元。
权志龙经纪公司将推机器人女团,12 月先在澳门办演唱会
权志龙所属经纪公司 Galaxy Corporation 计划于 2027 年 1 月面向全球推出机器人女团,此后再分别推出虚拟 AI 女团和真人女团。Galaxy 创始人兼 CEO 崔龙湖在接受《每日经济新闻》采访时表示,公司希望把三种形态连接起来,形成一体化运营的团队。
在女团推出前,Galaxy 将于今年 12 月在中国澳门举办一场小型机器人 K-pop 演唱会,随后计划把演出带到迪拜、东京等地。崔龙湖表示,机器人硬件采购自中国,配套软件由韩国团队开发,再叠加 K-pop 内容;宇树科技和巨星传奇是其合作伙伴。
Galaxy 已在首尔运营机器人主题乐园。按崔龙湖披露的数据,乐园运营约六个月,总到访量超过 5 万人次,其中约 3.3 万人次付费入园。园内除 K-pop 演唱会和跆拳道表演,还提供机器人操控、绘画与食品制作等互动体验。
五月天演唱会大电影正式官宣登陆 Apple Vision Pro 平台
近日,五月天演唱会大电影正式官宣将登陆 Apple Vision Pro 平台,为持有这款空间计算头显设备的乐迷带来全新的沉浸式观演体验。
不同于传统平面银幕或流媒体平台的常规播放模式,登陆 Apple Vision Pro 的版本将充分结合头显的高分辨率 Micro-OLED 屏幕与空间音频技术。用户在佩戴设备观影时,不仅可以获得宛如置身演唱会现场内场前排的巨幕环绕视野,还能通过空间音效清晰捕捉到现场乐器演奏的方位感与万人合唱的声场氛围,把线下巡演的视听张力直接搬进客厅。
随着越来越多头部艺人演唱会与体育赛事加速接入空间视频生态,硬件设备的影音娱乐体验正在变得愈发立体。
郑渊洁用 AI 将童话做成动画,授权账号免费发布
据澎湃新闻报道,郑渊洁 11 月 17 日恢复停更两年的自媒体,称自己虽然不懂动画制作,已经用 AI 将自己的童话作品制作成动画短片。
他授权抖音账号「郑渊洁童话世界」(抖音号 83425404836)发布这些短片,供读者免费观看,并表示接下来将把《不踩井盖的小蛇》做成动画。
郑渊洁在原文中表达了对创作门槛变化的判断:不懂动画的作家能借助 AI 制作动画,不懂文学创作的人也可能借助 AI 写作。他认为,免费欣赏文化作品、更多人参与创作会成为趋势,并提醒作家尝试付费 AI 后再判断它对职业的影响。
#欢迎关注爱范儿官方微信公众号:爱范儿(微信号:ifanr),更多精彩内容第一时间为您奉上。




