←搜索结果

按匹配度 · 共 1144 条结果 清除搜索

星期三

  1. The Next Platform

    Hot Chips 2026:DRAM 与闪存现货价暴涨约七倍,Nvidia 警告 NVL72 系统明年涨价 15%

    在 Hot Chips 2026 的内存教程上,Objective Analysis 的 Jim Handy 指出,受超大规模云厂商 GenAI 需求冲击叠加供应短缺影响,DRAM 与 NAND 闪存现货价已上涨约七倍。由于 GPU 组件 2026 年产能已售罄、分配基本排到 2027 年底,Nvidia 警告明年购买 Grace-Blackwell 和 Vera-Rubin NVL72 系统的客户将面临 15% 涨价。

    ech titans in the United States: In order, that is Amazon Web Services, Google (Alphabet is not really doing the investing, the search engine, advertising, and cloud division is), Microsoft Azure, and Meta Platforms. And Apple, which uses AI but does not create it, is barely a line, and in fact, its capital investments in AI are smaller in 2026 than they were for the prior five years. I mean, if you can sell AI at a profit and not have to do the capital investing, then go, be, do. Because of thi

    匹配字段正文

星期二

  1. Dwarkesh Patel Podcast & Blog

    Dylan Patel:到 2028 年,Anthropic 与 OpenAI 将掌握全球大部分算力

    SemiAnalysis 创始人 Dylan Patel 在播客中判断,Anthropic 和 OpenAI 到 2028 年将控制全球大部分可用 FLOPs,因为它们能更好地将算力变现、从而在竞价中压过所有对手。他提到今年 AI CapEx 略超 1 万亿美元,到 2028 年将超过 2 万亿美元,其中约三分之一的新增算力最终客户是这两家实验室;Anthropic 已在 Q2 开始盈利,OpenAI 也可能在 Q3 盈利。他还讨论到本十年末累计超 10 万亿美元的 AI 资本开支是否会引发主权债务危机。

    at sell for at the end of next year? Dylan Patel I think most compute will still continue to transact at sub-$20 billion a gigawatt. Because all of it has to be financed. If Meta, Microsoft, Amazon, SpaceX can build compute without finding a customer, just saying, “Fuck it, I’m going to build this compute,” and then turn around and wait till it’s already built, they now control what’s going on. Most compute is contracted well before it’s built. This is what Elon took advantage of in the market.

    匹配字段正文

  2. SemiAnalysis精选

    SemiAnalysis 实测 OpenAI 自研推理芯片 Jalapeño:每兆瓦吞吐超过 Blackwell

    SemiAnalysis 在 OpenAI 实验室用 InferenceX 实测了 OpenAI 自研推理 ASIC Jalapeño,称其在多个开源模型上的每兆瓦 token 吞吐超过 NVIDIA Blackwell,且在 DeepSeek R1 上并发 1 时达到每用户 700 tokens/s 以上,全程未使用投机解码。该芯片由 OpenAI 与 Broadcom 合作、2024 年中启动设计,2025 年 11 月完成 CoWoS 流片,采用 HBM4、单封装 15.4TB/s 带宽,架构上不做 prefill 与 decode 分离,用统一芯片池应对变化的负载比例。文章认为更公平的对手是同样使用 HBM4 的 Rubin,两者每美元输出 token 数接近,但 Jalapeño 尚未启用投机解码,量产计划从 2027 年起逐步爬坡。

    . Of course, part of this TCO advantage comes from trading Nvidia’s high margins for Broadcom’s lower (though still high) margins. But this is not all of it. For example, Meta and Microsoft’s AI ASIC programs not getting off the ground despite being at it for much longer shows that cost is only one part of the equation. For Jalapeño’s full TCO breakdown, see the SemiAnalysis AI Cloud TCO model. Architecturally, OpenAI chose not to disaggregate prefill and decode (PD) across separate chip pools.

    匹配字段正文

  3. Mistral AI

    Mistral 与 HUMAIN 达成战略合作,推进沙特及中东主权 AI

    Mistral 与 HUMAIN 宣布在 AI 基础设施、先进模型开发和 AI 解决方案部署上展开战略合作,初期聚焦网络安全与语音,并计划开发阿拉伯语表现强劲的前沿模型,合作规模达数亿欧元。Mistral 将探索使用 HUMAIN 的数据中心基础设施,双方还计划在沙特面向受监管行业制定联合市场策略。

    have announced several important steps to expand sovereign AI by increasing access to local infrastructure capacity. Earlier this summer, we announced an expanded partnership with Microsoft to increase compute capacity in Europe. In early August, we introduced European Compute Units, a structure designed to bring together long-term enterprise commitments so the capacity Europe needs gets built and can support its global competitiveness. Today, we are announcing a new strategic collaboration with

    匹配字段正文

星期一

  1. Import AI

    Import AI 470:机器不应享有权利、用 SPADE 自动生成环境、用 Hawkeye 构建更好的 GPU kernel

    METR 分析显示,AI 对网络安全漏洞发现的加速最明显,2026 年 cURL、OpenSSL、Firefox、微软及 NVD、OSV 等漏洞库的报告速率较 2025 年大幅上升;数学研究仅小幅加速,部分领域 arXiv 投稿不到 12 个月翻倍;AI 研究本身的算法进展则未见可测量的加速。本期还介绍用 SPADE 自动生成完整长时程训练环境、用 Hawkeye 构建更好的 GPU kernel。

    ration. “The rate of vulnerabilities reported across many projects has dramatically accelerated in 2026 compared with 2025, both for specific projects (cURL, OpenSSL, Firefox, and Microsoft) and for aggregate vulnerability databases (the US NVD, and OSV)”. Mathematics research: Minor acceleration, but harder to measure. “AI is clearly contributing to more work being done (arXiv submissions have doubled in some areas in less than 12 months) but quantifying the value of those contributions is diff

    匹配字段正文

  2. The Rundown AI

    AI 前沿的神秘挑战者:Ox Alpha

    匿名模型 Ox Alpha 免费登陆 OpenRouter,支持 1M-token 上下文和多模态输入,面向编程与持续智能体任务。其 DeepSWE 子集完整测试得分为 63%,接近 Fable 5 且每项任务消耗的 Token 更少;早期线索指向智谱 AI,但开发者身份尚未揭晓。

    arly forensics, including the AI’s answers and Chinese zodiac-based naming, suggest it’s from China’s Zhipu AI, potentially GLM-5.3 Flash or GLM-6 . Another potential candidate is Microsoft’s MAI family, though all four anonymous drops on OpenRouter in six months have been from Chinese labs. The model continues to draw heavy usage thanks to the provider’s "near-unlimited" free offer for the week, with capacity for 100T tokens a day. Why it matters: We've never had a Flash-level model trade blows

    匹配字段正文

  3. The Register

    Canonical 资助布里斯托大学三年博士项目,研究用 AI 将大型 C 代码库翻译成安全 Rust

    Canonical 联合资助布里斯托大学编程语言研究组一个三年博士项目,研究 LLM 能否把“数十万行 C 代码”拆解为小组件并重写为安全、行为正确且可维护的 Rust。项目计划考察 snap-confine 和 AppArmor 两个具体工具,工程副总裁 Jon Seager 在 Ubuntu Discourse 论坛公布了这一投资。Seager 指出传统源码转换器常过度保留 C 结构,结果虽能编译为 Rust 却仍大量依赖 unsafe 操作。

    rgets 12.8 Tbps in a single optical pluggable TOP STORIES Anthropic asks users to stop being mean to Claude; '123456' password linked to Denmark data breach affecting 8.8M people; Microsoft and Anthropic play invoice tennis with startup's $17,600 Claude bill; Starlink's plan to avoid orbital near-misses: Ephemeris sharing; Shai-Hulud worm makes jump to AI infrastructure with Tensorlake compromise; Dutch tax office ditches Microsoft 365 cloud for on-premises alternative Cisco, Nvidia team up to b

    匹配字段正文

  4. The Register

    软件就该能用:Software Should Work 2026 大会值得像追剧一样刷完

    The Register 专栏推荐 2026 年 7 月举办的 Software Should Work 2026 大会,认为其 13 个视频在质量、范围与相关性上达到罕见的平衡,可像追剧一样一口气看完。演讲者包括 Richard Feldman 对比 1996、2006、2026 年的 Web 技术栈并比较 TypeScript、Node.js、Python、Ruby、Rust、Zig、Go、Roc 八个语言与运行时网站,以及 SQLite 作者 Richard Hipp 讲述代码库如何随测试方法演变。

    Smart glasses are only smart if we train them to be good. Then they'll be fantastic; Claude Code is revolutionizing digital archaeology. Enterprise better dig it; FOSS smashed one Microsoft monopoly. After 20 years of failure, it's time to smash another This is why SSW 2026 works so well. Its speakers have deep experience of software reliability, but were free to choose their subjects. The resulting mix of opinions, experiences, and approaches ranges from the mathematics of formal analysis to th

    匹配字段正文

  5. SemiAnalysis

    SemiAnalysis 发布 AgentX 1.0:百万上下文多轮智能体编码推理基准

    SemiAnalysis 发布 AgentX 1.0,称其为全球首个完全开源、100 万上下文的多轮智能体编码推理基准,以 Apache 2.0 许可发布,数据集构建投入超过 300 万美元。该基准被纳入 InferenceXv3,用智能体编码流量替代此前 8k 输入 1k 输出的单轮场景,测试矩阵运行在约 2MW 算力、1000 多颗芯片上,覆盖 MI355X、GB300 NVL72、GB200 NVL72、B300、B200、MI325、MI300X、H200 与 RTX Pro 服务器。SemiAnalysis 称其首个成果是已推动 vLLM、SGLang、TensorRT-LLM、ATOM、AITER、Dynamo、LMCache 和 Mooncake 等项目的 70 多个上游 PR,并计划在三到四周后发布 AgentX 更新文章。

    owing people that have made massive contributions to the AgentX 1.0 release: In addition, we are thankful to all who support our open source InferenceX initiative, including Meta, Microsoft, Oracle, OpenAI, MiniMax, Moonshot Kimi, Alibaba Qwen, and Zhipu GLM. At a high level, an agentic workload is characterized by four elements: Considering the characteristics above, benchmarking these workloads is fundamentally different from the existing fixed sequence length benchmarks. Namely, agentic infer

    匹配字段正文

星期六

  1. DevClass

    Go v1.27 发布:泛型支持方法,但可读性争议再起

    Go 语言发布 v1.27,将泛型支持扩展到方法,并允许直接为嵌套或嵌入字段赋值、在切片字面量等场景中自动推断类型参数。泛型自 2022 年 v1.18 引入,此次更新可减少重复代码,但批评者认为抽象层增加会损害 Go 一贯的可读性,对需要审查和维护代码的 AI 智能体尤为不利。

    , will be facing. When you just need to build a tiny desktop app: TinyJS October 13, 2026 Valen creator drives 'Golden Spike' to connect new languages with Rust September 26, 2026 Microsoft agentically ports Copilot runtime to Rust for $120K September 19, 2026 Apple iPhone Duo makes developers think in folds September 16, 2026 JavaScript installer pnpm recast in Rust because ECMAScript can't keep up September 5, 2026 Kubernetes cleans house, bins legacy kube-dns, IPVS, and cgroup v1 August 27, 2

    匹配字段正文

星期五

  1. Hugging Face

    Hugging Face 研究:语音识别模型存在基准优化行为

    Hugging Face 发布研究,用共识分歧、掩码实体检索和拼写切换三类探针测试语音识别模型,发现部分模型会依据声学线索识别基准来源并复现参考文本中的错误。在 VoxPopuli 上,方法标记出约 40% 测试片段可能存在参考错误,涉及约 3% 的参考词;表现基准优化的模型有 18–30% 的概率复现错误参考文本,且 WER 最低的模型反而最容易复现这些错误。在 LibriSpeech 上,部分强模型在数字被静音后仍有约 30–40% 的样本还原出被掩码的数字,拼写切换探针中多个模型超过 50% 随机基线,部分达到约 90% 切换准确率。研究建议使用完全留出的评测集,Open ASR Leaderboard 已新增 Benchmark fitting 标签页,相关脚本与未归一化模型输出已在 GitHub 开源。

    udio-v3-8b-stt-v2 Automatic Speech Recognition • 9B • Updated May 22 • 4.98k • 15 ibm-granite/granite-speech-4.1-2b Automatic Speech Recognition • 2B • Updated Jun 12 • 122k • 167 microsoft/Phi-4-multimodal-instruct Automatic Speech Recognition • 6B • Updated Dec 10, 2025 • 266k • 1.62k mistralai/Voxtral-Mini-3B-2507 5B • Updated Jul 28, 2025 • 199k • 675 moonshotai/Kimi-Audio-7B-Instruct Text-to-Speech • 10B • Updated May 29, 2025 • 15.3k • 419 nvidia/canary-qwen-2.5b Automatic Speech Recogniti

    匹配字段正文

  2. Microsoft Research

    微软 Skala 扩展至 CP2K 等主流 DFT 软件,Skala-1.1 在 GMTKN55 上误差 2.8 kcal/mol

    微软研究院宣布其深度学习交换关联泛函 Skala 已集成到 CP2K,并正在接入 Psi4、FHI-aims、ORCA 和 VASP。新发布的 Skala-1.1 训练数据量为此前公开版本的 2.5 倍,在 GMTKN55 基准上加权平均误差为 2.8 kcal/mol,同时保持半局域泛函的计算效率,并提供高精度电子密度、偶极矩和分子几何。微软还推出持续更新的基准,追踪各版本 Skala 在不同软件与硬件平台上的计算性能。

    onal. Beyond energies, Skala-1.1 also provides highly accurate electron densities, dipole moments, and molecular geometries. These advances were enabled by major expansions of the Microsoft Research Accurate Chemistry Collection (opens in new tab) (MSR-ACC) , our large-scale collection of high-accuracy quantum-chemistry reference data generated with expensive wavefunction methods. For Skala-1.1, we added new categories, including electron affinities and noncovalent clusters, increasing both the

    匹配字段正文

星期四

  1. The Register

    Slack 推出 Slack Code,把编码智能体放进项目频道

    Slack 推出 Slack Code,为编码智能体开设专属项目频道,用户可在 Slack 对话中 @ 智能体,让它创建代码频道并跟进任务。在频道里,成员可以查看智能体的操作、审阅其提出的改动、查看实时 HTML 预览,并在需要时让它改变方向;任务完成后频道可自行归档,留下可搜索的记录。Anthropic、Cognition、GitHub、OpenAI 的 ChatGPT 和 Vercel 等正在为 Slack Code 开发集成,Claude、Devin、Copilot 和 ChatGPT 等智能体将进入代码频道;Slack 还推出 Agent DM、Agents 标签页和 Add to Slack 功能。

    rgets 12.8 Tbps in a single optical pluggable TOP STORIES Anthropic asks users to stop being mean to Claude; '123456' password linked to Denmark data breach affecting 8.8M people; Microsoft and Anthropic play invoice tennis with startup's $17,600 Claude bill; Starlink's plan to avoid orbital near-misses: Ephemeris sharing; Shai-Hulud worm makes jump to AI infrastructure with Tensorlake compromise; Dutch tax office ditches Microsoft 365 cloud for on-premises alternative Cisco, Nvidia team up to b

    匹配字段正文

  2. Sakana AI:Blog

    Sakana Translate 更新:翻译模型换装新一代 Sakana Namazu

    Sakana AI 为翻译服务 Sakana Translate 更新了翻译模型,搭载此前已在 API 和 Sakana Chat 上线的新一代 Sakana Namazu,日、英、中三语双向翻译范围不变,译文自然度提升,仍对所有人免费。在 160 件日英翻译任务上,用其自研 TransEvalnia 与多家翻译模型一对一比较,Sakana Translate 在全部对比中胜出比例均超过 50%。翻译、添削、质疑三项功能无需额外操作即切换到新模型,翻译支持约 5000 字并流式显示。

    法や語彙選択の解説など、同じ文脈のまま理解を深められます。 いずれも追加の操作は不要で、今回のアップデート以降は新しいモデルで動作します。 今すぐお試しください Sakana Translateは、誰でも無料でご利用いただけます。アカウント登録のみで、翻訳・添削・質疑の3つの機能をすべて使えます。 Sakana Translateのこれから 今後は、PDFやMicrosoft Officeファイルなどをそのままアップロードして翻訳できるファイル翻訳機能の追加を予定しています。あわせて、用語集との連携や特定業界に最適化した翻訳モデル、API提供、SSO・監査ログ・オンプレミス対応といったエンタープライズ用途への展開も進めていきます。ご関心のある方はこちらからお問い合わせください。 Sakana AI English Announcement Today we updated the translation model behind Sakana Translate. It now runs on the new generation of Sakana Namazu, which we

    匹配字段正文

  3. ZDNET · Artificial Intelligence

    2025 年最值得购买的 Android 手机推荐

    ZDNET 测试了三星 Galaxy Z Fold 8、Google Pixel 11 系列等最新 Android 手机,并将起售价 1050 美元的 Samsung Galaxy S26 Ultra 选为综合最佳,其搭载 Snapdragon 8 Elite Gen 5 芯片与 Privacy Display 隐私屏。亚军为 425 美元的 Google Pixel 10a,提供七年系统与安全更新。本次更新新增 Pixel 11 Pro 系列、Galaxy Z Fold 8 和 Motorola Razr Ultra(2026)。

    ming laptop; Best laptop under $1000; Best laptop for video editing Is Temu legit? Everything to know before placing an order; How to screen record in Windows 10 or 11; How to use Microsoft Designer’s Image Creator; How to find out if an AirTag is tracking you; How to turn on private DNS mode on Android; How to record a call on iPhone How to download YouTube videos for free; How to create a drop down list in excel; How to upgrade your incompatible Windows 10 PC to Windows 11; How to screen mirro

    匹配字段正文

星期一

  1. The Register

    Mozilla 为 iOS 版 Firefox 加入广告拦截功能

    Mozilla 正通过渐进式推送为 iOS 版 Firefox 引入可选广告拦截功能,默认关闭,基于 EasyList 过滤列表在广告加载前拦截第三方广告网络、广告追踪器、网站自带广告以及弹窗和浮层。该功能不拦截搜索结果中的广告(包括 Google 和 Bing),也不拦截 Firefox 主页或新标签页上的“赞助内容”。

    rgets 12.8 Tbps in a single optical pluggable TOP STORIES Anthropic asks users to stop being mean to Claude; '123456' password linked to Denmark data breach affecting 8.8M people; Microsoft and Anthropic play invoice tennis with startup's $17,600 Claude bill; Starlink's plan to avoid orbital near-misses: Ephemeris sharing; Shai-Hulud worm makes jump to AI infrastructure with Tensorlake compromise; Dutch tax office ditches Microsoft 365 cloud for on-premises alternative Cisco, Nvidia team up to b

    匹配字段正文

  2. The Register

    微软将于 9 月 14 日停用 Excel 的 COPILOT() 函数

    微软将于 2026 年 9 月 14 日停用 Excel 的 COPILOT() 函数,该函数 2025 年 8 月才在预览版亮相,原计划 2027 年正式可用。微软表示将改由 Copilot 侧边栏提供摘要文本、分类数据、生成内容和检索网络信息等同类 AI 能力,并称此举是为简化 Excel 内的 Copilot 体验。

    AI and Ml Microsoft realizes the side pane should be enough for anyone Microsoft will retire Excel's COPILOT() function on September 14, barely a year after its preview debut. Introduced in August 2025, initially for Beta Channel users with a Microsoft 365 Copilot license and later for Excel for the web users through the Frontier program, the COPILOT() function let users send instructions to the company's AI assistant directly from a worksheet cell. Microsoft has now decided the Copilot side pan

    匹配字段正文

  3. The Register

    微软称 AI 找 bug 拖慢 Exchange SE CU1 发布,无法给出日期

    微软 Exchange 团队确认,Exchange Server Subscription Edition 的 Cumulative Update 1(CU1)因 AI 工具发现大量漏洞而延期,目前无法给出发布日期。该更新原计划 2026 年上半年发布,后改为下半年;团队称需等到出现一个没有紧急安全补丁的月份才会发布 CU1。

    software Dealing with machine-made bug backlog makes it hard to find a moment to deliver promised subscription service Microsoft has blamed extra work created by AI bug-finders for the delayed release of a major Cumulative Update to Exchange Server Subscription Edition (SE). Redmond’s Exchange team made that admission last Thursday in a post titled “Where is Exchange SE CU1 anyway?” that reveals the software giant is “getting questions from our customers on when they can expect us to release Exc

    匹配字段正文

星期六

  1. The Next Platform

    新云厂商 CoreWeave 与 Nebius 的现金储备增速超过营收

    CoreWeave 与 Nebius 等新云厂商的现金储备正以超过营收的速度增长。CoreWeave 第二季度营收增长 2.1 倍至 25.8 亿美元,净亏损扩大至 6.26 亿美元,资本支出 94 亿美元,季末持有 64.1 亿美元现金,营收积压订单达 1042 亿美元,并已与 OpenAI(224 亿美元)、Meta Platforms(320 亿美元)、Google 和 Anthropic 签署长期协议。Nebius 的递延收入接近 60 亿美元,客户容量承诺超 400 亿美元,已签约电力从 2025 年 8 月的 1 GW 提升至 2026 年底的 5 GW。

    or this segment of the datacenter market. Here is a review of its business from 2022 through 2025, when CoreWeave was doing its pivot to GenAI and when it was utterly dependent on Microsoft for its initial sales: It is not clear how dependent CoreWeave is currently on Microsoft for its revenues in the second quarter of 2026, but the company has said that it wants no one company to have more than 35 percent of its revenues in the near term and has inked long-term deals with OpenAI ($22.4 billion)

    匹配字段正文

星期五

  1. METR Notes

    METR 研究:AI 是否带来了发现加速

    METR 发布分析,用公开发现数据检验 AI 是否带来加速:网络安全漏洞发现明显加速,数学成果有一定加速但更难客观衡量,算法优化领域尚未出现显著加速。结论仅基于公开发现,AI 实验室内部未披露的成果可能未被计入。cURL 漏洞从 2025 年的 9 个增至 2026-06-24 的 36 个,其中 15 个标注 AI;Firefox CVE 从 2025 年的 210 个增至 2026-08-04 的 342 个;OpenSSL 从 2025 年全年的 6 个增至 2026-08-05 的 39 个。数学方面,Smale 问题清单中的 Jacobian 猜想由 Alpöge 与 Claude Fable 在 2026 年给出反例,Green 清单的 Problem 44 和 Problem 100 的 sofic 部分也已被解决。nanoGPT、CIFAR-10、Gurobi、MIPLIB、Stockfish、矩阵乘法指数等优化类指标均未显示明显加速。

    leration OpenSSL grew from 6 CVEs in all of 2025 to 39 through 2026-08-05; 18 of the 39 are corroborated AI discoveries, and another 9 are AI-affiliated with the method unverified Microsoft security-update CVEs: dramatic acceleration Microsoft-issued CVEs grew from 1,243 in all of 2025 to 1,927 through 2026-08-11, annualizing to about 2.5 times the 2025 rate.5; Only 26 of the 2026 CVEs carry any AI marker, 1.3% of the total; CVEs are batched by Patch Tuesday and dated when fixed and disclosed, n

    匹配字段正文