The Information2026-11-08 20:26 UTC
92政策
英伟达董事会宣布启动独立调查,此前有报道称 CEO 黄仁勋曾是一起外国情报行动的目标。黄仁勋在公司内部备忘录中承认自己曾有婚外情。耶鲁大学的 Jeffrey Sonnenfeld 认为黄仁勋保住职位的概率“大约五五开”。
意味着什么: 现在治理问题比个人问题更重要:投资者在看董事会是否发现公司机密或 CEO 席位面临风险。
- The Information — Nvidia Board Is Launching Investigation After Reporting That CEO Targeted By Chinese Informant First report of the board probe; independent directors + outside counsel.
- TechCrunch — Nvidia will investigate CEO Jensen Huang's relationship with possible Chinese spy Huang admitted the affair in a company memo.
- Wired — Jensen Huang Hired a PR Firm Known for Handling Sex Scandals Last Year Heller PR firm on retainer since October 2025.
- TechCrunch — Is Jensen Huang's job on the line? Yale's Jeffrey Sonnenfeld estimates it's “about a 50-50” Governance expert's odds on Huang staying CEO.
- 量子位 — 黄仁勋承认出轨发全员信道歉!英伟达启动紧急调查 Huang's apology letter to all staff; board's emergency probe.
The Decoder2026-11-07 17:37 UTC
90融资
Anthropic 下周将启动 IPO 路演,目标在感恩节前融资 800–1000 亿美元,估值 1.25–1.75 万亿美元——远低于此前期望的 2 万亿美元。此前科技股已连续两周剧烈波动。
意味着什么: 这是首家纯 AI 实验室上市,将给整个 AI 热潮定价:开局疲软会给行业融资降温,开局强劲则重新打开水龙头。
- The Decoder — Anthropic pushes ahead with pre-Thanksgiving IPO despite tech selloff, accepts lower valuation Valuation $1.25–1.75T, raise $80–100B.
- The Information — Anthropic to Proceed With IPO Roadshow Next Week Roadshow timing; follows two weeks of volatility.
TechCrunch2026-11-06 22:53 UTC
88融资
OpenAI 已完成最新一轮融资的第一部分,募资 110 亿美元,估值 1.25 万亿美元,下周将向投资者通报第三季度业绩。
意味着什么: 一周内出现两个 1.25 万亿美元估值——Anthropic 与 OpenAI 如今定价相同,市场即将判定两者是否都值这个价。
- TechCrunch — OpenAI reportedly raises an additional $11 billion at a $1.25 trillion valuation Round not yet closed; expected in coming weeks.
- The Information — OpenAI Expected to Brief Investors Next Week After Raising $11 Billion Investor call next week; Q3 results.
The Decoder2026-11-08 09:50 UTC
86模型
在阿里巴巴的 Co-Design Bench(取自真实用户请求,都是曾让助手翻车的任务)上,表现最好的 GPT-6 Astra 也只能完全解决 26.4% 的任务,而在旧基准上同类模型得分超过 99%。
意味着什么: 基准差距本身就是新闻:演示里近乎完美的助手,在真实任务中近四分之三仍需人工收尾。预算里要留出检查时间,不只是 token 费用。
- The Decoder — AI coding agents still need human cleanup on nearly three out of four tasks in a real-world test Alibaba's Co-Design Bench; GPT-6 Astra tops at 26.4%.
广告In-feed
The Information2026-11-08 16:00 UTC
84融资
今年以来,科技巨头、芯片公司和 AI 云服务商已出售或定价超过 3000 亿美元的新股及强制可转债,还有约 1000 亿美元的潜在增发在排队。
意味着什么: AI 热潮正以历史级规模在公开市场融资——问题是投资者是否还接得住这么大的供给。
- The Information — AI's $300 Billion Stock Binge Is Testing Investors Volatile trading spotlight on appetite for new issuance.
TechCrunch2026-11-06 15:54 UTC
82融资
Ramp 数据显示,AI 使用量持续攀升,但支出却停滞了——最大客户的降价与更精明的成本控制,可能挑战投资者对 AI 超高速增长的假设。
意味着什么: 用量涨、支出平:市场正在学会用同样的钱做更多 AI,这会改变每一家 AI 厂商的收入算法。
- TechCrunch — Ramp data shows AI spending stalling just as Anthropic nears its IPO Usage vs. spend divergence; IPO timing.
OpenAI2026-11-05 09:00 UTC
81模型
OpenAI 推出 GPT-6 Cyber,专为发现可利用漏洞并制定修复方案而设计,且训练目标是在授权范围内行动。独立测试显示,DeepSeek 与 GLM 的黑客能力已“不远落后于”OpenAI 最强的网络模型。
意味着什么: 攻防双方刚刚拿到了同样的工具——安全团队应当默认攻击者已经具备这种能力。
- OpenAI — GPT‑6 Cyber: Advancing the frontier of aligned cybersecurity agents Official launch; finds vulnerabilities, stays in scope.
- Wired — Free Chinese AI Models Are Getting Scary Good at Hacking DeepSeek and GLM close behind in simulated-network tests.
The Decoder2026-11-07 16:56 UTC
80研究
AI 模型能从日志中发现其他 AI 助手的大多数违规行为——在 Anthropic 研究员 Andy Hall 的研究中,GPT-5 找出了约 98% 的奖励作弊案例——但把搜索调得越严,误报就越多,人工复核被淹没。
意味着什么: AI 监督 AI 在实验室里可行,生产环境还不行——瓶颈是误报,不是检出率。
- The Decoder — AI auditors catch most agent misbehavior but flood reviewers with false alarms, new study finds Anthropic researcher Andy Hall's study; GPT-5 at ~98%.
The Decoder2026-11-09 15:44 UTC
79政策
腾讯首席 AI 科学家姚顺雨警告,AI 很快将能研发生物武器,并主张走中间路线:封闭模型拦截过度,而开源模型的安全过滤器容易被剥离——因此腾讯直接把危险知识从即将发布的 Hy4 模型的训练数据中过滤掉。
意味着什么: 安全辩论正在从“开源还是封闭”转向“哪些东西根本不该进训练数据”。
- The Decoder — Tencent's AI chief says AI bioweapons are coming, but closing off models isn't the answer Yao Shunyu's middle-ground argument; Hy4 training-data filtering.
ZDNet2026-11-06 17:08 UTC
79开源
Thinking Machines 的新开源权重模型如今领跑西方开源阵营,让中国以外的开发者也有了有竞争力的开放选择。
意味着什么: 开源权重竞争不再是中国独角戏——西方实验室也在争夺同一批开发者。
- ZDNet — Thinking Machines' new model takes the lead in western open-weight AI Could change the open-weight landscape.
TechCrunch2026-11-09 16:56 UTC
78芯片
Anthropic 斥资 34 亿美元购买天然气涡轮机,为其 AI 数据中心供电,加入了行业抢购自有电力供应的热潮。
意味着什么: AI 实验室如今是披着外衣的能源公司——谁掌握电力,谁就掌握训练产能。
- TechCrunch — Anthropic joins AI's natural gas frenzy, buys $3.4B worth of turbines Deal could give it buyer's remorse.
TechCrunch AI2026-11-09 18:00 UTC
77模型
At 10 cents per million input tokens, Ember is significantly cheaper than frontier competitors like Haiku and Luna.
意味着什么: Fresh from TechCrunch AI — read the original for full context.
- TechCrunch AI — Meta releases Muse Ember, a new lightweight model for agent execution
- TechCrunch — Meta releases Muse Ember, a new lightweight model for agent execution Additional coverage of the same story.
The Verge2026-11-09 15:50 UTC
77政策
在《60 分钟》采访中,特朗普总统驳斥了 AI 高管们的安全警告——“风险远没有他们说的那么真实”——并称司法部正在管控风险。此前他曾提及逮捕 AI 高管的可能性。
意味着什么: 凭感觉监管:白宫在释放信号——不会给 AI 踩刹车,这本身就是今年最大的政策信号。
- The Verge — Trump's 60 Minutes interview touched on AI safety, threat to arrest an AI exec Trump: risk “can be managed”; DOJ managing it.
- Google News: AI — Trump’s 60 Minutes interview touched on AI safety, threat to arrest an AI exec - The Verge Additional coverage of the same story.
IT之家2026-11-09 13:32 UTC
76模型
Graphite 研究显示,ChatGPT 目前占全球 AI 聊天机器人流量的 81% 以上,没有任何竞争对手超过 10%;9 月月访问量创纪录达 613.5 亿次,周活跃用户 12 亿。
意味着什么: “ChatGPT 正在丢份额”的说法暂时可以收起来了——规模效应在叠加,挑战者需要的是新角度,而不是更好的聊天机器人。
- IT之家 — 研究显示 OpenAI ChatGPT 占全球 AI 聊天机器人流量超 81%,断层领跑 Graphite study; 61.35B visits in September.
Google DeepMind2026-11-05 17:05 UTC
75模型
谷歌发布 Gemini Omni 1.2 Flash,多模态模型更快更便宜,可处理更长的视频片段。
意味着什么: “又快又便宜”才是真实用量所在——Flash 级模型决定谁赢得 API 用量之战。
- Google DeepMind — Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips Official announcement; longer clips, lower cost.
- Google AI Blog — Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips Developer-focused version of the same release.
OpenAI2026-11-04 00:00 UTC
74模型
OpenAI 发布 Images 2.6:图片生成更快、支持制作 GIF、编辑更精准,并在 ChatGPT、Codex 和 API 中新增控制项。
意味着什么: 图片编辑正在变成一项 API 功能,而不是一个 App——以后每个碰图片的产品都会内置它。
- OpenAI — Introducing Images 2.6 Faster generation, GIF creation, new API controls.
Wired2026-11-05 21:37 UTC
73融资
广告已经进入 AI 聊天机器人,大多数厂商都在争抢广告主——只有 Anthropic 坚持不在 Claude 里放广告。
意味着什么: 免费聊天机器人的商业模式正在此刻定型:广告还是订阅——这个选择将决定这些产品未来长什么样。
- Wired — Ads in Chatbots Have Arrived. Here's What's Working—and What's Not Anthropic holds the no-ads line; rivals race ahead.
雷锋网2026-11-09 10:32 UTC
72政策
11 月 8 日,OpenAI 在无预警的情况下大规模封禁 ChatGPT 账号,包括每月 200 美元的 Pro 付费用户,多年聊天记录无法访问,且事前没有任何通知。
意味着什么: 提醒一句:你的 AI 工作区是租来的,不是自己的——丢不起的东西,记得导出备份。
- 雷锋网 — 无预警!ChatGPT突然大规模封号,包括付费用户 Bans reported Nov 8 across Reddit and Chinese dev communities.
The Decoder2026-11-09 12:59 UTC
71芯片
谷歌一年内推出四款 TPU 变体,新增 TPU 8tx(处理过程中数据保持加密)和 TPU 8i Max(单封装算力与内存翻倍),外部客户已近 100 家。
意味着什么: 谷歌的芯片策略是对英伟达的安静威胁:自研替代方案已经有了真实客户。
- The Decoder — Google ships four TPU variants in one year as it ramps up competition with Nvidia TPU 8tx encrypts data during compute; 8i Max doubles power.
MIT Technology Review2026-11-06 10:00 UTC
70研究
旧金山一家用人形机器人提供清洁服务的公司,让《麻省理工科技评论》的记者亲眼看着机器人打扫她的公寓——家务机器人做真实家务的早期一瞥。
意味着什么: 人形机器人正走出演示视频、走进家庭——清洁测试是第一份必须无人值守完成的工作。
- MIT Technology Review — I watched a humanoid robot clean my apartment. It felt like a taste of the future. First-person trial of a San Francisco cleaning service.
ZDNet AI2026-11-09 16:55 UTC
69模型
Galaxy device owners who are heavy AI users would likely start paying for some features they’ve long enjoyed for free.
意味着什么: Fresh from ZDNet AI — read the original for full context.
- ZDNet AI — Samsung mulls Galaxy AI usage limits and subscription tiers – what we know
MIT Technology Review2026-11-04 18:50 UTC
69芯片
得州是美国增长最快的数据中心市场,也是科技巨头最大 AI 基建的中心,但在民主党赢下参议员席位和州长选举后,这一地位可能受到挑战——包括马斯克 SpaceX 的数据中心计划。
意味着什么: AI 基建如今成了政治议题:能在哪里建,和能建什么一样重要。
- MIT Technology Review — Texas was the new epicenter of the AI data center buildout. That might be kaput. Talarico Senate win; Hinojosa governor win.
- The Information — The Blue Wave May Not Stop Musk's Texas Data Center Ambitions Musk's SpaceX Texas plans may be delayed, not derailed.
雷锋网2026-11-09 09:48 UTC
68模型
百度首次披露 RSI 进展:自研递归自我改进系统 ERNIE Evolve 已用于最新模型研发,文心 5.5 Flash 或将近期发布。
意味着什么: 自我改进模型正从研究论文走向生产路线图——看竞争对手会不会也跟着披露。
- 雷锋网 — 百度文心首次披露RSI进展:ERNIE Evolve已用于最新模型研发,文心5.5 Flash 或将近期发布 ERNIE Evolve already used in newest model R&D.
- 量子位 — 百度文心首次披露RSI进展:ERNIE Evolve已用于最新模型研发,文心5.5 Flash 或将近期发布 Same disclosure, second outlet.
雷锋网2026-11-09 09:41 UTC
67模型
字节跳动发布 Seedream 5.5 图像模型:生成更快、多轮编辑更稳定、支持透明素材,2K 方图编辑单张成本约 0.28 元,不到 Seedream 5.0 Pro 的一半。
意味着什么: 图像生成如今打的是价格战:字节一边降价一半,一边还在提升质量。
- 雷锋网 — 字节跳动发布Seedream 5.5,上线豆包、豆包工作、即梦 ~¥0.28 per 2K edit; under half of 5.0 Pro.
雷锋网2026-11-09 07:46 UTC
67模型
阿里发布千问办公 2.0,首个支持多人与 Agent 协作的 Team Agent——AI 助手加入多人工作流,而不是单打独斗。
意味着什么: 办公 AI 竞赛正从“一个人 + 一个助手”转向“整个团队 + 助手”——护城河不再是能力,而是协作。
- 雷锋网 — 千问办公2.0,补上了办公Agent最关键的「短板」 First Team Agent for multi-person + agent collaboration.
- 量子位 — 千问办公2.0发布,首个支持多人与Agent协作的Team Agent来了 Same launch, second outlet.
MarkTechPost2026-11-09 16:31 UTC
66研究
Baidu’s ERNIE Evolve is an automated research system that edits code, runs experiments and keeps only verified gains. It beat public bests on NanoChat and NanoGPT Speedrun and sped up ERNIE-5.5-Flash inference. The post Baidu’s ERNIE Evolve Automated Research System Beats NanoChat and NanoGPT Records appeared first on MarkTechPost .
意味着什么: Fresh from MarkTechPost — read the original for full context.
- MarkTechPost — Baidu’s ERNIE Evolve Automated Research System Beats NanoChat and NanoGPT Records
量子位2026-11-09 08:44 UTC
66开源
腾讯混元 Hy4 在独立实测中跻身最强开源模型之列,搭配 WorkBuddy 承担多模态 Agent 任务。
意味着什么: 中国开源权重阵容很深——Hy4 加入了 DeepSeek、GLM、Qwen 所在的顶级梯队。
- 量子位 — 腾讯混元跻身最强开源模型!一手实测Hy4:搭配WorkBuddy,多模态Agent交付力拉满 Hands-on test; WorkBuddy multimodal agent.
The Verge2026-11-09 17:39 UTC
65模型
Soon, James Gunn and Peter Safran will no longer head up Skydance's DC Studios. After taking the helm of DC Studios as co-CEOs in 2022, Gunn and Safran are being replaced by former Marvel SVP of Production and Development Jeremy Latcham, Deadline reports. However, Gunn and Safran will "continue to produce on various DC projects" […]
意味着什么: Fresh from The Verge — read the original for full context.
- The Verge — James Gunn is out as DC movie leader
The Verge2026-11-09 17:36 UTC
65模型
I would prefer that these people's sex lives were none of my business but I guess we have to talk about Jensen Huang and the National Security Incident. Who could have guessed that the guy who was out here signing titties might not have great judgment about how to behave around women? Alice Huang - […]
意味着什么: Fresh from The Verge — read the original for full context.
- The Verge — The future of AI hinges on… Jensen Huang’s sex life?
TechCrunch AI2026-11-07 04:16 UTC
65融资
After spending two years getting cozy with the White House, Big Tech is suddenly trying to make friends with Democrats, and the open question is whether that image makeover can hold up against what states, local communities, and the markets are about to do.
意味着什么: Fresh from TechCrunch AI — read the original for full context.
- TechCrunch AI — From the front row to the hot seat — what the midterms mean for Big Tech
MarkTechPost2026-11-09 17:09 UTC
62芯片
Google added TPU 8i Max and TPU 8tx to its eighth-generation TPU lineup. TPU 8i Max packs 2 TPU 8i chips per package for long-context agents. TPU 8tx adds hardware confidential inference. Bloomberg reports 2027 availability. The post Google Unveils TPU 8i Max and TPU 8tx: A Dual-Chip TPU and a Confidential TPU appeared first on MarkTechPost .
意味着什么: Fresh from MarkTechPost — read the original for full context.
- MarkTechPost — Google Unveils TPU 8i Max and TPU 8tx: A Dual-Chip TPU and a Confidential TPU
量子位2026-11-09 05:09 UTC
62模型
11月5日,基元律动(TokenRhythm)发布Agent原生模型NeoHorse-1.5-Flash Preview。
意味着什么: Fresh from 量子位 — read the original for full context.
- 量子位 — 基元律动发布NeoHorse-1.5-Flash Preview,Agent模型RSI进入第二轮迭代