SemiAnalysis CEO Dylan Patel 在 CASPA 第35届年度大会上指出，AI 基建最大供应链瓶颈不是电力而是芯片基板，日本供应商扩产谨慎导致关键材料难以快速增加。伯恩斯坦资深分析师 Stacy Rasgon 预计半导体供应链紧张可能延续至 2028 年。Patel 还估算，建设 1 吉瓦算力基础设施的成本已从 2024 年约 400 亿美元升至如今超过 600 亿美元，两年涨幅逾 50%；同时英伟达在 CoWoS 与 HBM 资源中的占比已开始下滑，预计 2027 年进一步下降。
#行业动态#数据/训练#现象/趋势
Many of you don't understand how serious this is, so let me break it down. In computer science there is a famous kind of puzzle called 3-SAT. You are given a long list of yes-or-no switches and a set of rules about how they must be arranged, and you have to find a combination that satisfies every rule. The obvious way to solve it is to try every combination, but each new switch doubles the number of possibilities, so a puzzle with a few hundred switches would take longer than the age of the universe to brute force. For decades, the smartest people in the field found clever tricks that made this a little faster, but every trick still hit the same wall: the time required kept exploding as the puzzle grew. About 25 years ago, two computer scientists, Russell Impagliazzo and Ramamohan Paturi, working with Francis Zane, turned that wall into a formal idea. They proposed the Exponential Time Hypothesis, which says this explosion is not a failure of human cleverness but a law of nature. No algorithm, however brilliant, will ever solve 3-SAT without the time blowing up exponentially. Nobody could prove it, but almost everyone believed it, and it became one of the foundations of modern computer science. Researchers used it as a measuring stick. If you could show that solving some other problem quickly would also crack 3-SAT quickly, you had shown that problem was hopelessly hard too. Hundreds of papers, across scheduling, graphs, logic and much more, rest on this one assumption. Now an internal AI model at OpenAI claims to have broken it. It says it found an algorithm that solves these puzzles meaningfully faster than the hypothesis allows, and it has checked the proof in a computer verification system called Lean. And here is the remarkable part: the very people who proposed the hypothesis, Impagliazzo and Paturi, along with MIT's Ryan Williams, are now leading the public review of whether a machine has disproved their own idea. So what happens next? If the disproof holds, a large part of theoretical computer science has to be revisited. Every result that said "this problem cannot be solved faster, assuming ETH" loses its foundation, and researchers will have to work out which of those walls were real and which were simply where we stopped looking. It does not mean everything becomes easy. The new algorithm is still slow in absolute terms and this is not the famous P versus NP problem being solved. But it means our map of what is possible was wrong in a place we were very confident about, and it was an AI that redrew it. If the disproof fails, the hypothesis survives and the field breathes out, but we will still have watched an AI produce serious, checkable work on one of the deepest questions in the discipline, closely enough that the founders of the idea had to take it seriously. Either way, something has shifted.
#其他#现象/趋势#OpenAI
last week we closed the Training Agents series IRL at the GitHub offices in SF 4 talks, a panel and audience Q&A. almost 2h on post-training agents with open models > Learnings from enterprise RL (@t2govind Microsoft) > Environments and benchmarks in industry (@VBarres86633 Mercor) > How to train your agents with Miles (@nblintao RadixArk) > Why we built OpenEnv Arena (@xdotli BenchFlow) full recording + slides https://t.co/cIkm20wEdV
#行业动态#Agent#数据/训练
Today, we’re introducing the next chapter of Concurrence and the world’s greatest leap toward Superintelligent Care. When I was 13 I lost my mom to cancer that should’ve been caught earlier. We’re building a system that sees every patient’s future in time to change the outcome https://t.co/MOkLgjFqs8
#安全/对齐#行业动态#现象/趋势
Today we announce the launch of the Giotto family of models. We are unveiling Giotto Alto and Giotto Solo, the smartest European sovereign AI models on the market. Giotto Alto is our flagship model built for advanced reasoning capabilities, free for everyone and available on the App Store and Google Play. It is hosted on European cloud infrastructure, so your data stays in Europe. Giotto Solo is the smartest model that runs on a single GPU, so organisations can keep their AI on their own infrastructure. We are a Swiss lab with a European heart and we built this so that Europe can run advanced AI on its own terms. Use Giotto: https://t.co/KaVd59ExBd
#现象/趋势#行业动态
文章基于2020年初至2026年9月国内公开披露的212张AI相关医疗器械注册证，梳理医疗AI诊断在产品化、注册证、价格立项三道闸门上的商业化卡点。其中177张为影像诊断AI软件产品，占84%；医保局将人工智能辅助诊断列为不额外加价的扩展项，AI诊断缺乏独立收费资格。作者认为院端更倾向把AI视为设备厂商的服务增值与谈判筹码，并提出把效率换算成收入、切入医生稀缺场景、转向体检机构与药企商保付费三条短期思路。
#行业动态#政策/监管#数据/训练
2026年第一季度全行业上线微短剧约12.8万部，其中AI微短剧达12.2万部，占比超过95%。横店第一季度短剧开机量同比锐减四分之三，摄影棚空置率逼近80%，每天放出的普通群演有效通告只有700到800个。文章对比真人短剧50万到100万元的单部成本与AI短剧3000到5000元的流水线成本，认为可被标准化拆解的表演劳动最先被替代。
#现象/趋势#视频#行业动态
微信原生 AI 助手“小微”上线已 130 天，入口从官方披露的 16 处扩展到第三方实测的 28 处，覆盖聊天、搜索、公众号、视频号与 PC 端。微信通过 6 月 8 日发布的开发者接入指引，要求开发者把小程序拆成原子接口和原子组件并编写 SKILL 说明书，未接入的小程序将无法被微信 AI 推荐及调用。文章还提到开发模式需通过平台评测（服务交付、交互体验、性能质量三项 80 分以上，场景覆盖 60 分以上）才具备被调用资格，而调用排序规则尚未公开。
#现象/趋势#Agent#MCP/工具调用
OpenAI 开发者体验与社区负责人 Vaibhav Srivastav 在采访中表示，GPT-6.1 Sol 能达到 GPT-6 Astra 约 95% 的性能，成本只有后者的五分之一，他自己日常主要使用 Sol。他建议新项目或不熟悉的问题用 Astra 做对抗性审查，Ultrafast 档位面向欺诈检测、金融调查等实时场景，速度超过 300 tokens 每秒。他还介绍了常驻智能体 OpenAI dots 如何汇总 Slack、GitHub、Gmail 和日历信息并代为处理 PR，以及 Auto-review 在模型行动前按自定义策略做检查。
#行业动态#Agent#OpenAI
一个使用 OpenAI 模型的团队在 PHP 中发现两个重大漏洞，新版本 PHP 随后发布并呼吁站点运营者更新，欧洲部分政府网站因更新一度下线。文章指出，负责协调 PHP 安全问题的 PHP Foundation 只有一名全职员工，执行董事 Elizabeth Barron 称 PHP 严重缺乏资金；Network Solutions 对 121 名小企业站长的调查显示，45% 从未听说过 PHP，71% 不清楚自家网站是否使用 PHP。Cloudflare 部署了新的 Web 应用防火墙防护并推出 PHP 漏洞悬赏竞赛，奖金从 100 美元到 1 万美元。
#安全/对齐#开源生态#现象/趋势
a16z 合伙人 Ben Horowitz 与 Michael McGuiness 撰文，以 Databricks 从 Spark、Lakehouse、Lakebase 到智能体的四代技术演进为线索，讨论智能体如何改变企业。文章称 Databricks 已突破 70 亿美元年化收入、同比增长 80%，估值 1900 亿美元，每天有超过 1700 万个数据库由智能体启动。作者提出企业需要面向智能体的记录系统，把定义、权限和评估结果沉淀下来，并认为未来更多人的工作会转向指定任务、委派执行和检查结果。
#行业动态#Agent#数据/训练
The Verge 报道，Anthropic 计划在 IPO 中采用一套前所未有的治理结构，以在上市同时保住其 AI 安全使命。公司作为公共利益公司运营，董事会七个席位中四个由长期利益信托任免，创始人通过 Founder LLC 掌握 50.1% 投票权，普通股东几乎无法左右董事会。多位法学教授认为这在法律上让 Anthropic 免受股东逐利压力，但也指出创始人自身持股、融资需求与创始人离职后信托日落条款都可能削弱这些保护。
#行业动态#安全/对齐#Anthropic
Latent Space 播客邀请 Anthropic 的 Katelyn Lesse 和 Angela Jiang 讲解 Claude 平台背后的工程，包括 Opus 5.5 的发布、把模型“大脑”与智能体“手”分离、API 扩容和构建可靠的托管智能体。节目还涉及用 Rust 重写 Messages API、凭据与沙箱隔离、长时运行智能体的 harness 设计，以及安全分类器与模型路由。
#行业动态#Agent#部署/工程
Latent Space 播客访谈 Anthropic 工程负责人 Katelyn Lesse 与产品负责人 Angela Jiang，围绕 Opus 5.5 的发布展开，讨论模型写作风格调整、按每美元智能定价以及为防蒸馏而要求跨轮保留 thinking blocks 等 API 变化。访谈还介绍了 Claude Managed Agents 把会话与 harness 同执行代码的沙箱分离的思路，凭据通过代理单独挂载，沙箱故障不会带走整个会话；Dreaming 则被实现为一个托管智能体检查另一个智能体的历史并写回记忆。团队表示随着模型发布的安全工作变重，近期精力更多转向安全，技能、记忆等功能的推进相对放缓。
#行业动态#Agent#部署/工程
凤凰科技旗下数读社分析称，Anthropic 招股书披露约 6370 亿美元算力承诺，其中博通、SpaceX、谷歌、亚马逊均达千亿美元，而 7 月 ARR 仅 650 亿美元。文章认为三季度营收约 165 亿美元、环比增速降至约 43%，投资者对 IPO 估值预期从 2 万亿美元降至约 1.5 万亿美元，同时 DeepSeek V4.1 Flash、Kimi K3.1 Pro 等中国模型在性能和价格上持续施压。
#行业动态#OpenAI#Anthropic
OpenAI 公布了对三个数学难题的解答，其中一项是对计算机科学长期假设指数时间假说（ETH）的否定，Anthropic 随后也发布了自己的 ETH 反驳。MIT 教授 Ryan Williams 与 UC San Diego 的 Russell Impagliazzo、Ramamohan Paturi 受 OpenAI 邀请协助解读，并发布公开信征集反馈、讨论 AI 贡献的规范。Williams 表示，若该结论成立，过去十年大量依赖 ETH 的论文需要重新审视，他还曾向 OpenAI 科学负责人 Dan Roberts 确认是否影响现有密码系统，对方称已检查且认为不会导致破解。
#论文/研究#推理#OpenAI
TechCrunch 分析指出，Anthropic 在 IPO 临近时通过公益公司身份、长期利益信托（LTBT）和创始人特别股三层结构，把董事会多数席位的选举权交给信托，创始人七人合计持有 50.1% 股东投票权并可选出三名董事，普通股东对包括董事人选在内的多数重大决策几乎没有话语权。董事会还可将 LTBT 提名或罢免特定董事的决定交由员工投票，创始人控制权在 IPO 十年后或创始人少于三人时终止。
#行业动态#Anthropic#现象/趋势
路透社报道字节跳动、阿里巴巴和腾讯对英伟达RTX PRO 5500的订单需求合计超过300万张，按每张约1.14万美元估算规模超350亿美元，主要用于AI推理。文章同时提到Kimi K3.1 Pro上线后额度消耗过快、月之暗面致歉并补偿25%额度，以及Anthropic已提交IPO申请。作者指出350亿美元是采购意向而非确定收入，按五年摊销每年约70亿美元设备成本，还需叠加服务器、电力和运维支出。
#行业动态#数据/训练#部署/工程
36氪梳理发现，2025年2月至2026年9月的19个月间，破产、解散、清算和业务线关停的机器人公司超过20家，其中包括百亿估值独角兽深兰机器人和曾获美团、字节跳动投资的盈合机器人。文章指出，深兰与LG的订单并未大规模执行，盈合机器人的百亿基地未见投产，无境动力近10亿元订单中近5亿元属于框架协议；宇树招股书显示2025年前9个月人形机器人收入的73.6%来自科研教育，IDC数据显示2025年人形机器人出货中商演、科研、数据采集占78.4%。有AI投资人称宇树上市后没有再向投委会提交具身智能项目。
#行业动态#具身智能#现象/趋势
字节、腾讯、快手在近一个月密集推进互动内容布局：抖音上线互动空间广场和互动剧专区，将《归墟》《零号档案》等热门AI短剧改编为互动版本，并于10月19日发布互动剧和互动影游合作征集；腾讯的空弧TDream小程序开启删档测试，造化工坊9月17日更新无限画布工作流等专业功能；快手则上线“快手互动内容世界”小程序，集中展示互动剧、模拟养成和轻量小游戏。文章指出，入口和创作工具已逐步完善，但互动剧多线剧情成本高于普通短剧、分账机制尚未明朗，用户是否愿意在关键节点停下来选择仍待市场检验。
#现象/趋势#视频#行业动态