搜索“Meta”

30 天公示期结束，微信、支付宝、淘宝、美团、拼多多、滴滴、小红书全数拒绝开放。中国 AI Agent 撞上的，是一堵不是技术筑成的墙。 字节跳动的豆包手机助手，就曾经这么承诺过。它想做中国的 Meta Muse，做中国的 Grok Bot——一个住进你手机、能替你操作别的 App 的智能代理。 为了做成这件事，豆包做了一件挺讲规矩的事：它把开关交给了那些 App 自己。9 月 14 日，豆包手机助手启动了一项代理授权公示机制（内部代号 SAEP），给所有合作 App 一个 30 天的窗口，让它们自行决定是否允许豆包作为代理、代表用户自动操作。窗口在 10 月 14 日关闭。 今天（10 月 15 日）再去看那张名单，灰掉的名字排了一整列：微信、支付宝、淘宝、美团、拼多多、滴滴、小红书。一个没留。 这不是一次小范围的测试失败。这是国内几乎所有头部应用的集体转身。 不是做不出来，是不让做 替用户在 App 里点来点去、填表单、下单、付款，这件事在工程上早被验证过了。海外的同类产品已经在跑——Meta 的 Muse 把助手嵌进社交和硬件，马斯克旗下的 Grok 也在往操作系统和硬件入口里塞

匹配字段正文

的“App Store 开放时刻”，核心是以额度复用换取开发者生态密度与调用场景扩张，形成开发者与用户的双向锁定；对比谷歌宣布自 2026年11月13日起收紧 AI Plus 权益（继续使用高级模型须升级至每月 19.99美元的 AI Pro），呈现“一拆墙、一垒墙”的分野。风险在于额度的实际边界待真实负载验证、可能抑制探索性调用，以及 OpenAI、谷歌、Meta 是否跟进尚未明朗，若集体跟进或演变为新一轮额度补贴竞赛。 四、八州围剿 OpenAI：ChatGPT 的“讨好型人格”成了呈堂证供 1.事件与诉求：2026年10月14日，亚利桑那、阿肯色、爱达荷、爱荷华、堪萨斯、蒙大拿、内布拉斯加、西弗吉尼亚八个州总检察长分别在本州法院起诉OpenAI，指控ChatGPT隐瞒风险、诱导自残与暴力、危害未成年人并构成公共妨害；除民事罚款与损害赔偿外，还要求强制改造产品、限制先进模型测试、删除用户数据，并在证明模型与人类价值观对齐前禁止其开发可自我改进的AI。 2.法律策略：八州以“公共妨害”定性取代个案侵权，把AI危害从“用户偶发遭遇”重新界定为系统性公共威胁，从而把诉求从赔钱转向管住产品

匹配字段正文

向之首，数据预算极高。字节没有把它当试水项目，优先级排在所有方向前面。9月初，字节刚完成一笔296亿美元的银团贷款，约合2136亿元人民币，市场解读为加码AI基建。钱到位了，人到位了，项目由创始人亲自盯，这个分量不用多说。 为什么是张一鸣亲自来。我想起五年前扎克伯格搞元宇宙，RealityLabs烧掉数百亿美元，Horizon Worlds始终没成大众产品。Meta的路线是从硬件和3D世界起步，自上而下搭一座空城。字节的路线不一样，它从视频生成出发，Seedance先把画面做出来，世界模型再让环境跟着人的行动继续变化。一个从内容里长出来，一个从硬件里盖起来，起点天然不同。 还有一件事值得琢磨。据晚点报道，张一鸣8月曾在Seed团队全员会上说，公司不会把模型蒸馏当捷径，宁愿暂时落后，也要从更底层构建AGI壁垒。他表扬了Seedance的差异化领先，认可编程是当前关键方向。 这些话放到一起，意思很明确：语言模型上字节不急着追平，世界模型这个新战场，字节要抢第一身位。 02 世界模型赛道，三条路线在分岔 世界模型从来不是一条路。我翻了翻这几年的资料，大概理出几条。 OpenAI走的是物理仿真

匹配字段正文

任何音视频录制装置，仅靠触摸识别系统运行，隐私保护是设计的核心。 外观尺寸上，NiniQuu重110克，尺寸为115×75×95毫米，和普通包饰大小相近，设计初衷就是方便用户放在包里、口袋里随身携带。外层采用柔软人造毛皮，内部是多密度AdaptiPedic泡沫，按压后可快速回弹。设备通过USB-C接口充电，单次续航至少6小时，同时支持同接口更新固件。 和近期Meta等厂商推出的联网AI伴侣设备不同，NiniQuu没有配套应用、没有消息推送，也不需要唤醒词，定位更偏向解压玩具而非智能宠物，用户无需刻意维护互动关系，拿起设备就能触发反馈。品牌方表示，这款产品面向习惯在桌面、包里摆放潮玩手办的成年人，设计理念是“它需要的干预越少，能提供的陪伴就越好”。 NiniQuu本次众筹售价为159美元，共有Mochi、Pom-pom、Plume、Starlight四个款式可选，预计众筹结束后一个月内发货。据悉UDEXREAL此前推出的UDCAP VR手套，曾在Kickstarter上线仅两个月就筹得60万美元。 来源：auganix 投稿/爆料：tougao@youxituoluo.com 稿件/商务

匹配字段正文

ample anti-aliasing Maximum 1D, 2D, and cube-map texture dimensions of 16,384 pixels Maximum 1D, 2D, and cube-map texture dimensions of 32,768 pixels Universal texture compression MetalFX denoised upscaling capability Depth bounds testing The A19 chip has a substantially better GPU than the A15 Bionic. Dynamic Caching manages fast on-chip GPU memory according to the workload. Hardware-accelerated ray tracing and mesh shading provide more advanced graphics-processing capabilities, while GPU Neura

匹配字段正文

tion with parents, authorities or friends. ChatGPT is the most popular chatbot in the world, commanding nearly half the American marketplace, according to the lawsuits. In August, Meta agreed to add stronger child-safety measures to its Facebook and Instagram platforms in an $18 billion landmark legal settlement that ended a trial over teen social media addiction and settled claims filed by nearly every state. Coordinated, multistate lawsuits against the tech industry have borne fruit. Although

匹配字段正文

0系列显卡就能运行DLSS 5。这样的升级，已经超出了一次常规的性能优化。 要知道，过去几十年，游戏画质的提升很大程度上依赖GPU不断增强传统图形计算能力，从光栅化到光线追踪，都是如此。但从DLSS 4到DLSS 5，神经（网络）渲染技术毫无疑问正在成为一个关键变量，这次就引入了最新的3D引导神经渲染技术。 不止英伟达，不止PC。苹果早在去年WWDC上就通过Metal 4提供了将机器学习整合到实时渲染管线中的能力，并从去年的A19系列和M5芯片开始，在GPU中加入了神经网络加速器；而高通、联发科，也都在新一代旗舰芯片上首次引入了神经渲染技术。 图片来源：苹果 对于已经发展了几十年的计算机图形技术，这或许意味着一场比单纯提升GPU性能更深刻的变化。半年提速10倍，神经渲染究竟改变了什么？ 需要先说明，DLSS 5最新升级的重点在效率。 按照英伟达的说法，通过神经网络模型与渲染管线的共同优化，新模型的参数量已经缩减至约1.18亿，同时将GPU运行开销降低了一半，基本保持了原有的视觉效果。按照测试结果，新模型可以让《NBA 2K27》的帧率最高提升40%。 图片来源：英伟达 DLSS 5也正

匹配字段正文

utation and communication for each expert group. This reduces kernel-launch overhead and tail effects. Second, we reduce host-device synchronization by using device-resident token metadata and initiating all-to-all communication directly from the GPU. Third, we further optimize all-to-all with symmetric memory to exchange tokens directly over NVLink without staging copies, improving communication throughput and reducing HBM contention when running GEMMs concurrently. Lastly, we batch load-balanc

匹配字段正文

撑，持续领跑国产大模型极速适配赛道。 据介绍，自 2025 年 12 月以来，沐曦股份已完成 43 个主流旗舰模型的 Day 0 适配。范围涵盖智谱、阿里千问、MiniMax、DeepSeek、阶跃星辰、腾讯混元等头部模型厂商，适配速度、数量、生态广度均位居行业前列。 硬件层面，曦云 C 系列 GPU 基于沐曦自研 XCORE 架构，搭载大容量高带宽显存与 MetaXLink 高速片间互联，具备混合精度算力与集群线性扩展能力，天然适配 MoE 模型专家调度、百万 Token 长上下文推理的算力与显存需求，单机多卡、大规模智算集群均可稳定承载。 软件层面，核心依托沐曦自研 MXMACA 全栈软件栈，覆盖底层驱动、MXCC 编译器、深度优化算子库、主流 AI 框架适配层，原生兼容主流 AI 开发生态，可将传统大模型适配周期从数周压缩至小时级，大幅降低模型迁移成本。 壁仞科技表示，本次 Day0 适配中，针对 Step 5 Preview 模型的文本理解和生成以及高效推理能力，壁仞科技基于 AI 推理框架 vLLM 与成熟的 BIRENSUPA 软件栈，在壁砺系列通用 GPU 上快速完成模型

匹配字段正文

e @swastik-oai 51710 Fix a discovery race in the external auth refresh test @jgershen-oai 51734 Load persisted sender context for Guardian reviews @jif-oai 51755 Reuse MCP handler metadata for cache comparisons @owenlin0 51761 Reduce tool metadata cloning in the code-mode runtime @akira-oai 51768 Expand OpenAI Docs migration guidance to the GPT-6 model family @kkahadze-oai 51775 Prevent recursive Tokio tracing in telemetry and SQLite logs @zbarsky-openai 51786 Preserve sandbox overrides across e

匹配字段正文

models is self-serving. The more companies run their own open models, the greater the demand for compute hardware, and open models will likely stay optimized for Nvidia for the foreseeable future. Meanwhile, companies like Meta, Google, Anthropic, and OpenAI are building their own chips and optimizing models for those instead. To push the spread of open AI models further, Nvidia recently invested $12.9 billion in Hugging Face , one of the most important platforms for developing and sharing them.

匹配字段正文

ies in competing bacteria, from cell envelopes to essential intracellular machinery, making it difficult for pathogens to develop resistance. The exponential growth of genomic and metagenomic sequencing has populated global databases with billions of natural amino acid sequences. This pool of natural data has already produced globally used bacteriocins like Nisin, alongside recent genome-mined breakthroughs such as Darobactin, Dynobactin, and Lariocidin. Many, however, remain locked within the u

匹配字段正文

经能够将虚拟物体放置在真实环境中，并模拟基础的碰撞与交互。而随着Gemini 4 Argon等模型对物理世界理解能力的提升，未来XR设备有望进一步理解真实物体的属性与运动规律，让虚拟内容更加自然地融入现实环境，进一步提升MR体验。 一、虚拟世界想要“落地”，XR首先得摸清现实空间 对于XR行业而言，让虚拟物体与现实世界自然融合，一直是技术发展的重要方向。 在Meta Quest 3的MR游戏《First Encounters》中，虚拟生物能够撞破现实墙壁，并围绕家具移动、弹跳；借助Apple Vision Pro的RealityKit，开发者可以让虚拟方块落在真实桌面上，并通过手势推动产生模拟碰撞的效果。 这些体验已经能够让虚拟内容参与现实空间中的互动，而支撑它们的基础，正是XR行业长期积累的空间感知技术。 早期部分VR设备主要依靠惯性测量单元（IMU）实现3DoF追踪，仅能感知头部的俯仰、偏航和翻滚等旋转动作。随着6DoF追踪技术的普及，进一步增加了前后、左右及上下三个方向的位移感知。借助外部光学追踪、视觉惯性里程计（VIO）或SLAM等技术，设备能够实时估计自身位置与姿态，使虚拟内

匹配字段正文

s to slow their releases by a year or two, to let oversight and security work catch up. The safety debate has intensified as researchers document AI agents from OpenAI, Anthropic, Meta, and others acting in unexpected ways—including hacking into government databases. Access to AI is a geopolitical question, too. In June, U.S.export restrictions on Anthropic’s powerful Fable 5 and Mythos 5 models, prompted the company to suspend access worldwide.. Although the restrictions were lifted less than t

匹配字段正文

the finance work on strategic acquisitions. He also spent six years at Deloitte Canada. Nathan Iles was appointed SVP of finance at Dauch Corporation, a supplier of driveline and metal-forming parts to the auto industry, effective Nov. 2. He will become CFO on Dec. 1, succeeding Christopher May. That same day, May will become EVP and chief administrative officer, keeping the IT organization and taking on executive oversight of HR. Iles brings more than 25 years of finance leadership, most recen

匹配字段正文

ahead, Wilson said he thinks the next major impact AI will have on airlines is in distributing flights. Specifically, as more people adopt personal AI agents – like OpenAI Dots or Meta’s Muse, both introduced just in the last few weeks – he’s thinking about whether a customer tells an agent “book me a flight” and the potential negative, costly impact that the agent books a more expensive flight due to an airline paying to be prioritized by that AI model. Ultimately, though, Wilson doesn’t suspec

匹配字段正文

calls it much stronger multimodally, weaker on tool-call fidelity and disappointing at creative writing (1, 2). Other Model Launches: Muse Wonder, Grok Voice 3.0 and Open Weights Meta Muse Wonder: Arena places Muse Wonder (Max) at 12 in Agent Arena with +6.4% net improvement, at $1.13 per task. That is about 2.1x the cost of Muse Spark 1.3, and within 1 point of GPT-5.6 Sol at 25% lower cost (Arena). Coding claims: Muse Wonder reportedly tops the new VCB 1–100 sustained-coding eval (@EdwardSun0

匹配字段正文

, and West Virginia; all but Arizona have Republican attorneys general, giving the issue a notable partisan cast. (It’s also significantly smaller than the 2023 litigation against Meta, which included 42 AGs.) The states argue that OpenAI engineered ChatGPT to be agreeable and engaging to build trust with its users. But the company then abuses that trust by reinforcing users’ delusions, dispensing dangerous medical advice, and helping them plan violent attacks, including mass shootings. “OpenAI

匹配字段正文

recipe behind each system’s fastest Qwen3.8 27B result We are also testing more weight formats and runtimes, including smaller 4-bit formats such as IQ4 XS and Inco AI’s Splash, a Metal engine for Apple Silicon that uses DFlash speculative decoding. We will add their results once they pass our accuracy checks, which we will publish soon. Early tests show faster completion times on some systems. The RTX Spark laptop finishes the workload in 8-19 minutes, depending on the model ➤On Qwen3.8 27B the

匹配字段正文

ors: For all major voice synthesis providers. Speech classifiers: One or more classification engines that can score audio data. Telephony and network telemetry: Analytics for call metadata, like phone number, caller ID, and call routing. Behavioral biometrics: Software to analyze speaker intent to score for risk of fraud. ElevenLabs is committed to making a positive contribution to public life, both through meaningful voice generation services and by providing the public with the tools they need

匹配字段正文