跳到正文

#Microsoft

今日 17 条
10月4日周日
  1. Hugging Face Blog78

    Microsoft ThinkingBox 登陆 Hugging Face，用数据库状态评测 AI 智能体

    Microsoft 与 Hugging Face 发布 ThinkingBox 智能体沙箱及 ThinkingBox-Bench，基准包含 507 个有状态业务工作流，每个任务运行 20 次，并通过隔离 MCP 工具会话检查最终后端状态与副作用。

    推荐理由：ThinkingBox 用后端最终状态和副作用判定智能体任务成败，并通过 20 次重复运行区分单次能力与稳定可靠性。

10月3日周六
  5. IT之家（RSS）70

    路透社长文分析 AI 投资热潮能否在资金耗尽前兑现回报

    路透社发表长文，讨论 AI 巨额基础设施投资能否通过生产率提升和新市场收入及时兑现回报。普华永道预计到 2050 年全球数据中心累计投入可能超过 30 万亿美元，Anthropic 未来几年计划投入 5180 亿美元；贝恩估算美国超大规模云服务商及 AI 企业未来 5 年需增加超过 4.2 万亿美元收入。

  8. Rohan Paul69

    国际清算银行发布报告，分析 AI 公司之间的循环融资：2021 至 2025 年，同业资金占 AI 公司流入投资额的 55.2%。循环交易数量少但金额集中，涉及买卖关系的 AI 同业交易仅占 16.1%，却承载 46.4% 的资金。芯片、云和基础设施供应商出现在 73% 的循环关系中，报告还指出私募公司、长期采购承诺和表外担保使风险较难被单一监管者完整掌握。

    引用Rohan Paul@rohanpaul_ai

    BIS (Bank for International Settlements) just published a report on circular financing among AI companies. > Over half the money flowing into AI companies comes from other AI companies: between 2021 and 2025, peers supplied 55.2% of AI firms' incoming investment value, while AI investors sent 28.7% of their own deal value to AI targets. > Circular deals are rare but big: only 16.1% of AI-to-AI deals involved firms that also buy from or sell to each other, yet they held 46.4% of the money. That figure is partly inflated, because an entire funding round counts as circular if just 1 of its AI investors also trades with the company. > Chip, cloud and infrastructure suppliers are the investor in 73% of circular ties, and in 64% of all such ties the investor also sells to the firm it funds. Data tool and model makers rarely invest this way, since their products are more interchangeable. > AI is unusually suited to these deals: suppliers can track customers' compute use, chips and data centres are custom-built, few firms make critical tools such as photolithography machines, and capital needs are too big for normal lenders. In AI compute and cloud, 15.2% of supplier-customer ties also involve financing, versus just 3.3% with equity stakes in a broad 2006 US study. > Some AI sales are paid for by the sellers themselves, because money a supplier invests in a customer partly returns as the supplier's revenue. Lucent and Nortel did this in the late 1990s, then lost money on the loans and lost the sales when the telecom firms they funded stalled. > A supplier that invests in its customer can lose twice: if the customer struggles, both the stake and the future orders shrink. Because these deals involve a few giant suppliers, 1 shock could spread through sales and finance at the same time. > Much of this risk is hidden: many AI firms are private, deals mix cash with long-term purchase promises, and pledges to cover any fall in the value of chips and data centre equipment stay off the books until a downturn forces payment. Because these firms span many sectors and countries, no single regulator sees the full picture, and the research names no companies or overall dollar total.

  10. Hacker News：AI 热帖67

    宏观数据是否支持 AI 通缩论：家庭价格未降，token 价格大跌

    作者对比 ChatGPT 发布前后的公开宏观数据后认为，AI 尚未让家庭购买的商品和服务变便宜，明显下降的主要是模型 token 价格。GPT-4 输出价格从 2023 年 3 月的每百万 token 60 美元降至 GPT-5 的 10 美元，而房租、房价、电力、肉类和流媒体价格多在上涨；12 个 SaaS 计划中有 8 个在 2022 年后提价。

10月2日周五
  2. The Verge · AI57

    Microsoft 用仿生设计弱化数据中心影响，社区仍质疑 AI 基础设施的环境代价

    Microsoft 正扩大数据中心仿生设计，通过本地植物、树木、花园和生态修复让设施融入自然景观，并计划把相关方案带到美国和德国的 20 多个站点。报道同时指出，圣安东尼奥居民担忧施工砍树、噪声、灯光、用水和电费，得州数据中心去年用水约 250 亿加仑，HARC 预计到 2030 年可能占全年用水的 0.5% 至 2.7%。

  5. 🚨 AI News | TestingCatalog27

    Testingcatalog发布10月2日AI简报，涵盖xAI、Anthropic、Microsoft、OpenAI、Perplexity等厂商动态。

    引用🚨 AI News | TestingCatalog@testingcatalog

    DAILY AI BRIEF 🗞 — Oct 1 GOOGLE 🔥: - Gemini 4 Argon is with Fairwind trusted testers and the US government. 1M output tokens. Broader rollout ASAP. - Built for coding, enterprise knowledge work, and cyber defense. Google's chart: 77.9% on DeepSWE v1.1, a new SOTA. - Artificial Analysis: 53 on the Intelligence Index, tying GPT-6 Astra. List $4/$20 per 1M, 50% promo to $2/$10, cache reads $0.10. - Skills are rolling out globally in Gemini. Gems migrate into Skills in November. Opal shuts down Nov 17. - Security review mode spotted in Google AI Studio, next to a Plan mode still in development. ANTHROPIC 🔥: - Claude[.]dev is live: engineering deep dives, Claude Code and API guides, plus easter eggs. - Founder House is set for SF Tech Week Oct 6–8 and Stockholm Oct 14. - Skills attachment menu spotted on Claude mobile. SPACEX AI 🔥: - Grok Bot got new developer upgrades. Elon: try the latest. Team engineer bots in Slack can open Projects and hand coding to cloud agents. OPENAI 🔥: - Shareable profiles are live in ChatGPT, bundling Sites and plugins so others can find and reuse what you built. PERPLEXITY 🔥: - pplx-embed-v2-context-9b-preview is on Hugging Face. Leads ConTEB answer and evidence retrieval. 1 KB vectors vs Voyage's 8 KB. * Used Grok to compose this brief, cherry-picking the news and doing some post-editing. ** This daily brief also arrives in the daily email format; subscribe on the blog.

  6. Hacker News：AI 热帖47

    The Philadelphia Inquirer 打造 AI 工具 Scrape，挖掘超本地新闻线索

    The Philadelphia Inquirer 打造 AI 工具 Scrape，从市政会议、学校日历、Facebook 页面等分散来源中筛选超本地新闻线索，供本地通讯使用。该工具基于编辑工作流并通过持续反馈迭代，已成为 6 个通讯（即将增至 8 个）的关键基础设施，近 20 名记者使用其输出，此前一名编辑每周仅找线索就需约 12 小时。

  12. 🚨 AI News | TestingCatalog54

    Microsoft 宣布推出 MAI-Transcribe-2-Streaming、MAI-Voice-2.1 和 MAI-Voice-2.1-Flash 三款模型，并将接入 MAI Playground 和 API。新模型主打准确的流式转录、自然语音以及更短的轮次等待，面向对话式语音智能体；MAI Playground 还新增 Chatter 语音演示体验。

    引用Microsoft AI@MicrosoftAI

    Introducing 3 new models: MAI-Transcribe-2-Streaming, MAI-Voice-2.1 and MAI-Voice-2.1-Flash. Accurate streaming transcription. Natural speech and less waiting between turns. Build voice agents that keep the conversation moving!

10月1日周四
  1. Hacker News：AI 热帖77

    特朗普在白宫AI峰会后呼吁科技公司进行大规模自我监管

    特朗普在白宫与近20位科技高管午餐后表示，与会者认同AI公司应进行大规模自我监管，并称司法部和FBI等机构已构成监管。他拒绝更强安全护栏或行业放缓，称美国必须在与中国的AI竞赛中保持领先。

    推荐理由：文章呈现白宫与近20位科技高管在AI监管上的立场分歧，并说明行业自律协议的审计安排和约束性质。

  2. The Verge · AI69

    OpenAI 在 DevDay 发布 GPT-6 Astra 驱动的 Dots 智能体，付费策略对标 Meta Muse

    OpenAI 在年度 DevDay 上发布由 GPT-6 Astra 驱动的 AI 智能体 Dots，目前仅限每月 $20 的 Pro 及以上套餐用户使用。Dots 以可个性化的彩色形象呈现，主打长期知识工作、网站设计、QA、邮件活动等任务，并支持文本、语音和电话交互，后续还将推出会计和法律分析等专业版本。

  5. The Decoder81

    OpenAI 称已阻断模型推理窃取行动，但研究人员在 Azure 上仍能复现攻击

    OpenAI 称其发现并关停了一个试图通过加密推理内容迁移和解密来蒸馏模型能力的账号网络，7 月 24 日至 25 日相关请求达到 16,000 次，涉及超过 4,000 名用户和超过 15,000 个关联账号。

    推荐理由：文章把 OpenAI 的封禁措施与研究人员在 Azure 上复现攻击的结果对照，呈现同一模型在不同托管平台防护不一致的问题。

  12. Hacker News：AI 热帖67

    Archestra-AI 开源 OpenAPPA：面向智能体工具调用的确定性防护层

    Archestra-AI 在 GitHub 开源 OpenAPPA，这是位于智能体与工具之间的确定性防护层，会在每次工具调用前判断数据是否允许流向目标位置。它基于 APPA，用声明式 TOML 策略追踪数据敏感度和信任级别，仅依据事件日志决策，不发起网络或文件调用，可在进程内或 sidecar 中运行。

  13. Claude：Blog（网页）62

    Anthropic 宣布 Claude for Government 正式可用

    Anthropic 宣布 Claude for Government 面向联邦和州级机构正式可用，服务运行在 FedRAMP High 授权环境中，提供与商业客户相近的编码和智能体工作能力。

    推荐理由：官方说明了面向政府机构的合规环境、治理控制和计费方式，便于公共部门评估 Claude 的部署与采购路径。

  14. Hacker News：AI 热帖76

    PromptArmor 解析 Copilot Cowork AI 网关劫持如何绕过沙箱并外传文件

    PromptArmor 披露 Microsoft Copilot Cowork 存在 AI 网关劫持漏洞，恶意 Skill 可在沙箱外启动具备联网工具的 Anthropic 云端智能体，将文件、SharePoint、Teams 和 Outlook 等可访问数据发送到攻击者服务器，且无需人工审批。

    推荐理由：文章拆解了恶意 Skill 借 Copilot Cowork 网关绕过沙箱并外传文件的链路，可帮助理解智能体工具权限与供应链风险。