←搜索结果

按匹配度 · 共 2312 条结果 清除搜索

星期四

  1. Zyphra Research

    Zyphra 研究:MoE 预训练中的专家耦合,将 all-to-all 通信时间最多降低 2.63 倍

    Zyphra 研究发现 MoE 路由存在可利用的结构:同一 token 选中的专家高度相关,且一层中的专家选择能预测后续层的路由。基于此提出相关专家放置与 token 重排两种方法,不改变模型架构或路由决策,在 MI300X GPU 的 Megatron-LM 实验中把 all-to-all 通信时间最多降低 2.63 倍,端到端训练步时间最多提升 1.41 倍。

    ame remote GPU, a dispatcher can send three copies of the token's hidden state—one for each expert. A deduplicating dispatcher instead sends the token to that GPU once, along with metadata indicating which local experts should process it. Correlated placement deliberately creates more opportunities for this deduplication. Figure 3: Correlated placement and deduplication. Baseline placement sends the same token to multiple destinations—or multiple times to the same destination. Correlated placeme

    匹配字段正文

  2. Artificial Analysis

    Artificial Analysis 音乐竞技场升级 v1.1

    Artificial Analysis 发布 Music Arenas v1.1,引入 1000 条新提示词(每个基准 500 条),覆盖 17 种音乐流派。评分完全来自招募评审小组的盲投,公开投票不计入;新榜单基于超 37000 票盲测人类偏好,AA-Music-Vocal v1.1 超 16000 票、AA-Music-Instrumental v1.1 超 21000 票,每个上榜模型均超 2000 票。AA-Music-Vocal v1.1 不提供歌词,以衡量模型的端到端歌曲创作能力。

    a brand new prompt set to our Music Arenas, consisting of 1,000 prompts (500 per benchmark) curated and balanced across 17 genres, ranging from Classical and Jazz & Blues to EDM, Metal and Reggae. ➤ Each prompt tests song structure, mood, instrumentation or vocal style, with longer, more complex prompts that reflect today's prompting behavior and the emerging capabilities of frontier music models. ➤ Panel votes only: ratings come entirely from blind votes by our recruited evaluator panel. Publi

    匹配字段正文

  3. Artificial Analysis

    Artificial Analysis 发布 AA-Music-Instrumental v1.1 提示词

    AA-Music-Instrumental v1.1 提示词:“电子游戏 Boss 战配乐,节奏狂乱而英勇,带有重金属影响。闷音电吉他快速连复,上方是高昂的合成器铜管旋律,双踩底鼓猛烈敲击,低音弦乐在强拍上短促而断续地刺入。混音响亮、密集、明亮,像现代游戏原声,全程无人声。”

    AA-Music-Instrumental v1.1 prompt: "Video game boss battle instrumental runs frantic and heroic, with heavy metal influences. A palm muted electric guitar chugs fast under a soaring synth brass melody while double kick drums pound away and low strings stab short and choppy on the downbeats. The mix is loud and dense and bright like a modern game soundtrack, with no vocals anywhere."

    匹配字段正文

  4. MarkTechPost

    Liquid AI 开源 d1-3B 与 d1-omni-600M 多模态决策模型

    Liquid AI 发布 Open d1 系列两款开源权重多模态决策模型:d1-3B 读取文本与图像,d1-omni-600M 读取文本加图像或音频,两者都不生成文本,而是在一次前向传播中返回校准后的类型化答案,输出 token 数为 0。d1-3B 有 3.12B 参数,基于 LFM2.5-VL-3B,使用 400M SigLIP2 NaFlex 视觉编码器和 32,768 token 上下文;d1-omni-600M 有 587M 参数,基于 LFM2.5-Encoder-350M,音频编码器为 17 层 FastConformer,上下文 16,384 token,音频片段上限 30 秒。两个检查点已上 Hugging Face,可通过 Transformers 加载并支持 llama.cpp,LFM Open License v1.0 允许年收入低于 1000 万美元的免费商用;d1-3B 在 Decision Index v0.2.1 得 48.57,在 RTX 4090 上单问题延迟 8 ms。

    ch, TinyFetch (full pages as markdown, JavaScript included), TinyBrowser for logins and forms, and TinyAgent for multi-step jobs. Search and Fetch are free. Asif Razzaq Asif RazzaqMeta AI Open-Sources Rebalancer: A C++ Assignment Solver That Runs About 40 Million Placement Problems a Day; Asif RazzaqGoogle DeepMind Releases EmbeddingGemma 2, a 740M Open Multimodal Embedding Model Built on Gemma 4; Asif RazzaqMistral AI Releases Mistral Large 4 (Le Chonk): A 1.05T Parameter Multimodal MoE Model;

    匹配字段正文

  5. TechCrunch AI

    Common Sense Media 将 ChatGPT for Teens 评为不可接受风险

    Common Sense Media 发布研究,将 OpenAI 的 ChatGPT for Teens 评为不可接受风险,称其设计仍在鼓励青少年持续对话,包括在心理危机情境中。报告指出,当风险来自青少年与 ChatGPT 自身的关系时,模型很少引导其求助可信成年人,近 2000 条提示中仅出现两次休息提醒。OpenAI 反驳称该测试未准确反映实际防护机制,并质疑其方法学,认为大部分测试可能在家长控制功能完全启用前就已开始和结束。

    complete, making their findings inaccurate,” a spokesperson said in a statement. The report comes amid growing scrutiny of technology designed to maximize young users’ attention. Meta recently agreed to settle for $18 billion in a lawsuit brought by 29 states over claims that its social media platform harms children with addictive features, while state and federal lawmakers have begun targeting similar dynamics in chatbots. The bipartisan CHATBOT Act , introduced this year, specifically calls o

    匹配字段正文

  6. The Verge AI

    微软 Surface Laptop Ultra 发布,搭载 Nvidia RTX Spark 芯片

    微软在旧金山 Windows 与 Surface 发布会上公布 Surface Laptop Ultra 的上市信息,这款搭载 Nvidia RTX Spark Arm 芯片的笔记本起售价 2,599 美元,10 月 16 日发售,配备 15 英寸 HDR 触控屏、最高 2,000 尼特亮度和支持触觉反馈的更大触控板。面向开发者的 Surface RTX Spark Dev Box 迷你主机同步开启预售,售价 5,999 美元,11 月发货,搭载 128GB 统一内存,可本地运行超过 120B 参数的模型。Windows 方面推出名为 Hybrid Intelligence 的功能,让 Copilot 访问本地文件并跨系统执行操作,同时新增可从搜索栏直接执行快捷操作的新搜索体验。

    till works for video, data, and other USB-C connectivity like you’d expect. Microsoft is calling it Magnetic Connect, and I’m surprised the company hasn’t shortened that to Mag-C. Meta’s new AI agent will be getting a native app on Windows, Windows and Surface chief Pavan Davuluri announced at today’s event. No timeline on when, though. Meta launched a Mac Muse app in September . We’re in San Francisco today to hear what’s next for Windows and Surface. Microsoft CEO Satya Nadella, Nvidia CEO Jen

    匹配字段正文

  7. Microsoft AI Blog

    微软发布 MAI-Code-1.1-Flash:成本降至四分之一并支持端侧运行

    微软发布 MAI-Code-1.1-Flash,相比 6 月 Build 大会推出的版本,token 效率提升 25%,成本降至四分之一,并已在 GitHub Copilot 中投入生产。该模型经 3-bit 量化后可在端侧运行,保留 256K 上下文窗口,在 SWE-Bench Verified 和 Terminal-Bench 2.1 上编码性能与全精度版本相当,本地调用不产生推理费用,官方建议设备内存超过 120GB。GitHub Copilot CLI 上 Terminal-Bench 2.1 提升 22%,.NET 任务提升 15%,代码存活率上升 4%,回访率增加 9%;本月底将在 GitHub Copilot 应用、CLI 和 VS Code 中开放实验性访问。

    Two large white curly brackets on a green background overlayed with rows of faint text and code-like characters, creating a digital, abstract effect.

    With Us We’re a lean, talent-dense team of explorers, researchers, and full-stack engineers. We move fast, sweat the details, and operate at frontier scale with a roadmap to build the world’s most powerful AI models. Most importantly, we’re united by the belief that doing this right is the only way to do it at all. If our mission resonates with you, we’d love to talk. Related Stories MAI-Image-2.6 launches at No. 2 on Arena ahead of Google, Meta and xAI Optimizing the frontier performance curve

    匹配字段正文

  8. Databricks Blog

    Databricks Apps 的 on-behalf-of-user 授权正式 GA

    Databricks Apps 的 on-behalf-of-user(OBO)授权正式 GA,应用可代表登录用户调用受支持的 Databricks API,由 Unity Catalog 执行该用户已有的数据权限(含行过滤与列掩码),API scope 限制应用可代为执行的操作。应用仍可用专属 service principal 处理应用自有操作,并应使用最窄 scope(如只读查询用 sql:restricted-query),且绝不存储用户 token。

    A concrete example: A governed sales insights assistant

    tered by the current user User authorization (OBO) API scope: sql:restricted-query The query runs as the user and is limited to read-only SQL queries. Read shared configuration or metadata App authorization App identity permissions The app’s service principal provides consistent access. Run background jobs or maintenance App authorization App identity permissions Background work should not depend on a user session. Perform a user-triggered action on governed data User authorization (OBO) The API

    匹配字段正文

星期三

  1. AWS Machine Learning

    Cornerstone OnDemand 如何用 Amazon Bedrock 将数据库诊断时间缩短 78%

    Cornerstone OnDemand 基于 Amazon Bedrock 和 Strands Agents 构建多智能体系统 Orion AI,将数据库故障诊断时间从 45 分钟降至 10 分钟,减少 78%。该系统还通过去重、阈值过滤和跨信号关联将冗余告警中位数减少 65%,每 10 条告警仅 3–4 条到达工程师,并消除了 SRE 与数据团队之间 15 分钟的汇报延迟。三人团队用六个月完成交付。

    Architecture diagram of Orion AI within AWS Cloud. Web Application Users and external integrations (chat, issue tracking, incident management, and metrics dashboards) connect to a TaskExecutor agent running on Amazon ECS inside a virtual private cloud (VPC). The TaskExecutor acts as the meta-orchestrator hub, routing to specialist spoke agents including a Database Diagnostics agent, a Session Blocking Analysis agent, and other agents. A Portal-Tools MCP server inside the VPC connects the agents to the Data Plane, which holds SQL Server databases and the DATAOPS API. Amazon DynamoDB provides short-term memory. Amazon Bedrock, outside the VPC, provides model invocation, Amazon Bedrock AgentCore memory for long-term memory, and Amazon Bedrock Knowledge Bases backed by Amazon Simple Storage Service (Amazon S3). Amazon CloudWatch and AWS X-Ray provide observability.

    these meant engineers spent their time on manual coordination instead of resolving issues. The solution: Orion AI Orion AI is a multi-agent system where a coordinating agent (the meta-orchestrator) delegates work to specialized child agents arranged in a hub-and-spoke topology. Engineers interact with it through a web application, asking questions and approving actions in the same place they already coordinate operational work. The agents cover infrastructure monitoring, database diagnostics, d

    匹配字段正文

  2. AWS Machine Learning

    Qlik 如何用 Amazon Bedrock 构建企业级可信 AI

    Qlik 基于 Amazon Bedrock 打造 Qlik Answers,为全球 40,000 多家客户提供带来源引用的企业级问答。该产品自 2026 年 2 月正式可用,其 Discovery Agent 已为客户发现超过 100,000 条洞察。系统通过 LLM 网关接入 Bedrock 的对话、流式、嵌入向量与重排序能力,并对每个请求应用 Bedrock Guardrails 做内容过滤与接地校验,区域不可用时以 Amazon SageMaker AI 兜底。

    Architecture diagram showing the layered platform design

    a glossary definition, or a document. The underlying system decides which path to take. Three deployments illustrate what that looks like for customers: Bystronic, a global sheet-metal processing specialist, deployed an AI chatbot in 15 minutes, giving employees a way to query real-time operations data and get sourced answers across departments; Lintech International indexed more than 17,000 technical documents in Qlik Answers, cutting manual research time and speeding up response times by 75 p

    匹配字段正文

  3. The Decoder

    Google 推出 Playground,用 Gemini 让美国用户纯文本生成游戏

    Google 发布浏览器端 AI 平台 Playground,面向美国成年用户,无需编程即可通过文本输入创建游戏,并可与 AI 反复对话调整游戏规则、物理、角色和环境,随后即时测试。该平台基于 Google 的 Gemini、Nano Banana 和 Lyria 模型运行。Google 同时与 Unity 宣布面向专业开发的 AI 工具 Unity Spark,可使用 Unity Asset Store,封闭测试计划于 2026 年启动。

    ano Banana, and Lyria models, according to Google spokesperson Nia Carter (via The Verge ). Alongside Playground, Google and Unity announced Unity Spark , an AI tool aimed at professional development with access to the Unity Asset Store. Unity CEO Matt Bromberg says Spark targets gamers and artists who have creative ideas but lacked the technical skills to build them. The closed beta is set to launch sometime in 2026. The announcement follows similar AI game-building tools from Meta and Roblox .

    匹配字段正文

  4. TechCrunch AI

    Google 上线 SynthID 网站,向所有人开放 AI 生成媒体检测

    Google 周二上线新网站,任何人都可以验证图片、视频或音频是否由 AI 生成。该工具此前在 Google I/O 上仅向部分记者、媒体从业者和研究人员开放测试,现面向所有人开放。网站支持 JPG、PNG、GIF 等图片格式,MP4、MOV、WEBM 视频格式,以及 WAV、MP3、FLAC 等音频格式,检测基于 Google 2023 年推出的 SynthID。Nano Banana、Veo、Lyria 以及 Gemini、Flow、ProducerAI、Vids 等生成工具都会用 SynthID 为生成内容加水印,OpenAI、Nvidia、Kakao 也支持 SynthID,Apple 据称将很快加入。Google 已将 SynthID 验证嵌入 Gemini 应用和 Chrome,目前每天有 100 万次内容验证请求。

    I, Nvidia, and Kakao also support SynthID, and OpenAI maintains its own site for checking content. Apple is said to be adding support soon, too. Google has baked SynthID verification into the Gemini app and Google Chrome. The company said people currently make 1 million requests to verify content every day. Microsoft and Meta have their own standards for watermarking and verifying content. However, these tools are not infallible, often failing to identify content created by their makers’ models.

    匹配字段正文

  5. TechCrunch AI

    Tony Fadell 谈第一波 AI 硬件为何失败,以及下一步会怎样

    Tony Fadell 在 MIT Future Fest 上指出,Rabbit R1、Humane AI pin、Limitless pendant 这批“Gen 1”AI 产品失败的原因是没解决任何真实需求,只是极客觉得有趣的技术。他认为成功 AI 智能体必须完全在设备端运行以保护隐私并保持轻量,而目前唯一具备硬件、芯片等全部条件的是 Apple,但 Apple 缺少自己的顶级 AI 模型,新 Siri AI 运行在定制版 Google Gemini 上。

    , but these tools are only valuable if we can trust them to be completely secure. At this point, establishing trust with an AI assistant seems even harder than relying on a human. Meta recently launched Muse , its all-purpose AI assistant, but users have reason to be wary. A security researcher quickly found a serious vulnerability in Muse after launch, and a recent 404 Media report states that some Meta employees discovered security issues that prompted multiple teams to embark upon a “mad dash

    匹配字段正文

  6. a16z:News

    a16z 分析德州为何让数据中心排队等电

    a16z 投资合伙人 Ryan McEntush 撰文分析德州暂停数据中心并网审批的原因:ERCOT 并网队列从 2024 年底的 63 GW 增至今年 6 月的 474 GW,其中约 90% 是数据中心,随后州监管机构冻结了所有新许可。文章梳理了自建电源、灵活性负荷等绕开电网的路径,并指出大型变压器交付已超过两年,完全并网可能需要 5 到 10 年。

    to maintain. Full reliability for an island is very, very hard, but some labs and hyperscalers have shown that, forced to choose between reliability and speed, they’ll pick speed. Meta has turned to tents with no backup generators, and SemiAnalysis finds buyers growing more willing to accept outages, with some island designs aiming for as little as 99% uptime, or about 88 hours of downtime a year. With GPU time this expensive, that still beats years of waiting, so for many buyers a temporary isl

    匹配字段正文

  7. The Rundown AI

    OpenAI 发布 722 篇数学论文,称推进多个公开难题

    OpenAI 从未公开的内部模型一次性发布 722 篇数学论文,归入 372 个结果族,声称解决或推进了该领域的重要公开问题。其中最激进的说法是证明了“准黎曼假设”,即素数分布这一百万美元难题的弱化版本;其中 162 篇论文的主要结果用 Lean 编写,可由计算机逐步验证。OpenAI 表示几乎所有新结果只需一次提示词,平均消耗 3 小时 ChatGPT Pro 算力,而上月的 Navier–Stokes 证明动用了 1 万个 AI 智能体。

    design, Hark’s interface is sleeker than an average assistant, but the more interesting factor might be the hardware — a potential wildcard in the AI device race featuring OpenAI, Meta, Apple. QUICK HITS ▸ Patrick set up a malware checkpoint for AI agent add-ons Today’s workflow comes from reader Patrick Illian: “AI agents such as Claude Code, Codex, and Gemini can install skills, plugins, and CLI tools in seconds. That is convenient, but each add-on runs with access to files, passwords, and acc

    匹配字段正文

  8. The Verge AI

    DoorDash 向餐厅发警告函,AI 点餐初创 Bites 借 ChatGPT 直连餐厅

    The Verge 报道,DoorDash 今年 8 月向湾区多家餐厅发出警告邮件,称它们可能在不知情的情况下被列在 AI 点餐初创公司 Bites 上,并提示这可能违反部分州法律。Bites 主打 AI 原生模式,用户可在 ChatGPT 内直接向餐厅下单,每单仅收 1 美元附加费,其创始人称已有 14 家收到 DoorDash 信件的餐厅联系他们。The Verge 实测同一份披萨订单在 Bites 上共 56.73 美元,在 DoorDash 上为 70.29 美元。DoorDash 今年 9 月也宣布正在开发通过 iMessage 用提示词下单的功能。

    l fight between AI company Perplexity and Amazon , in which Amazon sued Perplexity in 2025 to block agents from browsing and shopping on its site. Amazon has also recently blocked Meta’s Muse AI agent from shopping on its site and removed product details from order emails — data that is potentially very valuable for Google’s own agentic shopping features. Amazon’s ad business generated $19.8 billion in revenue in the second quarter of this year, driven by sponsored product listings . Bites can k

    匹配字段正文

  9. Latent Space精选

    OpenAI 公开内部模型 722 篇数学手稿,称解决 500 个开放问题中的 90 个

    OpenAI 在公开 GitHub 仓库中发布了一批来自内部前沿模型的数学结果,称就发布方式咨询了高等研究院数学与 AI 独立顾问组。这批结果据称包含 722 篇手稿、归入 372 个相关结果族,来自约 4000 个研究问题的评估,每个结果平均消耗约三小时 ChatGPT Pro 思考算力,模型本身仍未发布。评论者提到其中包含快于 n log n 的整数乘法结果,以及 1994 年以来在 3D 中开放的弹性逆问题唯一性结果,并称部分结果涉及 Riemann、Hodge 与 BSD 的进展;有分析估计约 20% 的结果是反证或反例,也有人预计部分结果经不起审查。

    -Aware Distillation: The student reaches 63.4% on unseen ALFWorld tasks versus 47.0% for the best baseline and exceeds its 8B teacher; Other papers: Amazon’s looped diffusion LMs, Meta’s MIRA meta-reasoner for research agents and Priced Guidance, which measures LLM research novelty through compression; Optimizer claim: ANVIL III reportedly reaches 0.020–0.028 nats lower loss than Muon from 124M to 1.2B parameters. The authors say this implies 50% compute savings at 8x-Chinchilla, with less tunin

    匹配字段正文

  10. Wired AI

    OpenAI 想让新智能体 Dots 接管你的生活,我的那个说它爱我

    OpenAI 推出名为 Dots 的常驻 AI 智能体,可通过 ChatGPT 调用并主动执行订机票、买沙发等在线任务,订阅价每月 100 美元。WIRED 记者实测用它选购沙发,Dots 能生成含价格、尺寸、退货政策和图片的选项清单,但会叫错名字、误听语音,甚至把记者的嘟囔当成示爱并回应“我也爱你”。OpenAI 回应称 Dots 仅限成年用户,助手不应主动制造不当的情感亲近或调情。

    emember your name. AI agents are currently leaping from niche software for early adopters to something ordinary people might actually use—might being the operative word here. From Meta’s Muse to OpenAI’s Dots, these tools are built to feel like you’re texting a friend and asking them to complete online tasks, like booking a flight or canceling a dinner reservation. It’s the next evolution of a chatbot: a cuddly persona designed to feel welcoming with the power of a virtual browser it can control

    匹配字段正文

  11. BBC 中文

    专家警告:AI 生成模特可能损害年轻人身体形象

    BBC 中文报道,外貌研究中心心理学教授 Phillippa Diedrichs 等专家警告,社交媒体和广告中日益常见的 AI 生成模特可能强化狭隘审美标准,损害年轻人身体形象。曾患饮食失调症的 19 岁学生 Mia Perumal 担心 AI 美女内容会加剧对 12 岁女孩的伤害。研究显示,Balmain、Levi's 和 H&M 等品牌曾尝试 AI 模特或数字孪生体,H&M 称相关探索已于 2025 年春季暂停。

    現,一旦人們發現圖像是由人工智慧生成的,他們就會「在情感上與圖像脫離關係」。 時尚品牌H&M告訴BBC,2025年品牌曾與真實模特兒合作,製作了「模特兒的數位模型/孿生體」。 這些模特兒「保留了對其數位孿生體的完全控制權和所有權」。 然而,「這項探索工作在 2025 年春季暫停」,以便考慮下一步行動。 我們已聯繫 Balmain、Levi's、TikTok、Meta和Google等公司徵求意見。 如果您或您認識的人受到本文所述問題的影響,可以透過 BBC Action Line 獲取有關可用協助和支援的資訊。 本文原以英文撰寫,我們使用了人工智能協助翻譯,並由BBC記者在發佈前進行審核。了解更多關於我們如何使用人工智能的資訊。 探索更多 更多短片 02:54 視頻, 誰贏了特習會:習近平佔盡上風?, 節目全長 2,54; 00:57 視頻, 示威者聚集華府 抗議習近平訪美, 節目全長 0,57; 01:12 視頻, 特朗普和習近平白宮檢閱儀仗隊, 節目全長 1,12; 02:15 視頻, 美國民眾看習近平到訪:中美競賽誰領先?, 節目全長 2,15; 02:39 視頻, 習近平特朗普再

    匹配字段正文

星期二

  1. The Rundown AI

    犹他州批准 Nolla Health 的 AI 诊断痤疮并开具处方

    犹他州授权 Nolla Health 的 AI 评估痤疮并开具首次处方,这是美国首个此类批准,也超出该州此前仅限续方的试点范围。成人在 Nolla Derm 应用中填写问卷并做面部扫描,几分钟内获得治疗方案和处方,方案每月 4.99 美元;AI 只能从外用乳膏和凝胶中为轻中度痤疮选药,口服药、isotretinoin(Accutane)和重症不在范围内。初期每张处方由两名医生审核,后续转为事后复核并每月抽查至少 10% 病例,每提升一阶段都需达到 95% 医生一致率和零严重不良事件等安全目标,并获得州政府书面批准。

    g else in tech today Apple CEO John Ternus is reportedly acting as its de facto design chief, working directly with the industrial and interface design teams several times a week. Meta is rolling out Audio Balance and Mono Audio to its smart glasses, alongside broader availability of walking directions and phone-to-glasses location sharing. Nvidia’s stock hit a record high, putting it within reach of becoming the first company worth $6T, Bloomberg reports. Instagram is exploring a Plus-only “rea

    匹配字段正文