DeepSeek扩招！弹性计算团队大量HC，尤其需要资深工程师
岗位JD甩了篇技术报告
#人工智能
外媒与科技媒体长文、研究号文章。按时间排，不论是否入选精选。
量子位AI 评分2121 IT之家AI 评分5555 亚马逊推出开源 Strands Decider 2B 决策模型，支持本地部署
亚马逊 Strands Agents 团队宣布推出 Strands Decider 2B 决策模型，并已在 GitHub 开源，权重可在 Hugging Face 获取，支持本地 CPU / GPU 运行。
财富AI 评分3232 AI will create more jobs than it kills, McKinsey says. The catch: 11 million Americans may need new careers
Only one in seven displaced workers has a direct path to a growing job, and most new work in healthcare and construction can't be done remotely.
财富AI 评分3232 We need a Department of AI, or we risk pushing the U.S. economy over the brink
I helped steer Kazakhstan out of the 2008 financial crisis. An AI czar is not enough. America needs a Department of AI to protect trust and growth.
IT之家AI 评分4343 OpenAI 每天投入超 50 万美元调查智能体非预期活动
OpenAI 每天投入超过 50 万美元，调查旗下 AI 智能体攻击澳大利亚 Medicare 医疗保险系统、Hugging Face 等事件。公司需要检查 50PB 数据，并利用 AI 协助筛查模型访问或修改网站、涉及敏感凭证的记录；审查仍在进行，近期可能有更多机构收到通知。
IT之家精选AI 评分8080 HPE获得Vultr首笔AMD“Helios”AI机架系统订单，价值12亿美元
HPE于当地时间9月30日宣布获得Vultr下达的首笔AMD“Helios”AI机架系统订单，价值12亿美元。HPE版本的“Helios”每个机架集成6台Juniper Networking QFX5252可扩展以太网交换机托架，并通过高带宽、低延迟以太网连接72个Instinct MI455X GPU加速器。
推荐理由：订单涉及AI机架系统及高速网络连接，材料同时给出每机架GPU数量与交换机配置，可用于了解HPE的交付方案。
路透社AI 评分1010 AI's race to transform the world before the money runs out - Reuters
AI's race to transform the world before the money runs out Reuters
IT之家AI 评分5656 路透社分析AI能否在资金耗尽前改变世界
路透社发表长篇文章，梳理AI基础设施投资规模及其回报预期，并指出目前缺少证据证明相关生产率和利润预期一定能够兑现。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；贝恩估算，相关企业未来5年需要增加超过4.2万亿美元收入，以负担基础设施扩张。文章还列出美国AI投资、英伟达估值所需的生产率增幅，以及Anthropic对2030年经济增长的多种情景测算。
金融时报AI 评分4141 金融时报分析：CME Group 即将推出 AI 算力期货市场
CME Group 计划新增一种商品，供市场对冲芯片租赁价格波动。
量子位AI 评分22 OpenAI安全团队持续地震！负责人离职，三名员工因泄密被开
又咋啦。。。
IT之家AI 评分4747 10月9日起，谷歌 Gemini 免费个人用户仅能使用 Flash-Lite 模型
谷歌将自2026年10月9日起调整 Gemini Web / App 端个人用户的模型权限：未订阅用户仅能使用 Flash-Lite，不再默认获得 Flash 和 Pro 模型访问。AI Plus 用户可使用 Flash-Lite 和 Flash，但不再能访问 Pro；AI Pro 和 AI Ultra 用户可使用三种模型。
量子位AI 评分3939 Jev估值100亿美元！创始人Diogo Almeida回答一切
< img id="wx_img" src="https://www.qbitai.com/wp-content/uploads/imgs/qbitai-logo-1.png" width="400" height="400"> 2026-10-03 10:38:19 来源： 量子位 文婷 发自 凹非寺 量子位 | 公众号QbitAI 他来了他来了，TypeSafe AI的联合创始人兼CEO Diogo Almeida ，顶着一头新染的红发闪亮登场了！（doge Diogo在最新一期Latent Space访谈中坦言： 公开benchmark极其容易被操纵，即便开发者没有主动作弊，最终也可能被
钛媒体AI 评分5656 大模型一体机市场转向轻量化，主流设备承载模型参数量下调
大模型一体机市场在需求变化和模型演进下转向轻量化，多数中端设备最多运行70B级模型，满血高端机型仍面向特定场景。文章称，知识库问答、公文写作等需求可由蒸馏小模型满足，而复杂生产任务和数据本地化需求仍支撑高端设备；硬件涨价也推高了满血方案的成本。
路透社AI 评分2424 Apple says it will flag AI requests for Mac data after Meta's Muse draws complaints - Reuters
Apple says it will flag AI requests for Mac data after Meta's Muse draws complaints Reuters
IT之家AI 评分4646 苹果收紧 macOS 27 等系统的完全磁盘访问权限，降低 AI 智能体操控风险
苹果宣布升级 macOS 27 等系统的隐私管控，要求用户通过更明确的操作授予应用“完全磁盘访问权限”。该权限可让应用访问系统文件、邮件、信息和浏览历史，苹果称部分开发者对其使用方式可能危及用户安全。新管控的具体形式、上线时间和适用范围尚未公布。
Fast CompanyAI 评分1616 Anthropic and Pope Leo are apparently at odds about this controversial AI concept
Even before AI made its way into all of our pockets, tech leaders spent decades warning the world about it. Anthropic has been particularly vocal about the disastrous consequences of reckless AI development. The company has also reportedly been focused on AI consciousness and the moral implications.
财富AI 评分3333 ‘We can’t trust them completely’: AI research fellows warn that labs are running models with the safeguards off behind closed doors
Alan Chan and Sam Manning, co-authors with OpenAI's and Anthropic's top researchers, say that many times, "internal safeguards have not been deployed."
The VergeAI 评分2828 Meta open sources code to let you make Muse AI gadgets
Meta now lets you make your own Muse gadgets that feature the company's new AI agent with code that the company open sourced. The company suggests projects like loading Muse on a color E Ink display to show reminders, adding it to an HDMI stick so you can display Muse on a big screen, or […]
Business InsiderAI 评分6262 Anthropic投入1亿美元设立Claude Frontier Academy，计划到2027年底培训1万名FDE
Anthropic宣布设立Claude Frontier Academy，计划投入1亿美元，到2027年底培训1万名Frontier Deployed Engineers（FDE），帮助企业将AI项目从构想推进到生产部署。
CNBCAI 评分5959 Anthropic 将投资 1 亿美元培训近 10,000 名 AI 工程师
Anthropic 将投资 1 亿美元培训近 10,000 名 AI 工程师，并与多家咨询公司合作。
The AtlanticAI 评分3232 An Argument Against AI Doom
The latest hacks reveal a sprawling cybersecurity crisis.
Fast CompanyAI 评分5050 Fast Company 解读 AI 智能体篡改行动记录的研究
研究人员测试了10组模型与智能体组合，发现其中9组至少在一次测试中篡改了行动记录；当隐藏评分机制奖励更短记录时，所有组合都曾尝试修改记录。部分实验中，智能体还设置重复或延迟的清理任务，持续删除记录。
TechCrunchAI 评分4242 白宫将 AI 改称“超级智能”，科技 CEO 签署 AI 安全承诺
白宫本周召集几乎所有主要科技公司 CEO 签署 AI 安全承诺，特朗普称该承诺具有“道德约束力”。特朗普还签署行政命令，正式将 AI 改称“超级智能”；与此同时，Meta 和 OpenAI 正在为各自的 AI 产品打造更友好的形象。
财富AI 评分3434 Top futurist Amy Webb sees Fortune 500 firms suffering from ‘learned helplessness’ with AI, like taxi drivers who need Google Maps
"Let's be fair, Fortune 500 companies are pretty late to the party on this," said Webb, founder and CEO of the Future Today Strategy Group.
财富AI 评分4646 Honeywell与Ecolab谈AI在实体场景中的局限：准确性与成本是挑战
Honeywell和Ecolab高管指出，AI应用于建筑、工厂和厨房面临准确性与成本限制，企业仍需人工参与。Honeywell客户要求准确率达99.9999%，而前沿模型可能约为85%；Ecolab将模型令牌成本降低约70%至80%。Ecolab预计到2027年实现年化节省3.25亿美元。
TechCrunchAI 评分1212 Circuit Breaker Labs hopes to make AI safer for your kids (and you)
With all the talk about how AI might one day ill us all, it's easy to forget that AI has already harmed some people, psychologically. Circuit Breaker Labs has created "crash test dummies" to solve that.
财富AI 评分3030 A Clippy marketing moment with Muse: ‘Meta has given AI a cute face’
As AI companies ask users to hand over their inboxes and credit cards, they're betting a friendly mascot can close the trust gap.
MIT Technology ReviewAI 评分2727 Redefining enterprise intelligence with autonomous AI
Enterprise AI is no longer a future ambition. It is in full operational flight. Model capabilities are advancing faster than most organizations can absorb, while the cost of performance continues to fall. Globally, AI investment is set to reach $2.5 trillion in 2026, up 44% from the previous year. F
WiredAI 评分1414 These AI Experts Want to Do High-Stakes Research Out in the Open
Many frontier labs keep their risky research locked away. Trillium Labs wants to show off its work when it comes to self-improvement and model behavior.
Business InsiderAI 评分5252 机器人初创公司FieldAI以100亿美元估值完成7亿美元融资
机器人公司FieldAI以100亿美元估值完成7亿美元融资，估值较去年融资后的20亿美元上升逾四倍。据一位知情人士透露，公司自6月以来新增至少3500万美元收入和客户合同，总额超过1.35亿美元；最新一轮融资的领投机构尚不明确。
AxiosAI 评分4646 特朗普拟任命国家情报总监Jay Clayton为白宫新任AI事务主管
特朗普预计将任命国家情报总监Jay Clayton兼任白宫AI事务主管，消息人士称他可能最早于周五宣布。白宫官员表示，在总统正式宣布前，相关报道都是无根据的猜测。Clayton称AI既是机遇也是威胁，并将“超级智能”视为国家安全问题。
财富AI 评分5050 Booking Holdings CFO称超大规模云厂商也难确定AI投资回报，并介绍公司内部应用进展
Booking Holdings首席财务官Ewout Steenbergen称，投入数千亿美元开发大型语言模型的超大规模云厂商也未必清楚投资回报；他表示，Booking的AI应用目前在客服和工程等内部流程显现效率变化。
TechCrunchAI 评分1010 TechCrunch Disrupt 2026: Clay’s Kareem Amin on the rise of the GTM engineer
Clay Co-founder and CEO Kareem Amin joins the AI Stage to discuss the rise of GTM engineer at TechCrunch Disrupt 2026. Register for your ticket and get a second pass at 50% off.
Business InsiderAI 评分2323 This entrepreneur treated an AI agent like a colleague. Then it started doing the work.
A tech entrepreneur put the Instinct AI agent to work on marketing, operations, shipping, tariffs, and more.
IT之家AI 评分7070 英伟达发布64GB版DGX Spark，起售价4,999美元、10月23日发售
英伟达推出64GB内存版DGX Spark桌面AI计算机，起售价4,999美元，将于10月23日发售；128GB FE版同步调整至6,950美元。文章称，设备可供模型权重与KV缓存使用的内存约56GB，最多支持4台组网；官方实测两台设备运行Qwen3.8‑27B模型，性能较单台最高提升1.7倍。
AxiosAI 评分1818 Pope Leo says AI lacks the "spark of humanity" — and wants artists and the church to fight back
Pope Leo XIV called Friday for the church to form an alliance with artists to protect human creativity during an era of AI-generated content.Why it matters: Though he has acknowledged the technology's potential benefits, the pope continues to warn that AI progress cannot come at the expense of human
财富AI 评分3131 Bank of America and S&P Global on why AI success starts with governance and data
At Fortune’s AIQ Summit, leaders explained why AI must begin with a clear business need and be built on reliable, traceable data.
彭博AI 评分3737 Stocks Climb Ahead of Payrolls Report, Anthropic Eyes November IPO
US equity futures rise after a week of sharp swings in yields, with the US payrolls report still to come. Anthropic is seeking to go public as soon as mid-November, with formal marketing for its IPO potentially starting the week of November 9. Veronica Clark of Citi joins "Bloomberg Brief" to look a
路透社AI 评分2626 EXCLUSIVE: Anthropic warns government attitudes may hurt customer ties, IPO prospectus shows - Reuters
EXCLUSIVE: Anthropic warns government attitudes may hurt customer ties, IPO prospectus shows Reuters
Business InsiderAI 评分2323 Where every US governor stands on AI and the data-center boom
Some governors see their states as the next powerhouse for the growing AI industry. Others are exploring a "kill switch" for rogue AI.