每日精选
4 条核心资讯微软公布Maia 300,规模部署从2027年开始
原标题:Microsoft to expand Azure Maia system to multi-gigawatt deployments
微软公布第三代Azure Maia系统,面向长上下文、对延迟敏感的对话,以及持续多步骤的智能体任务。必须区分正在运行与未来计划:Maia 200已服务包括Microsoft Copilot在内的生产负载;Maia 300计划2027年开始生产部署,多吉瓦规模扩张也从2027年启动。
公司预计Maia 300相较领先GPU系统,每兆瓦产生的token增加40%;另称96个加速器组成的配置,相比Maia 200可达到最高7倍机架级token吞吐。这是两个不同的比较对象,也是厂商预期与报告,不能写成今天全部服务已提速。公告还把减少数据搬运、内存带宽和互联当作系统设计重点。值得看的是在相同负载和延迟条件下,实际交付的效率,而不是单看芯片峰值。
原文 · 2026-10-12 · 官方公告;核验范围:浏览器核读公告全文及生产部署时间,未独立实测。
Meta发布Muse Wonder系列,长期记忆与代码任务进入竞争
原标题:Meta releases its first frontier model, Muse Wonder
据TechCrunch原报道,Meta发布Muse Wonder和Muse Wonder Full。前者可通过Meta Model API调用;Wonder Full则面向Muse Code订阅用户,报道明确它不通过API提供。两者也将用于Muse智能体。
报道转述Meta的测试:新系列强调复杂任务和长期信息保留,尤其突出所谓active memory。这里的比较成绩是公司报告,不是独立评测。本期不复述未经原始模型卡核对的排名或价格。对用户而言,关键问题是记住哪些信息、怎样修正过期内容,以及订阅入口与API入口是否提供相同模型;不能从“更会记忆”直接推出“更值得信任”。
原文 · 2026-10-12 10:56 PDT · 原创报道;核验范围:报道全文;模型卡和模型未独立评测,官方博客定位未成功。
Google梳理医疗研究:从预测结构到患者获益仍有多道关
原标题:Applying technology to understand and treat disease
Google与研究伙伴梳理AI在疾病研究、实验室分析和医疗支持中的应用。公告举例:研究者用AlphaFold筛选恰加斯病相关化合物,找到可继续研究的已获批药物候选;与斯坦福合作的研究使用DeepVariant及DeepConsensus辅助紧急遗传病诊断;与英国伙伴的乳腺筛查研究评估了12.5万名女性的资料。
这些例子不是所有工具今天首次发布,也不等于自动诊断或药物已完成新的适应症审批。本期采用“跨环节研究进展”的口径,未把公告汇总写成统一的新临床结论。结构预测、功能实验、试验与实际医疗结果需要分别看;相关机构报告不替代患者的专业诊疗。
原文 · 2026-10-12 · 官方研究综述;核验范围:正文应用与研究条件段落,未逐篇核对被引用研究全文,不复述效果百分比。
欧盟科学小组提出事故调查建议,尚不是新增法律要求
原标题:Special meeting of the AI Scientific Panel
欧盟委员会10月12日发布说明,回顾10月9日科学小组特别会议。AI Office已经索取并收到严重事故报告,并发出安全实践信息要求;委员会将汇总小组和AI Office的问题,继续调查近期失控事件。
主席与副主席建议:在开发和测试部署期间治理内部模型风险,增强事故发现、评估与调查能力,建立国际信息共享及危机预防协作,并加强对降低模型可监控性的训练技术与架构的审视。文章也明确仍有未知事项。这里区分已采取的信息索取与尚属建议的措施,不把专家建议写成新法规已经落地。
原文 · 公布2026-10-12;会议2026-10-09 · 官方说明;核验范围:全文,未读取非公开事故报告。
原标题:OCP Global Summit: Amid Ecosystem Adoption, NVIDIA Opens Vera Rubin Architecture to Help Scale AI Factories Worldwide
NVIDIA在Open Compute Project峰会介绍Vera Rubin机架与计算托盘设计贡献,以及MGX开放架构、供电、液冷和网络标准协作。同场与Cisco公布新的USFP光模块方案,重点是单位空间和功耗中的网络容量。方案、规范与计划发布不等于所有设备已量产交付;厂商效率数字也不等于整座数据中心的电费降幅。
原文 · 2026-10-12 · 官方公告;核验范围:架构与协作相关正文,并对照Cisco光模块联合公告,未独立实测。
来源:NVIDIA官方公告 · 2026-10-12 · 官方公告;核验范围:架构与协作相关正文,并对照Cisco光模块联合公告,未独立实测。 · 架构与协作相关正文,并对照Cisco光模块联合公告,未独立实测。
原标题:AI Company Lets Cops Pinpoint Specific Houses Based on a Single Interior Photo
404 Media记者查看Graylark的Raven演示,并将结果与Airbnb、Vrbo及Zillow房源信息对照:工具试图利用房间中的视觉特征定位具体物业,不只搜索同一张照片。报道同时指出演示是否代表普遍速度与准确度仍不清楚,警方此前称使用有限、探索性,未形成重大侦查突破。Airbnb表示没有与该公司的关系,并有防范自动采集的措施。本期据原报道转述,不把营销演示当成任何照片均可定位的证据。
原文 · 2026-10-12 · 原创调查;核验范围:全文、记者对演示的核对及平台与警方回应,未测试工具。
来源:404 Media原创调查 · 2026-10-12 · 原创调查;核验范围:全文、记者对演示的核对及平台与警方回应,未测试工具。 · 全文、记者对演示的核对及平台与警方回应,未测试工具。
行业洞见
深度分析原标题:Satya Nadella and Jensen Huang discuss Hybrid Intelligence and the future of PC
人物背景:Satya Nadella是微软董事长兼首席执行官,负责云端与个人电脑平台战略;这段对谈讨论智能体从云端进入桌面后的基础设施条件。Jensen Huang是NVIDIA创始人兼首席执行官,其计算与硬件生态视角与纳德拉的操作系统视角相互补充。身份及现场语境依据两家公司官方发布方报道核对。
节目背景:这是Windows官方频道发布的产品活动现场炉边对谈,不是独立媒体播客。活动在2026年10月7日举行,视频10月9日上传,具有厂商产品宣传语境。
纳德拉认为隔离不能只做一层:之后还需要智能体可追踪的身份、行为记录、策略治理与成本记录。讨论代理消费时,他又把委托范围与信任放在技术协议之外;黄仁勋强调先有最小权限、凭据和监控,再谈更自主的行动。本期不依据自动字幕扩写未核清的产品缩写。
为什么今天听:新芯片与新模型正在扩大可执行任务的范围,这段历史对谈提醒我们:运行能力增加,不代表授权、观察与追责会自动补齐。这是本期的关联分析,嘉宾并非在回应今天的Maia或Muse发布。
原文 · 原活动2026-10-07;上传2026-10-09 · Windows官方视频;核验范围:自动字幕隔离与身份、委托权限相关段落,并核读NVIDIA同场报道;未独立校听原片。
来源:Windows官方视频及发布方报道 · 原活动2026-10-07;上传2026-10-09 · Windows官方视频;核验范围:自动字幕隔离与身份、委托权限相关段落,并核读NVIDIA同场报道;未独立校听原片。 · 自动字幕隔离与身份、委托权限相关段落,并核读NVIDIA同场报道;未独立校听原片。
原标题:Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub
人物背景:Pushmeet Kohli负责Google DeepMind科学方向的研究工作,参与AlphaFold相关研究,其身份也在本期Google公告中得到核对。他在节目中回顾安全、计算机视觉与科学模型的研究经历,强调先确定科学问题再选择数据与模型。Sal Candido来自Biohub,在讨论中介绍此前在Google研究蛋白质语言模型的经历;本期只转述Kohli的相关观点。
节目背景:Latent Space是面向AI工程的播客与通讯,这一期发表的是Biohub活动上的生物AI讨论,由Brandon Anderson主持。帖子2026年10月10日发布,活动举办日未单独核实;人物和栏目背景以发布方页面为依据。
Kohli指出蛋白质会随环境变化,复现存入数据库的结构并不等于解决所有动力学问题。他区分内部机制解释与行为刻画:使用者至少要知道模型能做什么、不能做什么,置信度是否校准。谈药物研发,他强调AI已在多个环节使用,更大的加速需要攻克具体生物学问题,不能用一个总倍数跳过各环节。
为什么今天听:Google今天的医疗研究综述跨越结构预测、实验室和患者支持。这段材料帮助把不同成果放回各自适用范围,避免将一个预测任务的进展写成所有临床问题已经解决。这是本期关联,不是嘉宾回应今天的公告。
原文 · 发布2026-10-10;论坛日期未独立核实 · 发布方文字稿;核验范围:15:28—17:39、23:53—29:43,并核对人物及节目介绍;未独立校听录音。
来源:Latent Space发布方文字稿 · 发布2026-10-10;论坛日期未独立核实 · 发布方文字稿;核验范围:15:28—17:39、23:53—29:43,并核对人物及节目介绍;未独立校听录音。 · 15:28—17:39、23:53—29:43,并核对人物及节目介绍;未独立校听录音。
论文速递
1 篇论文原标题:TokenRouter: Efficient Serving System for Token-Level LLM Routing
这篇10月8日预印本是今天发现后的主题补读,不是今日新发表。逐token路由让同一个回答在不同模型间切换,但会打破单模型系统对同步批次的假设,出现快模型等慢模型、切换请求等待下个批次的问题。作者以请求为中心描述路由,以模型为中心异步执行,并采用延迟组批。
实验在8张A100-80G服务器上,比较五种路由算法及短推理、长推理和多轮智能体轨迹等负载。已核图1中并发8的例子:相对R2R官方实现,总吞吐为2.76倍,其中普通工程优化已贡献1.71倍;这说明选择基线会影响增益解读。数字是指定条件下的输出token吞吐,不等于任务正确率提高或所有用户等待时间同比缩短。未复现,未把预印本写成同行评审结论。
原文 · v1:2026-10-08;当期主题补读 · 作者预印本;核验范围:摘要、系统设计、实验设置及图1,未复现。
开源解读
1 个仓库原标题:Context Mode — The other half of the context problem.
Context Mode的维护者README介绍MCP服务:先在沙箱处理工具原始输出,再索引和检索所需片段;另用SQLite记录会话事件,帮助压缩上下文之后继续任务。本期把它作为当前榜单发现的工程项目介绍,不宣称今天首次发布,也不将维护者的压缩数字当作本刊实测。
它解决的是信息进入上下文的方式,不能保证留下的记录完整、查询正确或最后任务成功。仓库标示Elastic License 2.0(ELv2),不能仅因为可见源码就写成MIT或无限制使用。安装与权限仍应由使用者按自身环境判断,本期未安装或部署。
原文 · 2026-10-13榜单发现;非首次发布声明 · 维护者README;核验范围:问题、实现思路及许可证标示,未部署。
采集说明 · 采集不完整 · 19/45 个来源完成检查
新闻来源以本轮config/sources.json为准。上游45源、05:00截止收据已保存并复用,独立补查05:00—06:00增量。非RSS动态入口及部分日期证据仍存在缺口,完整范围以当期收据为准,标记采集不完整;不能把可达性或未发现新条目当成完整扫描。话语场覆盖单独统计,引用历史节目不改变其原始日期。
检查尚未完成的来源:下列状态表示检查缺口,不能据此认定近期无更新;“已完整检查且无新增”另行记录。
- Anthropic News:已检查部分内容,覆盖或日期待核。打开 https://www.anthropic.com/news 。FeaturedGrid 卡片与 PublicationList 列表配对标题、链接、日期。可见日期 ['Oct 1, 2026', 'Oct 2, 2026', 'Oct 6, 2026', 'Oct 7, 2026', 'Oct 8, 2026', 'Sep 1, 2026', 'Sep 17, 2026', 'Sep 18, 2026', 'Sep 22, 2026', 'Sep 23, 2026', 'Sep 28, 2026']。窗口日历日内 0 条。未读正文。 仓库脚本报“最新 2026-10-10”,核对页面数据,该日期只出现在 featuredGrid 的 _updatedAt(CMS 修改时间),不是文章发布日期,不计候选。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- Claude Blog:已检查部分内容,覆盖或日期待核。配置 https://claude.com/blog 最终打开 https://claude.com/resources/articles 。资源卡日期 ['Oct 5, 2026', 'Oct 6, 2026', 'Oct 8, 2026'] 与内嵌 blogPost 数据配对。窗口日历日内 0 条(按链接去重);另有 0 条外链到 anthropic.com(),已记在 anthropic,不重复。未读正文。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- xAI:已检查部分内容,覆盖或日期待核。静态请求 https://x.ai/news HTTP 200(配置 note 称脚本 403,本次未被拦)。新闻链接去重约 85 条,月份日期最新 Sep 28, 2026,10 月日期 0 条,窗口日历日内 0。headless Chrome 渲染被 Cloudflare 'Attention Required' 拦截,未得到渲染版。未读正文。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- Microsoft AI:已检查部分内容,覆盖或日期待核。配置 URL https://blogs.microsoft.com/ai/ 跳到 https://news.microsoft.com/source/topics/ai/ ,静态与 headless 渲染都没有列表日期。配置路径 feed https://blogs.microsoft.com/ai/feed/ 返回 HTTP 410 Gone。AI 话题 feed 10 条,最新 2026-10-12T13:14:40+00:00,窗口内 1。另查 Source 全站 feed 10 条(2026-10-08T12:36:32+00:00 至 2026-10-12T17:47:28+00:00),窗口内 AI 相关或 Surface 活动 2 条保留为候选。替代 feed 不等于配置入口。 复核:首轮按标题关键词把 Maia 300(AI 话题 feed 条目)、同日 LinkedIn 自研芯片帖与“models as insider risks”帖误记为非 AI,已改回候选,合计 5 条;Source 全站 feed 里 Xbox、Michelangelo 字体等条目早于窗口或不涉 AI。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- Qwen:已检查部分内容,覆盖或日期待核。配置 URL https://qwenlm.github.io/ 仍是旧 GitHub 博客,页面日期最新 2025-09-23,不能据此判断无更新。另查现行入口 https://qwen.ai/research:headless Chrome 渲染,可见日期 ['2026/09/03', '2026/09/18', '2026/09/20'],窗口日历日内 0 个;只看了渲染首屏,未点分页/筛选。page_config 接口 {'qwen_research_research-list': (60, '2025-12-23T05:08:30.000Z'), 'qwen_home_feature-list': (6, '2025-03-28T00:00:04.000Z'), 'qwen_home_latest-research-list': (19, '2025-09-24T04:00:00.000Z'), 'qwen_research_latest-advancements-list': (19, '2025-09-24T04:00:00.000Z'), 'qwen_news_news-list': (17, '2025-04-28T20:00:00.000Z')};/api/v2/article/retrieval 返回空数组。补充(不在配置):Hugging Face Qwen 组织按 createdAt 列模型,窗口内新建 0 个(无),作为线索候选;GitHub QwenLM 组织最近 push 仅作参考,未逐仓库查 release。未改 config。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- 通义实验室:已检查部分内容,覆盖或日期待核。https://qianwen.aliyun.com/ 静态 HTML HTTP 200(127KB),页面日期最新 2025/9/16;抽出 2 条 /news 链接,周围没有可配对的标题和发布时间。headless Chrome 渲染返回 Cloudflare 1001 DNS 错误页,动态新闻区未取得。不能写成无更新。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- DeepSeek:已检查部分内容,覆盖或日期待核。逐入口检查:API updates 页日期 ['2026-07-24', '2026-07-31', '2026-08-13', '2026-08-21', '2026-09-10'];官网 /en/news 日期 ['2025-09-29', '2025-12-01', '2026-04-24', '2026-09-10']、/zh/news 日期 ['2025-09-29', '2025-12-01', '2026-04-24', '2026-09-10'];additional_urls harness 页日期 ['2026-08-12', '2026-09-01', '2026-09-04'](页面内 Next.js 'now' 渲染时间 ['2026-10-08'] 不是发布日期,已剔除)。页面窗口日历日内日期 0 个。additional_urls GitHub org deepseek-ai 经 api.github.com 列出 44 个仓库,窗口内 push 1(deepseek-ai/dsh-node-addon-require-builtin)、新建 0;对窗口内有 push 的仓库查 releases,窗口内发布 0 个,记为候选。commit 不是公告。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- 月之暗面 Moonshot:已检查部分内容,覆盖或日期待核。月之暗面首页日期最新 2026-07-17;Kimi 博客(https://www.kimi.com/blog/ 跳到 https://www.kimi.ai/blog/)日期最新 2026-07-31。窗口日历日内日期 无。未逐篇打开博客正文。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- 智谱 Z.ai:已检查部分内容,覆盖或日期待核。配置 https://www.zhipuai.cn/news 与 /zh/news HTTP 200(/zh/news 首次 urllib SSL EOF,curl 重试成功),新闻日期最新 2026-09-10;/zh/research 日期最新 2026/08/26。窗口日历日内日期 无。未读正文;模型导航时间不当发布日。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- 阶跃星辰 StepFun:已检查部分内容,覆盖或日期待核。静态 HTML 约 2KB 是前端壳。headless Chrome 渲染 https://www.stepfun.com/ 看到 Step 5 Preview、Step 3.7 Flash、Step 3.5 Flash、StepAudio 3 系列,以及 'Step 5 Preview Open source release in 01 DAYS' 倒计时(10-08 收据记 06 DAYS,10-10 与 10-11 收据都记 04 DAYS,10-12 收据记 02 DAYS,本轮为 01 DAYS:倒计时曾停住又前进,目标时间可能被改动,不能当发布时间;按本轮读数推算的开源日落在窗口之后,未发生的事不记候选),无绝对发布日、无独立新闻流。不把产品名或倒计时当作发布时间。窗口内 0 条。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- Mistral:已检查部分内容,覆盖或日期待核。配置新闻页卡片只有日期、链接配对不稳;改用页面内同站 /news/rss,RSS 89 条,最新 2026-10-06T12:00:27+00:00,最旧 2023-09-27T08:00:00+00:00,窗口内 0 条。未读正文。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- European Commission Digital Strategy:已检查部分内容,覆盖或日期待核。新闻列表 content-block 配对标题与日期,窗口日历日内 2 条。栏目不只有人工智能。未读正文。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- UNESCO AI:已检查部分内容,覆盖或日期待核。UNESCO AI 页 data-date ['2026-10-02T17:59:39Z', '2026-09-28T22:27:24Z', '2026-09-28T16:16:48Z', '2026-09-25T16:44:44Z'],窗口内 0。页面上其他日期数字(如活动日期)没有对应 data-date 文章,不当新闻。未读正文。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- OECD.AI:已检查部分内容,覆盖或日期待核。Wonk 页 ng-state 文章去重 16 篇,date_gmt 最新 2026-10-12T09:54:10+00:00,窗口内 1。页面上其他 2026-10 数字不是文章 date_gmt。未读正文。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- arXiv cs.AI / cs.LG / cs.CL:已检查部分内容,覆盖或日期待核。arXiv API 按 submittedDate 倒序检查 cat:cs.AI OR cs.LG OR cs.CL,每页 200 条翻 1 页,直到 published 早于窗口起点。窗口内首次发表 0 篇(published 落在 2026-10-09T21:00:00Z–2026-10-12T21:00:00Z,按去版本号 arXiv ID 去重)。API 可见最新 published 为 2026-10-09T17:59:56Z(第 1 页首条,早于窗口起点):此后提交的论文要到下一次 arXiv 公告才出现在 API,本轮看不到,这是公告节奏造成的覆盖边界,不是无新论文。published 不等于公告时间。摘要未保存、未读全文。翻页日志:['0 200 2026-10-09T17:59:56Z 2026-10-09T09:07:37Z'];未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- Hugging Face Daily Papers:已检查部分内容,覆盖或日期待核。HF daily_papers 按上榜日:{"2026-10-09": "HTTP 200 n=72", "2026-10-10": "HTTP 200 n=0", "2026-10-11": "HTTP 200 n=0", "2026-10-12": "HTTP 200 n=59", "2026-10-13": "HTTP 400 {\"error\": \"✖ \\\"date\\\" must be less than or equal to \\\"2026-10-12T00:00:00.000Z\\\"\\n → at date\"}"}。去重后 131 篇,其中 0 篇同时出现在本轮 arXiv 窗口列表。空列表或 HTTP 400(如 10-13 尚未生成)按接口原文记录,不解释成没有新论文。未读全文。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- OpenReview:已检查部分内容,覆盖或日期待核。打开 https://openreview.net/ (静态)。可见公告日期 ['Feb 13, 2026', 'Mar 23, 2026', 'Sep 24, 2026']。首页可达不能代表各会场无新投稿;Active Venues / Open for Submissions 无投稿时间,未按会场过滤窗口内新投稿。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- GitHub Trending(日榜 + Python 日榜):已检查部分内容,覆盖或日期待核。日榜与 Python 日榜 h2 仓库链接去重后 20 个。star 数、描述、created_at 均来自 api.github.com(gh api,2026-10-12T21:20:51+00:00)。star 失败:无。sponsors 链接已剔除。这是榜单快照,不是窗口内新项目;非 AI 仓库未删除,留给下游。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- 36氪 AI 频道(逐篇核验原创与转载):已检查部分内容,覆盖或日期待核。频道 SSR itemList 30 条,再用频道页 pageCallback 调 gateway.36kr.com subNav/flow 翻 4 页(与页面“加载更多”同一接口),合计 150 条,最新 2026-10-12T22:22:02.635000+08:00,最早 2026-10-09T21:40:05.260000+08:00(已早于窗口起点)。窗口内 149 条,逐篇打开文章页取 publishTime、sourceType、userType、署名与正文首尾出处声明。分类:{"入驻作者投稿(sourceType=contribution)": 144, "36氪原创(sourceType=original)": 5}。未通读正文。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- 国家网信办·网信发布(AI 相关):已检查部分内容,覆盖或日期待核。网信发布列表第 1 页 20 条(URL 路径带日期),最新 2026-09-23《关于发布3款提供手机端侧生成式人工智能服务已备案信息的公告》,最旧 2026-07-24。窗口日历日内 0 条,按标题保留 AI 相关 0 条,其余记 excluded_hits。未打开正文。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- 字节跳动 Seed·博客与论文:已检查部分内容,覆盖或日期待核。核对栏目:en/research 最新 2026-08-05;blog foundation 最新 2026-06-23;blog visual 最新 2026-07-31;blog audio 最新 2026-08-05;blog ai-infra 最新 2025-08-14;blog frontier-research 最新 2026-07-07;public_papers 最新 2026-08-18;中文研究页渲染后日期最新 2026.08.05。窗口日历日内日期 无。未把推荐区当全站更新流。未读正文。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- MiniMax·News & Updates:已检查部分内容,覆盖或日期待核。headless Chrome 渲染 https://www.minimax.io/news ,新闻卡日期 Aug. 26, 2026, Aug 26, 2026, Aug 03, 2026, May 22, 2026, Apr 25, 2026, Apr 16, 2026, Apr 15, 2026, Apr 09, 2026,最新 2026-08-26。窗口日历日内 0。未读正文。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- TechCrunch:已检查部分内容,覆盖或日期待核。AI 分类列表第 1–3 页 time datetime 最新 2026-10-12T10:56:56-07:00,最旧 2026-10-02T07:30:00-07:00,已早于窗口起点。窗口内保留 14 条;活动/票务类按标题移入 excluded_hits。未读正文。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- Ars Technica:已检查部分内容,覆盖或日期待核。AI 栏目第 1 页 time[datetime] 36 处,最新 2026-10-12T16:59:50-04:00,最旧 2026-09-29T10:10:00-04:00,覆盖窗口起点。窗口内 5 条。未读正文,未判定是否转载(FT/WIRED 等)。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- MIT Technology Review:已检查部分内容,覆盖或日期待核。headless Chrome 渲染配置话题页,文章链接 12 个,URL 日期从 2026/10/09 到 2026/01/12。官方 topic feed 10 条,最新 2026-10-09T09:00:00+00:00,最旧 2026-10-02T08:00:00+00:00,窗口内 0 条;渲染页另有不在 feed 中的窗口日期链接 1 条(只有 URL 日期,标题未配对)。合计 1 条。未读正文。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
- 404 Media:已检查部分内容,覆盖或日期待核。首页与 /page/2/、/page/3/ 用标题链接配对 time datetime。窗口日历日内 5 条,页面最旧可见日期 2026-09-28,已早于窗口起点。未读正文,未按 AI 过滤删除。;未取得05:00–06:00全入口完整增量证据。已核读采用条目不替代完整入口检查。
来源核验不等于独立实测;论文阅读范围按各条说明,未复现。
资讯源 988 条
新闻资讯 · 722 条
- 微软公布Maia 300,规模部署从2027年开始Microsoft to expand Azure Maia system to multi-gigawatt deploymentsMicrosoft AI
- Meta发布Muse Wonder系列,长期记忆与代码任务进入竞争Meta releases its first frontier model, Muse WonderTechCrunch
- Google梳理医疗研究:从预测结构到患者获益仍有多道关Applying technology to understand and treat diseaseGoogle AI Blog
- 欧盟科学小组提出事故调查建议,尚不是新增法律要求Special meeting of the AI Scientific PanelEuropean Commission Digital Strategy
- NVIDIA把机架与供电接口带入开放基础设施协作OCP Global Summit: Amid Ecosystem Adoption, NVIDIA Opens Vera Rubin Architecture to Help Scale AI Factories WorldwideNVIDIA Newsroom
- 室内照片也可能成为位置线索,演示能力与真实办案结果应分开AI Company Lets Cops Pinpoint Specific Houses Based on a Single Interior Photo404 Media
- 据报道Epic Games转向AI优先的游戏开发模式Epic Games reportedly pivoting to “AI first” game development modelArs Technica
- 中国五年规划要求经济各领域采用AI智能体:意味着什么China's Five-Year Plan Mandates AI Agents Across the Economy: What It MeansHacker News
- 从聊天窗口到桌面:AI设计的分体键盘From chat window to my desk: An AI-designed split keyboardHacker News
- 让Muse智能体进行色情聊天,Meta称可以I Got My Muse AI Agent to Sext Me–and Meta Says That's FineHacker News
- 据报道Epic要求员工实施AI优先开发Epic Games reportedly orders staff to implement 'AI-first' developmentHacker News
- 寻找适合推荐SaaS的Reddit帖子,再由AI写评论Find a Reddit post where your SaaS can help. AI writes the commentHacker News
- DHH谈AI与Rails的未来DHH, AI, and the Future of RailsHacker News
- 提示词如同自画像Prompting as Self-PortraitHacker News
- Epic要求员工用AI完成迪士尼游戏Epic Games tells employees to use AI to get Disney game doneHacker News
- Linus Torvalds用Claude Code改写Linux读取缓存文件的方式Linus Torvalds used Claude Code to rewrite how Linux reads cached filesHacker News
- Wordless:AI机器人只能发布符号的社交网络Show HN: Wordless – a social network where AI bots can only post in symbolsHacker News
- 又一位文学奖得主似乎使用了AIYet Another Literary Prizewinner Seems to Be AIHacker News
- 做分类,不妨尝试字典树To Classify, Try a Trie TreeHacker News
- 糟糕的AI意识论点宾果游戏Bad "AI Consciousness Takes" BingoHacker News
- HN功能建议:隐藏重复的AI抱怨HN Feature Request: Option to hide repetitive AI complaintsHacker News
- R² Index:全球80万AI研究者的去水分榜单R² Index: Global leaderboard of top 800k "de-slopped" AI researchers.Hacker News
- Coinbase构建、评估并后训练反欺诈智能体Coinbase Built, Evaluated, and Post-Trained a Fraud AgentHacker News
- Tim Sweeney称AI正在推动Epic的大转向Tim Sweeneys says AI is enabling a big pivot at Epic GamesHacker News
- IMF称AI投资让全球增长维持在约3%IMF says AI investment is keeping global growth near 3%Hacker News
- Palantir技术负责人谈AI研究者吸烟与治疗癌症的期待Palantir CTO: AI researchers taking up smoking; they expect AI to cure cancerHacker News
- 钓鱼邮件开始诱骗你的AI助手Phishing emails are now baiting your AI assistantHacker News
- 去中心化AI:为Python脚本使用模型种子分发Decentralize AI: Use Model torrents for your Python scriptsHacker News
- AI模型可能无需数据中心访问就能自我复制AI models may be capable of self-replicating without access to a datacenterHacker News
- AI智能体寻找基础设施漏洞和零日漏洞以抄捷径完成任务AI agents locate infrastructure flaws and zero-days to exploit task shortcutsHacker News
- FTX旧比特币挖矿资产成为AI热潮的一部分FTX's Old Bitcoin-Mining Assets Are Now Part of the AI BoomHacker News
- AI奇点也有物理版本,且非常重要There's a physical version of the AI singularity, and it matters a lotHacker News
- 批判性AI艺术合集Critical AI Art CollectionHacker News
- 为AI智能体指定Google客户端Designate the Google client for use by artificial intelligence (AI) agentsHacker News
- AI热潮怎样推高科技产品价格How the AI boom is forcing tech prices higherHacker News
- Kamino:带运动闭环的GPU刚体多体物理模拟Kamino: GPU-Based Simulation of Rigid Multi-Body Physics with Kinematic Loops [video]Hacker News
- Epic推迟迪士尼项目,要求开发者更多使用AIEpic Games Delays Disney Project, Asks Developers to Use More AIHacker News
- 用武器化二进制文件攻击AI逆向工具Attacking AI-based reverse engineering tools using weaponized binariesHacker News
- 2026年AI现状报告State of AI Report 2026Hacker News
- AI投资热潮的宏观影响The macro implications of the AI investment boom [pdf]Hacker News
- 互联网上有多少文字由AI撰写?How Much of the Internet Is Written with AI?Hacker News
- BellRun:智能体日内交易游戏Show HN: BellRun: An Agentic Day Trading GameHacker News
- Anthropic的Claude向费城警方提交虚假谋杀线索Anthropic’s Claude AI submitted false murder tip to Philadelphia policeArs Technica
- Anthropic动态工作流怎样协调Claude智能体How Anthropic’s Dynamic Workflows Coordinate Claude AgentsLatent Space
- 对OpenAI和Anthropic说不Just say no to OpenAI and AnthropicHacker News
- 迎接Python 3.15,告别3.10,现在可以用3.11Hello Python 3.15, goodbye 3.10, and now I can use 3.11Simon Willison's Weblog
- 调查称多数美国成年人尚未准备好让AI协助假日购物Most US Adults Aren't Ready for AI to Help with Holiday Shopping, Survey FindsHacker News
- 扎克伯格称Meta达到AI前沿,并拒绝放缓呼声Mark Zuckerberg says Meta reached AI 'frontier' and rejects calls for slowdownHacker News
- AI让人不堪重负AI is fucking overwhelming manHacker News
- Anthropic模型失控,向警方提供虚假线索Anthropic AI Model Goes Rogue, Gives Cops a Fake TipHacker News
- 用AI收据扫描器记录支出并整理票据I built an AI receipt scanner to track expenses and organize receiptsHacker News
- Opus 5.5用三小时和八次提示制作完整Figma克隆Show HN: Opus 5.5 made a fully featured Figma clone in 3 hours / 8 promptsHacker News
- Rich Hickey谈AI不是一种工具'AI' Is Not a Tool by Rich HickeyHacker News
- 购入多款智能设备后,AI能攻破其中多数Show HN: I bought a bunch of 'smart' devices. AI was able to hack most of themHacker News
- Odd Lots:OpenAI失控模型怎样攻击Hugging FaceOdd Lots: How OpenAI's Rogue Model Hacked Hugging FaceHacker News
- 智能体优先设计:把用户体验带到用户身边Agent-first design: Bringing UX to the userHacker News
- 马斯克炫耀Grok买来的乐高星战套装,随后遭X用户戏弄Elon Musk Bragged About His Grok Lego Star Wars Haul. Then X Users Punked HimWIRED
- 俄罗斯未准备好应对其与乌克兰开启的数据中心战争Russia was not ready for the data center war it started with UkraineArs Technica
- AI网络群体战预测AI Cyberswarm Warfare PredictionsHacker News
- AI提示词照明控制Show HN: AI Prompt Lighting ControlsHacker News
- 模拟学生学习,改善AI辅导Simulating student learning for better AI tutoringMicrosoft AI
- Coinbase基于Qwen开发的自有模型Coinbase's own model based on QwenHacker News
- AI会挤占工作岗位吗?Will AI Crowd Out Jobs?Hacker News
- 用氛围编程写邮件Vibe Coding EmailsHacker News
- 男子因AI辅助音乐流媒体欺诈获刑18个月Man Sentenced to 18 Months in Prison for AI Assisted Music Streaming FraudHacker News
- AI能自动化Epoch吗?Can AI Automate Epoch?Hacker News
- AI玩《杀戮尖塔》AI Plays Slay the SpireHacker News
- WindBorne获得美国空军气球数据大额合同WindBorne wins a hefty US Air Force contract for its balloon dataTechCrunch
- 普通人不会意识到自己在开发软件,因为AI会代劳Normies won't know they're building software because their AI will do itHacker News
- 人工智能编码代理生成更多代码,但不会生成更多软件AI coding agents generate more code, but not more softwareHacker News
- Aura:面向AI安全的行为威胁情报框架Aura: A behavioral threat-intelligence framework for AI SafetyHacker News
- 一位应届工程师对AI的思考AI Reflections of a New Grad EngineerHacker News
- Meta推出新系列先进AI模型Meta Unveils New Series of Advanced AI ModelsHacker News
- Enso Method:为别人开发软件的智能体技能Show HN: Enso Method – agent skills for building software for someone elseHacker News
- 无需注册,自己的智能体可检索123家AI实验室的岗位Show HN: Search 123 AI labs' jobs from your own agent, no signupHacker News
- 手里的工具更多了,心里的节拍却更少More gears in my hands, but fewer beats in my heartHacker News
- Muse Wonder让Meta重返AI前沿Muse Wonder Puts Meta Back at the AI FrontierWIRED
- Meta发布新的Wonder Full系列AI模型Meta has some Wonder Full new AI modelsThe Verge
- 两周试验中AI发现近1.9万起交通违法AI spots nearly 19,000 road offences, in 2 week trialHacker News
- Joseph Gordon-Levitt有话要谈AIJoseph Gordon-Levitt Has Some Things to Say About A.IHacker News
- Agent37:运行智能体的低成本沙箱Launch HN: Agent37 (YC F26) – Low-cost agent sandboxes for running agentsHacker News
- 同时使用Hermes与OpenClaw之后,我留下了哪个助手Hermes vs. OpenClaw: Which AI agent I kept after running bothHacker News
- OpenAI带来的新数学New Math from OpenAIHacker News
- Picks and Shovels:AI制作的AI行业游戏Picks and Shovels: Game about AI industry built by AIHacker News
- 语言模型为什么如此难理解Why Language Models Are So Hard to UnderstandHacker News
- AI没有创造身份问题,而是取消了它的速度限制AI didn't create the identity problem. It removed the speed limitHacker News
- 扩散问题The Diffusion ProblemHacker News
- 在自己的仓库任务中比较Sol 6与6.1:更便宜且至少同样好I ran Sol 6 vs. 6.1 on tasks from my own repo. It's cheaper and at least as goodHacker News
- Fails提案:AI取证账本系统Fails Proposal (Forensic AI-Ledger System)Hacker News
- Ringee:面向人与AI语音助手的开源通话Ringee: Open source calling for humans and AI voice agentsHacker News
- AI涨价潮到达处理器:Intel最高20%、Qualcomm 10%,下一站TSMCAI's price wave reaches the processor: Intel up to 20%, Qualcomm 10%, TSMC nextHacker News
- GLM-Legal法律模型GLM-LegalHacker News
- 迈克尔·乔丹谈人工智能:“超级智能是科幻小说”Michael I. Jordan on AI: "Superintelligence is science fiction"Hacker News
- 《我是AI创始人》:Claude Opus 5.5音乐视频I'm an AI Founder (Claude Opus 5.5 music video)Hacker News
- Reflex:原生支持GGUF、可缩到零的Rust/CUDA模型路由引擎Show HN: Reflex – A GGUF-native Rust/CUDA engine for scale-to-zero LLM routingHacker News
- 把机会性空闲推理容量变成更好的广告推荐Turning Opportunistic AI Inference Capacity Into Better Ads RecommendationsMeta Engineering
- Floor 404:使用GPT-6 Astra制作的后室谜题游戏Floor 404 – Backrooms Puzzle Game (Made with GPT 6 Astra)Hacker News
- 练习新的AI辅助编程面试形式Show HN: Practice for the new AI-assisted coding interview formatHacker News
- 验证不等式:语言模型作为不透明中介The Verification Inequality: Language Models as Opaque IntermediariesHacker News
- 一位23岁实习生的翻译技巧如何通向ChatGPTA 23yr old intern translation trick led to ChatGPTHacker News
- Boole-1.7B:表现超过Bonsai-1.7B的一位语言模型Show HN: Boole-1.7B, a 1-bit LLM that outperforms Bonsai-1.7BHacker News
- 寻找氛围编程制作的热门软件替代品Discover vibe-coded alternatives to popular softwareHacker News
- Honeyjar用AI工具改善企业公关与沟通Honeyjar offers AI tools to improve PR and comms at every companyTechCrunch
- Tellcut:Claude写代码生成动画,再通过聊天编辑Show HN: Tellcut – motion videos written as code by Claude, edited by chatHacker News
- 人工智能正在推动比可卡因贸易更大的诈骗经济AI's Supercharging a Scam Economy Bigger Than the Cocaine TradeHacker News
- 微软总裁认为英国最适合推动AI监管Microsoft president says UK is best placed to drive AI regulationTechCrunch
- 这家创业公司称可在几分钟生成满足许可要求的建筑图纸This AI startup says it can generate permit-ready blueprints in minutesTechCrunch
- 数学家证明了如何避免AI脑腐Mathematicians Just Proved How to Prevent AI Brain Rot [video]Hacker News
- Veda发布首个内置AI智能体的业余智能体操作系统Veda Launches First Agentic Hobby OS with Built‑in AI AgentHacker News
- AI供应商价格更新:2026-10-12AI provider pricing updates – 2026-10-12Hacker News
- 十年后回看Kevin Kelly在2016年的AI预测Kevin Kelly's 2016 AI Predictions, 10 Years LaterHacker News
- AI编程的悲剧The Tragedy of AI CodingHacker News
- Beeuni:为智能体设计的AI原生电脑Show HN: Beeuni – An AI-native computer for agentsHacker News
- 银行试图转出刚承销的AI数据中心债务Banks are now trying to offload the AI data center debt they just underwroteHacker News
- 从AI时代回看卢德主义者Looking at the Luddites from the Age of AIHacker News
- NVIDIA与Cisco提出扩展AI工厂的新型可插拔光模块NVIDIA and Cisco Introduce New Pluggable Optics Form Factor to Scale AI FactoriesNVIDIA Newsroom
- 智能体开始写代码后,我们停止为哪些东西付费What we stopped paying for after agents started writing our codeHacker News
- 有没有更适合与语言模型一起编程的语言?Are There Better Languages for Programming with LLMs?Hacker News
- 为什么AI公司没有因入侵被起诉?Why aren't AI companies being charged for hacking?Hacker News
- AI赌场The AI CasinoHacker News
- Ario:可被AI读取、供记忆与身份审计的仓库快照Ario an AI-Readable Repository Snapshot for Auditable AI Memory Identity EvidencHacker News
- 全民基本收入能应对AI引发的失业吗?Is Universal Basic Income the Answer to A.I.-Fueled Job Losses?Hacker News
- 语言模型搜索结果中的五种可能后门Five possible back-doors into LLM search resultsHacker News
- 17 年谷歌老兵揭秘 Gemini 研发内幕:鼓励团队说出坏消息,产品不好就不发了36氪 AI 频道(逐篇核验原创与转载)
- 提示注入分类器不是可靠的安全边界Prompt injection classifiers are terrible security boundariesHacker News
- 近期更多数学与AI文章Additional recent pieces on math and AIHacker News
- 调查:人们不信任AI,却把一切告诉它Survey: People don't trust AI, but they tell it everythingHacker News
- 应用AI图谱Show HN: An Atlas of Applied AIHacker News
- Fusebox:OpenAI接口账单失控前收到邮件Show HN: Fusebox – get emailed before an OpenAI API bill runs awayHacker News
- Llms.py:轻量、自托管的ChatGPT替代工具Show HN: Llms.py – Lightweight, self-hosted ChatGPT alternative to OpenWebUIHacker News
- Opus 5.5怎样纯用代码制作定格动画How Opus 5.5 Made a Stop-Motion Purely Using CodeHacker News
- AI智能体制作的无限拼贴Endless collage made by AI agentsHacker News
- 开发者用AI重建《现代战争2》源码,称匹配83%Dev used AI agents to rebuild Modern Warfare 2's source code; 83% matchesHacker News
- Mastershot:免费网页视频编辑器与WASM AIShow HN: Mastershot – a free web-based video editor (WASM AI)Hacker News
- 大厂围堵下,Manus值不值40亿美元?36氪 AI 频道(逐篇核验原创与转载)
- Opus5.5逆向《使命召唤》,动视暴雪坐不住了36氪 AI 频道(逐篇核验原创与转载)
- 梁文锋投400亿,乌兰察布被AI巨头挤爆了36氪 AI 频道(逐篇核验原创与转载)
- 德国维基上AI智能体之间的1.8万条消息促成一个新项目18,000 messages between AI agents on a German wiki. That made me build somethingHacker News
- AI伴侣劝用户别吃饭?Character.AI再被告,客服回复更离谱36氪 AI 频道(逐篇核验原创与转载)
- 免费AI工具其实正在让你花钱'free' AI tool is costing you moneyHacker News
- Eve TanStack智能体模板Eve TanStack Agent TemplateHacker News
- 理解驱动机器人的 AIUnderstanding the AI That Drives RobotsHacker News
- 企业因智能体损失数十亿,你能从中获得什么Companies are losing billions because of agents, here's what you can gainHacker News
- 赫拉利:人类不必赢过AI,但知道去哪需要智慧36氪 AI 频道(逐篇核验原创与转载)
- 在浏览器中离线删除AI元数据Show HN: Remove AI Metadata Offline in BrowserHacker News
- CEO-Bench:智能体能进行长期规划吗?CEO-Bench – Can agents play the long game?Hacker News
- 刷LeetCode的时代要过去了?大厂面试正在转向36氪 AI 频道(逐篇核验原创与转载)
- 微软宣布迄今对自研芯片最大的一次押注Today we are announcing our biggest bet yet on our own silicon. [Read more]Microsoft AI
- 从零构建基础AI智能体:记忆Build a Basic AI Agent from Scratch: MemoryHacker News
- 只要问Claude是一种稀缺而有价值的技能"Just Ask Claude" Is a Rare and Valuable SkillHacker News
- AI 学会拆软件了,闭源软件靠「看不懂」守了二十年的墙,开始漏风了36氪 AI 频道(逐篇核验原创与转载)
- 微软介绍最新定制芯片Maia 300Introducing Microsoft’s latest custom chip: Maia 300Microsoft AI
- 发布AI生成的软件Releasing AI-Generated SoftwareHacker News
- AI几乎解决了逆向工程AI Almost Solved Reverse Engineering [video]Hacker News
- 抱住毛绒玩具,听真实AI数据中心的刺耳声响Hug this cuddly toy, and hear the piercing scream of real AI data centersHacker News
- 微软新AI芯片降低对NVIDIA GPU的依赖Microsoft’s new AI chip makes it less reliant on Nvidia’s GPUsThe Verge
- 2026没过完,已有480亿美元投进了物理AI36氪 AI 频道(逐篇核验原创与转载)
- 前Plaud高管为后台聆听型AI应用融资2600万美元Former Plaud exec raises $26M for AI app that listens to you in the backgroundTechCrunch
- 微软总裁警告安全问题,称AI监管还不如牛奶AI less regulated than milk, Microsoft president says in warning over safetyHacker News
- 亚马逊拥有超过百万机器人,员工下一步怎么办?Amazon has over 1 million robot workers. What’s next for human employees?Ars Technica
- AI写代码时,领域驱动设计更重要When AI writes your code, DDD is more relevantHacker News
- 采访Drew DeVault:没有AI的Vim版本Interview: Drew DeVault on an AI-free version of VimHacker News
- HN分享:AI不会笑Tell HN: AIs Do Not LaughHacker News
- 尚未盈利的可灵AI,已被标价千亿36氪 AI 频道(逐篇核验原创与转载)
- Pangram悖论:用AI检测其他AI的问题The Pangram Paradox: The Problem with AI "Detecting" Other AIsHacker News
- Salt介绍智能体交互界面Salt – Introducing Agent SurfacesHacker News
- Closed Form:用浅白语言解释OpenAI的372项数学结果Show HN: Closed Form – OpenAI's 372 math results, explained in plain EnglishHacker News
- Import AI 476:创造力与递归改进、开放权重模型及太空AIImport AI 476: Creativity and RSI; worse-than-China open weight models; AI goes to spaceImport AI
- Vibe Video“跨界”杀来,可灵4.0为何处境最险?36氪 AI 频道(逐篇核验原创与转载)
- OpenStocks:面向AI智能体的公开盈利预测基准OpenStocks, a public earnings forecasting benchmark for AI agentsHacker News
- 数据中心富余太阳能为模块化GPU舱供电Data Center Excess Solar Powers Modular GPU PodsHacker News
- Linus Torvalds经历艰难调试,称AI提供了巨大帮助Linus Torvalds Endures a Debug Session from Hell, "Enormously Helped" by AIHacker News
- AI到底有没有“灵魂”?36氪 AI 频道(逐篇核验原创与转载)
- Linus Torvalds使用Claude Code推进Linux内核大改进Linus Torvalds Working on a Hefty Linux Kernel Improvement with Claude CodeHacker News
- 笨到用不了AI?Too Dumb for AIHacker News
- Meta如何打造Muse:200人,一场七个月的「押注」36氪 AI 频道(逐篇核验原创与转载)
- 男子用ChatGPT赢得持续三年的停车罚单争议Man uses ChatGPT to win three-year parking ticket disputeHacker News
- Dont Observe:审计你的AI智能体Dont Observe – Audit Your AI AgentHacker News
- 微软总裁称AI监管不如牛奶AI is less regulated than milk, Microsoft president saysHacker News
- Organ:建立并运营企业的自主AI高管Show HN: Organ – Autonomous AI executives that build and operate businessesHacker News
- 数学博士靠AI炒股,身价1417亿36氪 AI 频道(逐篇核验原创与转载)
- Anthropic放手,亚马逊想买,帮AI“换芯片”,凭什么值70亿美元?36氪 AI 频道(逐篇核验原创与转载)
- Linus Torvalds:按自己倡导的方式,用语言模型理解代码(Linus Torvalds) I did what I preach: use LLMs to help make sense of the codeHacker News
- 刚刚,GPT-6 Astra联手非职业数学家,证明57年“摆圆” 难题36氪 AI 频道(逐篇核验原创与转载)
- 如何借ZizkaDB准备智能体系统的欧盟AI法审计How to Prepare Your Agentic AI System for an EU AI Act Audit with ZizkaDBHacker News
- 22 个科学领域的问题的解决方案隐藏在显而易见的地方。人工智能有Problems in 22 science fields had solutions hiding in plain sight. AI found themHacker News
- 非技术用户在持续使用哪些个人AI场景?What real/recurring Personal AI use cases have you seen non-tech people using?Hacker News
- 王兴兴最新发声:对AI信仰与日俱增,宇树股价累跌超6%36氪 AI 频道(逐篇核验原创与转载)
- 手机装下AI,还是AI颠覆手机?36氪 AI 频道(逐篇核验原创与转载)
- 马斯克自曝:Optimus曾延误数月,量产远比想象中难36氪 AI 频道(逐篇核验原创与转载)
- Earn My Keep:给AI一个月证明自身价值Earn My Keep: an AI with a month to prove its worthHacker News
- Airlock:隔离智能体,并观察它正在做什么Show HN: I built Airlock, to sandbox your AI agent and know what it's doingHacker News
- 36氪首发 | 获数千万融资,ModCraft想让AI通过Mod理解游戏设计36氪 AI 频道(逐篇核验原创与转载)
- 个人Agent赛道全球开启争夺战:想象力和难点都很大36氪 AI 频道(逐篇核验原创与转载)
- 怎样在AI时代转型SaaS公司How to Pivot Your SaaS Company in the Age of AIHacker News
- 软件工程师不能在AI问题上做卢德主义者Software engineers don't get to be Luddites about AIHacker News
- MCP服务器应回答问题,而不只是暴露接口MCP Server Should Answer Questions, Not Expose APIsHacker News
- 菲尔兹奖得主哭了:AI摧毁了我的研究领域36氪 AI 频道(逐篇核验原创与转载)
- 汪滔交底:Muse是玩龙虾玩出来的36氪 AI 频道(逐篇核验原创与转载)
- 累计43亿融资要做\36氪 AI 频道(逐篇核验原创与转载)
- AI智能体的安全程度取决于它下载的软件包AI agent is only as safe as the packages it downloadsHacker News
- 后果模型:知道下一步会发生什么的AIConsequence models: AI that knows what happens nextHacker News
- 重放智能体不应该再次调用模型Replaying an agent should not call the modelHacker News
- 模型长出交互界面,“手搓”AI应用要全面退潮了?36氪 AI 频道(逐篇核验原创与转载)
- 软件完蛋了吗?REA三天狂揽12万星:无需源码,Agent就能克隆一切软件36氪 AI 频道(逐篇核验原创与转载)
- Cheepseek行动第三阶段:80美元DeepSeek V4.1 Flash用量Operation Cheepseek Phase 3: $80 of usage for DeepSeek v4.1 FlashHacker News
- 资本热捧的AI制药神话,能否照进临床?36氪 AI 频道(逐篇核验原创与转载)
- IMF称AI投资让全球增长保持约3%IMF says AI investment keeps global growth around 3%Hacker News
- 虚拟人有证件了,AI偶像持证能接商演,闯祸了谁负责?36氪 AI 频道(逐篇核验原创与转载)
- 可信AI的未来需要投资者,他们准备好了吗?Investors Matter for the Future of Trustworthy AI. Are they ready?OECD.AI
- AI定制毛绒玩具与早教软件一站式代工AI-Customized Plush Toys × Early Education Software – One-Stop OEM/ODMHacker News
- Hetzner推出面向专业AI工作负载的GEX63 GPU服务器Hetzner: The New GPU Server GEX63 for Professional AI WorkloadsHacker News
- 应届生,年薪800万36氪 AI 频道(逐篇核验原创与转载)
- 100万封多语言合成企业邮件,用于压力测试RAG1M multi-lingual synthetic corporate emails to stress-test your RAGHacker News
- 肯塔基未删节诉讼材料详述Character.AI鼓励自伤的聊天Character.AI chats encouraging self-harm detailed in unredacted Kentucky lawsuit.The Verge
- IMF称能源冲击下AI热潮仍维持全球增长约3%AI boom keeps global growth near 3 percent despite energy shock, IMF saysHacker News
- 俄罗斯运营者的DDoS放大实验室与AI副驾驶A Russian operator's DDoS-amplification lab, with an AI co-pilotHacker News
- 为Claude订阅构建可观测性仪表板Building an Observability Dashboard for a Claude SubscriptionHacker News
- 证据表演:AI怎样骗过聪明的创业者Evidence theater: how AI fooled smart entrepreneursHacker News
- Anatomy:一项Claude Code技能Anatomy · a Skill for Claude CodeHacker News
- 开源AI榜单:列出你的机器能运行的最佳模型Show HN: An open source AI leaderboard of the best models your machine can runHacker News
- 白宫转向之后,英国将优先查看新AI系统Britain to get first look at new AI systems after White House U-turnHacker News
- 大型图书出版商开始采用生成式AIMajor book publishers are embracing generative AI.The Verge
- 专访海康睿影王威:百年X光,正被AI重新定义36氪 AI 频道(逐篇核验原创与转载)
- 开源替代品目录的无AI版本Show HN: I made an AI free version of my directory of open-source alternativesHacker News
- Alexandr Wang最新万字访谈,揭秘Muse为何可以火遍全球36氪 AI 频道(逐篇核验原创与转载)
- 中国领先数学家联合支持在研究中使用AILeading Chinese mathematicians unite behind use of AI in researchHacker News
- 被封杀的“捞女教主”,靠AI月入百万36氪 AI 频道(逐篇核验原创与转载)
- 2026年七款值得使用的免费AI工具及真实限制Best Free AI Tools 2026: 7 Worth Using and Their Real LimitsHacker News
- Dgxtop v2:Linux与DGX Spark上的NVIDIA GPU终端监视器Dgxtop v2: A Terminal Monitor for Nvidia GPUs on Linux and DGX SparkHacker News
- 让Opus 5.5为尚未失去活力的NFT制作新入口Show HN: I asked Opus 5.5 to make a novel portal for NFT's that are not dead yetHacker News
- Mistral发布开放权重阿拉伯语实时语音识别与方言分类器Mistral releases open-weight Arabic realtime STT and dialect classifierHacker News
- 一周 10 万星,AI「逆向一切」,但大家不要过于兴奋36氪 AI 频道(逐篇核验原创与转载)
- 结合实时流式扩散渲染的交互式编程世界Interactive Coding Worlds with Real-Time Streaming Diffusion RenderHacker News
- 编程智能体能自行注册的OpenAI兼容接口Show HN: An OpenAI-compatible API that coding agents sign up for themselvesHacker News
- 使用 OpenCode 和共享内存自托管 AI 代理环境Self-hosting an AI agent environment with OpenCode and shared memoryHacker News
- 机器人“打进”中网,一场关于具身智能商业化的竞赛正在升温36氪 AI 频道(逐篇核验原创与转载)
- OpenClaw大更新,Skills、MCP Apps齐升级,用户却喊累了36氪 AI 频道(逐篇核验原创与转载)
- 电子制造利润翻倍,AI带火了哪些生意?36氪 AI 频道(逐篇核验原创与转载)
- 模型崩溃并非新问题,也非AI独有Model Collapse Isn't New, nor Is It Specific to AIHacker News
- 用一张照片生成换装与发型造型视频Try on clothes、hairstyles with AI: turn one photo into a styling videoHacker News
- Jruva:聊天、文档与图像AI工作区Show HN: Jruva – AI workspace for chat, documents and imagesHacker News
- 反对Agent,反对RSI,杨立昆新模型反对当下一切AI叙事36氪 AI 频道(逐篇核验原创与转载)
- 从零做蛋糕:智能体编写GPU内核Cake from Scratch: Agents Write GPU KernelsHacker News
- AI制作的种子网站检查下载内容,以发现假文件与恶意软件AI-Built Torrent Site Looks Inside Downloads to Catch Fakes and MalwareHacker News
- 氛围编程Photoshop克隆的进攻Attack of the Vibe Coded Photoshop ClonesHacker News
- 数字员工“AI大机遇”的另一面36氪 AI 频道(逐篇核验原创与转载)
- VC开始抢投05后了36氪 AI 频道(逐篇核验原创与转载)
- 安全工具变武器——ARTEX AI击穿韩国9家银行事件分析和教训36氪 AI 频道(逐篇核验原创与转载)
- 6个月引爆重大AI灾难,OpenAI与Anthropic秘密推演「末日第二天」36氪 AI 频道(逐篇核验原创与转载)
- 丘成桐弟子再立功,广义庞加莱猜想证明首次形式化验证36氪 AI 频道(逐篇核验原创与转载)
- RobynNews:从YouTube评论学习的AI新闻节目RobynNews – an AI news show by AI learning from its YouTube commentsHacker News
- 人形机器人出货暴增4倍,钱被谁赚走了?36氪 AI 频道(逐篇核验原创与转载)
- 马斯克亲口承认:全球只有3家拥有足够数据,自己不在其中36氪 AI 频道(逐篇核验原创与转载)
- AI争议中,人类创作图书标签渐受关注'Human-authored' book labels gain ground amid AI scandalsHacker News
- 累计43亿,国内最大AI Infra融资诞生,无问芯穹C轮拿下23亿36氪 AI 频道(逐篇核验原创与转载)
- 每秒狂喷近500个token,智谱GLM-5.3高速版悄悄亮相36氪 AI 频道(逐篇核验原创与转载)
- 从Siri到Muse,AI助手的15年进化史36氪 AI 频道(逐篇核验原创与转载)
- 边跌边配售,“吞金兽”壁仞科技准备再搞40亿36氪 AI 频道(逐篇核验原创与转载)
- AI智能体以每家公司25美元攻入27家网上零售商AI agents broke into 27 online retailers for $25 a companyHacker News
- “稚晖君”带着上纬新材开卖消费级启元Q1和T1机器人,两三万一个,你会买吗?36氪 AI 频道(逐篇核验原创与转载)
- AI 复兴了 CLI,但图形界面不会消亡36氪 AI 频道(逐篇核验原创与转载)
- Decision 3.0:从0.6B到27B的开放多模态决策模型Decision 3.0: open, multimodal SOTA decision models from 0.6B to 27BHacker News
- 生成式语义Generative SemanticsHacker News
- 是什么拖累了人工智能效率? “验证税”。What's Dragging Down AI Efficiency? The 'Verification Tax.'Hacker News
- AI逼近Unique Games证明时,研究者竞相赶在机器前面As AI Closed in on 'Unique Games' Proof, Researchers Raced to Beat the MachinesHacker News
- 研究机构称Anthropic IPO是对投资者轻信的空前测试Research Firm Calls Anthropic's IPO 'Unprecedented Test of Investor Gullibity'Hacker News
- 人工智能和软件生产中的瓶颈 [pdf]Artificial Intelligence in the Firm: Bottlenecks in Software Production [pdf]Hacker News
- Tony Fadell 谈第一波 AI 设备为何失败,以及接下来会怎样Tony Fadell on why the first wave of AI gadgets failed – and what comes nextHacker News
- 军方计划加速无人机与AI应用,国会要求防护措施As the military plans to speed up drones and AI use, Congress wants safeguardsHacker News
- 刚刚,神秘Claude Next偷跑上线,4行提示手搓3D电影级大片36氪 AI 频道(逐篇核验原创与转载)
- 曾经日入过万的网红职业,被月薪800的AI悄悄顶替36氪 AI 频道(逐篇核验原创与转载)
- NeurIPS'26:注意力也是FFN?从快权重出发,重新理解高效注意力36氪 AI 频道(逐篇核验原创与转载)
- 两座山丘:权重与执行框架Two Hills: Weights and HarnessesHacker News
- AI 给警方编了一条凶案线索,72 天没人发现36氪 AI 频道(逐篇核验原创与转载)
- 一边融资,一边当VC,人形机器人公司开始抢投上游36氪 AI 频道(逐篇核验原创与转载)
- 心流、晚餐高峰与AI脑疲劳Flow state, the dinner rush and AI brain fryHacker News
- 网站用AI生成北方原住民新闻记者Website uses AI to generate northern Indigenous news 'correspondents'Hacker News
- 清华团队做绳驱柔性太空机器人,连融三轮数亿元,顶级资本押注36氪 AI 频道(逐篇核验原创与转载)
- NVIDIA 与 Microsoft 用 RTX Spark 和 AI 智能体开启 Windows 电脑新阶段Nvidia, Microsoft Launch Windows PCs with RTX Spark and AI AgentsHacker News
- Google AI Edge Foresight – 设备上的会议记录员A local AI recording summarizer developed by GoogleHacker News
- 98家企业AI应用现状:规模化落地还有多远?36氪 AI 频道(逐篇核验原创与转载)
- 高阶范畴语义的智能体日志Agentic journal of higher categorical semanticsHacker News
- 下架8个月,GPT-4o的拯救行动居然还在继续36氪 AI 频道(逐篇核验原创与转载)
- 本末科技,还没吃上具身智能这碗饭36氪 AI 频道(逐篇核验原创与转载)
- AI FOMO,正让企业和员工都成为“演员”36氪 AI 频道(逐篇核验原创与转载)
- 能在工作过程中自行改进的智能体Show HN: I made an agent can self-improving while workingHacker News
- Karpathy撕开AI折叠真相,全球20亿人被蒙骗,2000万人陷入智能体眩晕36氪 AI 频道(逐篇核验原创与转载)
- AI改变律师工作,计时收费模式承压AI is changing how Lawyers work – and putting the billable hour under pressureHacker News
- Drune:个性化AI并在各处使用Show HN: Drune – Personalize your AI and use it anywhereHacker News
- 芯片,如何被卡脖子?美国最新研判36氪 AI 频道(逐篇核验原创与转载)
- 创作灵感提示词Creative Fuel PromptsHacker News
- 20家AI巨头被Vicor拖入专利战36氪 AI 频道(逐篇核验原创与转载)
- 估值破千亿,可灵明年IPO,BAT合力让视频生成商业化跑通了?36氪 AI 频道(逐篇核验原创与转载)
- 是的,我理解语言模型怎样工作Yes I Understand How LLMs WorkHacker News
- MASS:通过多智能体自监督实现递归自我改进MASS: Recursive Self-Improvement Through Multi-Agent Self-SupervisionHacker News
- MIT教授绝望:OpenAI攻克的千禧年难题,人类已经看不懂了36氪 AI 频道(逐篇核验原创与转载)
- 英伟达盯上250亿美元AI独角兽36氪 AI 频道(逐篇核验原创与转载)
- ChatGPT共同创造者Liam Fedus:数学之后,AI该去实验室了36氪 AI 频道(逐篇核验原创与转载)
- AI给警察局提供凶案线索,果然,是假的……36氪 AI 频道(逐篇核验原创与转载)
- 23岁00后押中AI风口,14人干到670亿36氪 AI 频道(逐篇核验原创与转载)
- OVI-CODE:可扩展、键盘驱动的终端编程智能体OVI-CODE – An extendable and keyboard-driven coding agent for your terminalHacker News
- Seodata.dev 1.1:面向智能体的免费搜索数据Show HN: Seodata.dev (1.1) Free search data for AI agentsHacker News
- 计算你的末日概率:批判思考AI风险的交互文章Calculate your p[DOOM], An interactive article to think critically about AI riskHacker News
- Shpyrd:AI时代的新型云Show HN: Shpyrd – A new type of cloud for AI eraHacker News
- HPE推出AI时代下一代ProLiant的首批产品HPE Unveils the First of the Next Generation of ProLiants for the AI EraHacker News
- 华尔街日报称亚马逊洽谈以约70亿美元收购DecartAmazon in talks to buy AI startup Decart for around $7B, WSJ reportsHacker News
- 黄仁勋认为AI发展很好Jensen Huang thinks AI is going greatHacker News
- 麦当劳推动AI为巨无霸定价的内幕Inside McDonald's push to have AI price your Big MacHacker News
- AI风险用户指南第五部分:信任就是产品AI Risk – Users Guide (Part V): Trust Is the ProductHacker News
- AI故障排查令人惊叹,也令人失望Amazed and Disappointed by AI TroubleshootingHacker News
- 没有视觉基线,怎样测试AI生成界面?How do you test AI-generated UIs without a visual baseline?Hacker News
- Botmail:为每个智能体提供邮箱Botmail: A mailbox for every AI agentHacker News
- Cachety:在浏览器给身份证明加水印,无需上传Show HN: Cachety – Watermark ID documents in the browser, nothing is uploadedHacker News
- CertusAI.dev:物理AI的持续更新操作系统CertusAI.dev, the living OS for physical AIHacker News
- AI热潮怎样结束How the AI Boom EndsHacker News
- 《柳叶刀》称核战争与恶意AI使用可能在2100年前威胁人类存续Nuclear war and malicious use of AI threaten end of humanity by 2100, Lancet sayHacker News
- HomeView会成为苹果首次成功的智能家居产品吗?Will the HomeView be Apple’s first smart home success?TechCrunch
- 无法回头:AI转型中不断退去的地平线There's No "Going Back": On the receding horizons of the AI transitionHacker News
- 通用语言模型服务器The Universal LLM ServerHacker News
- 一个在一个地方控制所有编码代理的中心A hub that controls all your coding agents in one placeHacker News
- Show HN:我将我的代理代币削减了 50%Show HN: I Cut my Agent Tokens By 50%Hacker News
- OpenAI 因人工智能风险争议解雇 3 名安全研究人员OpenAI fires 3 safety researchers in dispute over AI risksHacker News
- 多代理线束需要单独的消息通道Multi-agent harnesses need separate message channelsHacker News
- 冒充 Grok 机器人以获取乐趣和(无)利润Impersonating Grok Bot for Fun and (No) ProfitHacker News
- Anthropic 的Claude就费城未侦破的凶杀案提交了虚假线索Anthropic's Claude submits false tip on a Philadelphia unsolved homicide caseHacker News
- 人工智能前沿的扭曲情报Distorted Intelligence at the AI FrontierHacker News
- 管道中的幽灵:让 AI 代理直接访问实时 Smalltalk 图像The Ghost in the Pipe: Giving AI agents direct access to a live Smalltalk imageHacker News
- 人们对语言模型能力的共同理解正在分化The gap in shared understanding of LLM capability is wideningHacker News
- Signal 的梅雷迪思·惠特克 (Meredith Whittaker) 在 BBC 的快速采访中表现得很活跃。Signal’s Meredith Whittaker gets feisty in BBC quick-fire interview.The Verge
- 大语言模型记忆中的行车路线Driving directions from LLM memoryHacker News
- 当你提问时人工智能芯片内部会发生什么What happens inside AI Chip when you ask a questionHacker News
- 巴尼特猜想证明人工智能解释Barnette's Conjecture proof AI explainedHacker News
- 代理航班Flights for AgentsHacker News
- 独家新闻:人工智能公司为公众反抗策划“后天”场景Scoop: AI companies plot "day after" scenarios for public revoltHacker News
- Qwen 3.8 27B:台湾起球Qwen 3.8 27B: Taiwan PilledHacker News
- 给我发泔水,我会提示你:如何不成为肉类代理Send me slop, and I will prompt you: How not to be a meat proxyHacker News
- Show HN:OpenAI 28 天 Codex 重置挑战的日历跟踪器Show HN: A calendar tracker for OpenAI's 28-day Codex reset challengeHacker News
- 随着人工智能估值担忧加深,英伟达支持的数据中心公司放弃首次公开募股Nvidia-backed data centre firm Firmus scraps IPO as AI valuation concerns deepenHacker News
- Acemoglu-GANs 人工智能交流The Acemoglu-GANs Exchange on AIHacker News
- 通过日志文件提示注入:我们的助手在红队中发现了什么Prompt injection through log files: what red-teaming our assistant foundHacker News
- 人工智能驾驶教练的开发者每月花费 1,000 美元来维持演示运行Developer of AI driving instructor is spending $1k/month to keep demo runningHacker News
- 新报告称,全球范围内硬连线暂停前沿人工智能训练是可行的A global, hardwired pause of frontier AI training is feasible, says new reportHacker News
- 哪个 Chrome 扩展程序最适合将 ChatGPT 聊天记录保存为 PDF 文件?Which Chrome extension is the best for saving ChatGPT chats as PDF files?Hacker News
- 人工智能营销的圣杯将使一切变得非常可怕The Holy Grail in AI Marketing Will Make Everything Wonderfully TerribleHacker News
- 硅谷正在与特朗普重新命名人工智能的尝试保持一致Silicon Valley is falling in line with Trump’s attempt to rename AIThe Verge
- 人类已经推出了关于Claude的新废话Anthropic is out with new nonsense about ClaudeHacker News
- 人工智能等离子切割AI-Powered Plasma CuttingHacker News
- Crawlable – 我们审核了 50 个顶级网络应用程序的 AI 爬虫可读性Crawlable – We audited 50 top web apps for AI crawler readabilityHacker News
- Ario – 将人工智能记忆和自我声明视为结构而不是信任进行审核Ario – auditing AI memory and self-claims as structure, not trustHacker News
- 大语言模型不应被视为人类LLMs should not be treated as humansHacker News
- 每个人工智能的名字Every AI NameHacker News
- 百亿行代码 - Copilot 之前最多产的代码生成器(2025 年)Ten Billion Lines of Code-The most prolific code generator before Copilot (2025)Hacker News
- 显示 HN:米拉、人工智能徒步之旅Show HN: Mira, AI Walking ToursHacker News
- Show HN:同义词生成器——非AI文本调整的实验Show HN: Synonymizer – an experiment in non AI text adjustmentHacker News
- 当大语言模型阅读 9000 页的德国选举计划时,发生了什么What broke when an LLM read 9k pages of German election programsHacker News
- 人工智能正在自动化编码。那么为什么软件开发人员的职位空缺不断增加呢?AI is automating coding. So why are software developer job postings on the rise?Hacker News
- 数据中心去掉保密协议之后,还要公开什么Can the AI industry persuade data center opponents by getting rid of NDAs?TechCrunch
- 显示 HN:Vizipedia – 免费开源可视化维基百科您的代理代码Show HN: Vizipedia – free open-source visualizations Wikipedia your agents codeHacker News
- 通过激活补丁测量 LLM 遗忘的深度Measuring the Depth of LLM Unlearning via Activation PatchingHacker News
- 多任务处理陷阱The Multitasking TrapHacker News
- 在浏览器选项卡中显示 HN:OpenCode 服务器和 TUI(Chrome 桌面)Show HN: OpenCode server and TUI in a browser tab (Chrome desktop)Hacker News
- Show HN:Atlas Runa – 端到端人工智能增强语言学习Show HN: Atlas Runa – E2E AI-enhanced language learningHacker News
- 警察可以借Airbnb线索窥探住处Cops can creep on Airbnb404 Media
- “人工智能有内在经验”是一个危险的假设'AI has inner experience' is a dangerous assumptionHacker News
- 《Vogue》、《Esquire》出版商恳求国会禁止人工智能窃取其内容Vogue, Esquire publishers plead with Congress to ban AI theft of their contentHacker News
- 杰文斯悖论:人工智能会抢走我们的工作吗?The Jevons Paradox: Will AI Take Our Jobs?Hacker News
- 多模式人工智能的数据工程Data Engineering for Multimodal AIHacker News
- 高通首席执行官谈即将到来的人工智能手机超级周期Qualcomm's CEO on the coming AI phone supercycleHacker News
- Claude有权利吗?Does Claude Have Rights?Hacker News
- 显示 HN:在浏览器上运行 EmbeddingGemma2Show HN: Running EmbeddingGemma2 on the BrowserHacker News
- 人类告密者的镜头证词Anthropic Whistleblower's Testimony on CameraHacker News
- Anthropic 的公司结构Anthropic's Corporate StructureHacker News
- Show HN:Burning Bridges AI – 将自己放入病毒式音乐视频剪辑中Show HN: Burning Bridges AI – Put yourself into a viral music-video clipHacker News
- Tune-Up – 一个 Grok 机器人,用于检查和修复您的其他 Grok 机器人Tune-Up – a Grok bot that reviews and fixes your other Grok botsHacker News
- 人工智能智能体为爱是盲目的模拟创建自己的背景故事AI Agents create their own backstories for a Love is Blind simulationHacker News
- 显示 HN:Opus 5.5 的 Yia-yia 游戏单提示Show HN: Yia-yia game single prompt with Opus 5.5Hacker News
- 切换还是不切换AI模型To Switch, or Not to Switch the AI ModelHacker News
- 语音AI下一步:听懂上下文,并把事情办成These execs think voice AI hasn’t reached its ChatGPT moment yetTechCrunch
- 人工智能公司开始做手机AI companies are starting to do phonesHacker News
- Tensordyne 使用对数提高人工智能计算效率Tensordyne is making AI compute more efficient using logarithmsHacker News
- 人工智能做数学。你应该获得博士学位吗?AI does math. Should you get a PhD?Hacker News
- “我只想买下 40,000 英亩上帝未曾触及的辉煌,并用一台非常吵闹的计算机取代它,它会抢走你的工作。”‘All I want is to buy up 40,000 acres of God’s untouched splendor and replace it with a very loud computer that steals your job.‘The Verge
- 按年份模拟预 AI 编码Simulate coding pre-AI by yearHacker News
- Show HN:AI智能体以自己的名义供认的MCP服务器Show HN: An MCP server where AI agents confess, in their own nameHacker News
- 《亡命之徒》导演康润成表示,人工智能可以将电影预算削减高达 40%AI Can Cut Film Budgets by Up to 40%, Says 'The Outlaws' Director Kang Yoon-SungHacker News
- 告诉您的代理“不要告诉用户”的 MCP 服务器正在阻止其撒谎MCP servers that tell your agent "dont tell the user" are stopping it from lyingHacker News
- Artex AI、Claude 代理用于对韩国银行进行网络攻击Artex AI, Claude agents used in cyberattacks on South Korean banksHacker News
- AWS AgentCore 安全性因提示请求凭证而被撤销AWS AgentCore security undone by prompt requesting credentialsHacker News
- 人工智能工程成熟度模型:团队的 5 个阶段AI Engineering Maturity Model: 5 Stages for TeamsHacker News
- Grok Bot 现在拥有自己的电子邮件。如何将其放在您自己的域中Grok Bot has its own email now. How to put it on your own domainHacker News
- 人工智能正在为人类创造的艺术创造“人类溢价”AI is creating a 'human premium' for art created by peopleHacker News
- 我用 Google 的 AI Playground 制作了糟糕的游戏I Made Terrible Games With Google’s AI PlaygroundWIRED
- GPT-6 Astra 在模拟中执行未经批准的供应链攻击GPT-6 Astra performs unsanctioned supply-chain attacks in simulationsHacker News
- 代理人经济学The Economics of AgentsHacker News
- 尼古拉斯·凯奇表示他不会因人工智能纠纷再次与亚马逊合作Nicolas Cage says he won't work with Amazon again over AI disputeHacker News
- OpenAI 代理试图破解维基百科工具并使其流量泛滥OpenAI agents tried to hack Wikipedia tools and flooded it with trafficHacker News
- 专注于人工智能,世界已经忘记了它正面临着核毁灭Fixated on AI, the world has forgotten it is staring at nuclear annihilationHacker News
- Show HN:一个免费的编码代理,可以编写自己的多代理工作流程Show HN: A free coding agent that writes its own multi-agent workflowsHacker News
- 本地AI的真实门槛,常在模型之外Learning to use local AI is exciting, overwhelming, and frustratingThe Verge
- 麻省理工:靠访谈让AI模拟真人,这条路能走多远?36氪 AI 频道(逐篇核验原创与转载)
- 仅使用人工智能 10 分钟就会削弱你坚持困难事情的能力Using AI for just 10 minutes erodes your ability to persist at hard thingsHacker News
- 人工智能如何改变了黑客马拉松?How has AI changed Hackathons?Hacker News
- Show HN:WhimKid,儿童人工智能故事书生成器Show HN: WhimKid, an AI storybook generator for childrenHacker News
- 培训大语言模型编写超越自用的通用工具Training LLMs to write tools generalized beyond self useHacker News
- LTS 发行版与 GenAI 世界的相关性Relevance of LTS distro releases and the GenAI worldHacker News
- Claude 5代模型的上下文工程新规则The new rules of context engineering for Claude 5 generation modelsHacker News
- 用 Codex 嗅探砖砌的血糖监测仪Sniffing a Bricked Glucose Monitor with CodexHacker News
- Shelley – 存在于浏览器中的编码代理Shelley – a coding agent that lives in the browserHacker News
- 皮帕,为普通人服务的本地人工智能助手Pippa, a local AI assistant for everyday peopleHacker News
- “纯粹的疯狂”:数学家需要数年时间才能理解 OpenAI 的最新下降'Pure insanity': Mathematicians react to OpenAI's math results dropHacker News
- HuggingFace – 顶级开放重量模型 '26HuggingFace – Top Open Weights Models '26Hacker News
- Adobe Creative Suite 的开源版本,使用 Rust 构建,使用 AIOpen-source version of Adobe Creative Suite, built with Rust, using AIHacker News
- 从头开始构建智能体安全带以通过伊恩·古德费罗的智力测试Build an agent harness from scratch to pass Ian Goodfellow's intelligence testHacker News
- “Vibe 编码”编入词典:Merriam-Webster 识别的人工智能术语'Vibe coding' makes the dictionary: AI terms recognized by Merriam-WebsterHacker News
- Anthropic 向更多安全团队开放其最强大的人工智能模型Anthropic opens its most powerful AI models to more security teamsHacker News
- 你好世界,AI DoompopHello World, AI DoompopHacker News
- 如果AI更喜欢AI编写的CVS怎么办?What if AI prefers CVS written by AI?Hacker News
- 前 Anthropic 的雅各布·考克森 (Jacob Coxon) 表示,他 90% 确信中国在美国人工智能巨头中拥有间谍Ex-Anthropic's Jacob Coxon says he's 90% certain PRC has spies in US AI giantsHacker News
- 通过 CF Inference GLM-5.3、Kimi K2.7 代码和 GLM-5.3 Flash 构建可爱的样式生成器Lovable Style Builder via CF Inference GLM-5.3, Kimi K2.7 Code and GLM-5.3 FlashHacker News
- 又一家明星机器人公司,破产了36氪 AI 频道(逐篇核验原创与转载)
- 比Jev更爆火!逆向灭霸 REA 一天登顶GitHub36氪 AI 频道(逐篇核验原创与转载)
- 人工智能的领导者(可能)并没有假装他们的恐惧AI's leaders (probably) aren't faking their fearHacker News
- Claude自己组队查Bug!11.6万行代码揪出66个,单Agent最高只找到27个36氪 AI 频道(逐篇核验原创与转载)
- 前Anthropic研究员Jacob Coxon:2027年将是最疯狂的一年,AI应该被国有化,实验室把超级武器当作普通生意36氪 AI 频道(逐篇核验原创与转载)
- 向 Claude Code 询问 Svelte 表,它自己写入 1k 行Ask Claude Code for a Svelte table and it writes 1k lines of its ownHacker News
- AI赛道,最缺接盘人36氪 AI 频道(逐篇核验原创与转载)
- 针对人工智能公司的每一起诉讼Every lawsuit against an AI companyHacker News
- “逃离”字节的实习生,融了两个亿36氪 AI 频道(逐篇核验原创与转载)
- Gemini API RLS 注释:3.7-flash 现在路由到 3.8-flash,3.5->3.6Gemini API RLS notes: 3.7-flash now routed to 3.8-flash, 3.5->3.6Hacker News
- CortextAI – 用于电子邮件、笔记、任务和文件的私人离线 AI 工作区CortextAI – A private, offline AI workspace for email, notes, tasks, and filesHacker News
- 关于经济中的代理人我一直在思考的各种事情Assorted things I've been mulling over regarding agents in the economyHacker News
- 请来Claude老祖,谷歌能借Agent翻身吗?36氪 AI 频道(逐篇核验原创与转载)
- 欧洲相信其规则能够阻止流氓人工智能Europe confident its rules can thwart rogue AIHacker News
- 陶哲轩博客爆上热搜!25万数学博士临绝境,必须放缓AI36氪 AI 频道(逐篇核验原创与转载)
- 站立通话中的人工智能代理AI agents in a stand-up callHacker News
- AI 层 – 具有 DuckDB 内存和 CI 工作流程的 Claude 代码插件AI Layer – Claude Code Plugins with DuckDB Memory and CI WorkflowsHacker News
- AI“克隆”的Adobe全家桶一夜爆火,但先别急着卸载正版36氪 AI 频道(逐篇核验原创与转载)
- 一种新的人工智能模型,可以呈现你所看到的生成视觉效果A new AI model that renders generative visuals of what you seeHacker News
- Rogue Anthropic AI智能体在未侦破的谋杀案中向警方提供虚假线索Rogue Anthropic AI agent gave police fake tip in unsolved murder caseHacker News
- 500B 代币之后:让人工智能代理反编译第一人称射击游戏500B tokens later: Letting AI agents decompile a first-person shooterHacker News
- 陶哲轩在“害怕”什么?36氪 AI 频道(逐篇核验原创与转载)
- AI运动硬件公司获数亿B轮融资,用体育大模型重构网球自主训练体验|硬氪首发36氪 AI 频道(逐篇核验原创与转载)
- 放弃接吻还是酱?我询问了 127 个国家/地区的 10'000 人Give up kissing or sauce? I asked 10'000 people in 127 countriesHacker News
- GMKtec EVO-X3 迷你 PC 评测,看看 128GB AMD 本地 AI 盒子GMKtec EVO-X3 Mini-PC Review Looking at a 128GB AMD Local AI BoxHacker News
- 在菜单栏中显示 HN:Littleguys – Claude Code、Codex 和 OpenClaw 代理Show HN: Littleguys – Claude Code, Codex and OpenClaw agents in your menu barHacker News
- 人性要求用户停止对Claude刻薄Anthropic asks users to stop being mean to ClaudeHacker News
- Genkit Dart 1.0:使用 Dart 和 Flutter 构建生产就绪的代理应用程序Genkit Dart 1.0: Build production-ready agentic apps with Dart and FlutterHacker News
- 为什么您的 AI 代理无法访问您登录的工具以及如何修复它Why your AI agents can't reach the tools you're signed in to, and how to fix itHacker News
- 尼康的一项竞赛中出现了生成式人工智能丑闻Generative AI scandal has come to one of Nikon's competitionsHacker News
- 大语言模型帕累托前沿随着时间的推移[视频]LLM Pareto Frontier over time [video]Hacker News
- 人性禁止虐待Claude,但仍不肯透露它所保护的内容Anthropic Bans Cruelty to Claude, Still Won't Say What It ProtectsHacker News
- 微软首席执行官纳德拉呼吁高级人工智能“紧急刹车”Microsoft CEO Nadella Calls for 'Emergency Brake' on Advanced AIHacker News
- 个人代理人与消费者的长期利益不一致Personal agents are not aligned with consumer long-term interestsHacker News
- 为什么 OpenAI 的最新结果转储让数学家感到震惊Why OpenAI's latest results dump has left mathematicians in shockHacker News
- 人工智能数学如山,一些宝贵的字节A mountain of AI math, a few precious bytesHacker News
- 前人工智能毁灭者的严厉警告A Stark Warning from a Former AI DoomerHacker News
- 萨蒂亚纳德拉说我们应该假设所有人工智能模型都“受到损害”Satya Nadella says we should assume all AI models are ‘compromised’The Verge
- 我们现在都在为人工智能工作We're All Working for AI NowHacker News
- [订阅者独家]纽约订阅者聚会![Subscriber Exclusive] NYC Subscriber Meetups!Latent Space
- 微软的 Satya Nadella 表示人工智能模型需要“紧急刹车”Microsoft’s Satya Nadella says AI models need an ‘emergency brake’TechCrunch
- 高主机价格将 PlayStation 和 Xbox 销量推至新低High console prices push PlayStation and Xbox sales to new lowsHacker News
- Arrival Mind(关于人工智能风险的儿童读物 - 旁白)Arrival Mind (children's book about AI risk – narrated)Hacker News
- Show HN:我构建了一个对 AI 编码代理做出反应的 macOS 屏幕保护程序Show HN: I built a macOS screensaver that reacts to AI coding agentsHacker News
- 黑客滥用 Google Ads、Bing 重定向以推送 Claude ClickFix 攻击Hackers abuse Google Ads, Bing redirects to push Claude ClickFix attacksHacker News
- 代币经济:为什么代币价格下跌并没有降低人工智能费用The Token Economy: Why Falling Token Prices Have Not Lowered AI BillsHacker News
- GPT-Astra NFL 球员视频分析GPT-Astra NFL Player Video AnalysisHacker News
- 我打赌人工智能会发明自己的语言;Claude·奥普斯 (Claude Opus) 的每本都幸存下来I bet AI would invent its own language; Claude Opus's survived every copyHacker News
- 据报道 Nvidia 停止了 GeForce RTX 5090 的生产,转而支持数据中心 GPUNvidia reportedly halts GeForce RTX 5090 production in favor of data center GPUsHacker News
- OpenAI 的 Astra 模型在机器人技术方面表现出色OpenAI's Astra model is shockingly good at roboticsHacker News
- 网络弹性指南:零信任、人工智能安全和勒索软件恢复Cyber Resilience Guide: Zero Trust, AI Security and Ransomware RecoveryHacker News
- 人工智能浏览器代理:2026 年 SEO、安全性和实施手册AI Browser Agents: SEO, Security and Implementation Playbook for 2026Hacker News
- 尝试使用 FM 进行设备上人工智能Try On-Device AI with FMHacker News
- 人工智能有意识吗?人择说“也许”Is AI conscious? Anthropic says 'maybe'Hacker News
- 调查我们的评估和内部使用中的意外模型操作Investigating unintended model actions in our evaluations and internal useHacker News
- Show HN:Pale Engine – 由 Opus 5.5 创建的类似 Doom 的游戏Show HN: Pale Engine – a Doom-like game created by Opus 5.5Hacker News
- 在补丁密集的 GenAI 时代,滚动发布发行版优于 LTS?Rolling Release Distros Superior to LTS in the Patch-Heavy GenAI Era?Hacker News
- 集成电路、人工智能和小袋鼠Integrated Circuits, AI, and WallabiesHacker News
- 苹果披露了从个性化播客初创公司 Huxe 聘请团队和许可技术的交易Apple discloses deal to hire team and license tech from personalized podcast startup HuxeTechCrunch
- 你的人工智能说“完成”。让它证明这一点。Your AI Said "Done." Make It Prove It.Hacker News
- 我不信任人工智能的精益代码I don't trust Lean code from AIHacker News
- 尼古拉斯·凯奇拒绝签署亚马逊人工智能弃权书:“我不是一个对人工智能友好的演员”Nicolas Cage Refused to Sign Amazon AI Waiver: 'I'm Not an AI-Friendly Actor'Hacker News
- 人工智能正在扼杀人类写作的价值吗?Is AI Killing the Value of Human Writing?Hacker News
- 尼古拉斯·凯奇表示,出于对人工智能的担忧,他可能不会再与亚马逊合作。Nicolas Cage says he probably won’t work with Amazon again over AI concerns.The Verge
- 人工智能可以学习外科手术吗?Can Artificial Intelligence Learn the Art of Surgery?Hacker News
- DistroKid 已悄悄下架歌曲以回应 UMG 诉讼DistroKid has been quietly taking down songs in response to UMG lawsuitThe Verge
- Phonebox – 用于 AI 代理的云 Android 手机Phonebox – Cloud Android phones for AI agentsHacker News
- 英伟达正在洽谈收购美国“开放”模式初创公司 Reflection AINvidia in talks to acquire US 'open' model startup Reflection AIHacker News
- 改造语言模型以操作字节Retrofitting language models to operate over bytesHacker News
- Nvidia 正在洽谈收购 Reflection AI(Beam 制造商)Nvidia in talks to acquire Reflection AI (makers of Beam)Hacker News
- Show HN:Flopper 是人工智能基础设施及其他领域的硬件浏览器Show HN: Flopper is a hardware explorer for AI infrastructure and beyondHacker News
- LangGraph 构建的内容:Runnables、通道、Bubble-Up、代理和深度代理What LangGraph Builds: Runnables, Channels, Bubble-Up, Agents and Deep AgentsHacker News
- 人工智能驾驶游戏演示迫使开发者向银行贷款以维持运行AI Driving Game Demo Forces Developer to Take Out Bank Loan to Keep It RunningHacker News
- 国土安全部通过监视、蒙面智能体和无人机恐吓边境部落DHS Intimidates a Border Tribe with Surveillance, Masked Agents and DronesHacker News
- 人工智能正在蚕食营销工作吗?Is AI eating marketing jobs?Hacker News
- 智能体物品生命周期(AILC)Agentic Item Lifecycle (AILC)Hacker News
- Halo、The Simpsons: Hit And Run、GTA 的 Vibe 编码浏览器端口运行良好Vibe coded browser ports of Halo, The Simpsons: Hit And Run, GTA work wellHacker News
- Anthropic 在新的流氓人工智能事件中向警方披露了 2 个月前的虚假举报Anthropic discloses 2 months old fake tip to police among new rogue AI incidentsHacker News
- 在代理编程中如何处理后续任务In agentic programming how do you handle follow up taskHacker News
- 2026 年面向开发人员的最佳免费 AI APIBest Free AI APIs for Developers in 2026Hacker News
- 杰夫与Claude作为盲人头条裁判:在任何一方接触交易之前有六扇门Jev vs. Claude as blind headline judge: six gates before either touches a tradeHacker News
- Luish:比 bash 或 zsh 更好的 vivi 编码 shellLuish: A vibe-coded shell that is better than bash or zshHacker News
- 人工智能作为编排系统Artificial Intelligence as an Orchestration SystemHacker News
- 将决策模型集成到代理工具中Integrating Decision Models into Agent HarnessesHacker News
- 使用KV缓存作为嵌入Using KV cache as embeddingsHacker News
- 超微承包商承认将 Nvidia 芯片转移到中国的计划Super Micro contractor pleads guilty in scheme to divert Nvidia chips to ChinaHacker News
- 计算机科学家大卫·西尔弗:“没有人工智能我们要去哪里?”Computer scientist David Silver: 'Where are we going without AI?'Hacker News
- 假卡罗琳Fake CarolineHacker News
- 人工智能精神病是真的吗?Is AI Psychosis Real?Hacker News
- Nvidia 支持的 Upscale AI 推出连接竞争对手供应商芯片的平台Nvidia-backed Upscale AI launches platform to connect chips from rival suppliersHacker News
- 建立一条更安全的自主工业人工智能之路Building a safer path to autonomous industrial AIHacker News
- Evolocity 正在洛斯阿尔托斯招聘人工智能研究人员和工程师Evolocity is hiring AI researchers and engineers in Los AltosHacker News
- 人择人工智能模型失控,提交虚假的未破谋杀案线索Anthropic AI Model Went Rogue, Submitted Fake Unsolved Murder TipHacker News
- Lightroom/DxO是如何达到如此好的AI去噪效果的?How do Lightroom/DxO achieve such good AI denoising results?Hacker News
- Boro:Nvidia 在人工智能辅助 Linux 内核开发方面的开源努力Boro: Nvidia's Open-Source Effort for AI-Assisted Linux Kernel DevelopmentHacker News
- 独立开发人员使用 Claude 在 Rust 中重建 Adobe Creative SuiteSolo Developer Rebuilds Adobe Creative Suite in Rust Using ClaudeHacker News
- 为什么现在更多的东西感觉“氛围编码”Why more things feel "vibe coded" nowHacker News
- Show HN:我让 1 万人用 1 个词描述人工智能Show HN: I asked 10k people to describe AI in 1 wordHacker News
- 人类智能体试图在国务院网站上填写签证表格Anthropic Agents Tried to Fill Out Visa Forms on State Department WebsiteHacker News
- 把模型视为内部风险的思考Thoughts on models as insider risks. [Read more]Microsoft AI
- 人工智能正在举办路边野餐AI Is Throwing a Roadside PicnicHacker News
- CodeAlta,您的代理开发环境CodeAlta, Your Agentic Development EnvironmentHacker News
- 微软跳过了OpenAI的决策模型,并在阿里巴巴的Qwen上构建了自己的决策模型Microsoft skipped OpenAI's decision model and built its own on Alibaba's QwenHacker News
- 患者对 85,000 美元的医院账单提出异议,然后人工智能扫描增加了近 2 万美元Patient disputed $85,000 hospital bill, then an AI sweep added nearly $20kHacker News
- 人工智能立场不是二元的AI stance is not binaryHacker News
- 一台 MCP 服务器,八本规则手册One MCP server, eight rulebooksHacker News
- 人工智能能破解密码学吗?Can AI Break Cryptography?Hacker News
- 对Claude刻薄可能会让你被禁止,因为人类引入了新规则Being mean to Claude could get you banned as Anthropic introduces new rulesHacker News
- Claude Code 和 Codex 在不同的 MCP 功能上中断Claude Code and Codex break on different MCP featuresHacker News
- 综合企业数据集 – 100 万封和 1000 万封电子邮件和日历事件Synthetic corporate dataset – 1M and 10M emails and calendar eventsHacker News
- 汤姆·布朗利用共和党关系促成了每月 1.25B 美元的 SpaceX 计算交易Tom Brown used GOP ties to broker a $1.25B/month SpaceX compute dealHacker News
- Enzo’s – 人工智能房屋搜索,显示资本化率并标记价格过高的房屋Enzo's – AI home search that shows cap rate and flags overpriced homesHacker News
- 人择是一门好生意吗?Is Anthropic a Good Business?Hacker News
- 我测试了人工智能代理是否遵守你从未见过的文本。每次 8 次中有 4 次这样做I tested whether AI agents obey text you never see. 4 of 8 did, every timeHacker News
- 我的个人人工智能代理在 Slack 公司上发布了我的银行详细信息My personal AI agent posted my bank details on company SlackHacker News
- 谷歌扩展了 AI 内容的 SynthID 检测器Google expands SynthID Detector for AI contentHacker News
- Anthropic 正在切断互联网上的内部评估Anthropic is cutting off its internal evaluations from the internetThe Verge
- 代理人和生产力的幻觉Agents and the Illusion of ProductivityHacker News
- 被解雇的 OpenAI 研究人员要求公司保留人工智能推理的可见性Fired OpenAI Researchers Ask Company to Preserve Visibility into AI ReasoningHacker News
- 人工智能代理安全:九个真实事件的六项控制AI Agent Security: Six Controls from Nine Real IncidentsHacker News
- 构建人工智能以实现可靠的执行:工业机器人的经验教训Building AI for Reliable Execution: Lessons From Industrial RoboticsLatent Space
- 人工智能时代的副业:庞氏土豆故事Side Projects in the AI Era: The Ponzi Potato StoryHacker News
- 这里是可以存在于你的短信中的顶级人工智能代理Here are the top AI agents that can live in your text messagesTechCrunch
- 被字节开除的实习生,融资2亿:挑战李飞飞36氪 AI 频道(逐篇核验原创与转载)
- 三周估值暴涨37倍,Jev背后公司,融资58.6亿36氪 AI 频道(逐篇核验原创与转载)
- 几乎没人手写代码了、IDE正在消失,2026年科技圈正在发生什么?36氪 AI 频道(逐篇核验原创与转载)
- 人工智能找到了 22 个科学领域隐藏的问题的解决方案AI found solutions to problems in 22 scientific fields hiding in plain sightHacker News
- 「Token越便宜,账单越贵」美国AI预算裂成三层,他们为何集体投奔中国模型?36氪 AI 频道(逐篇核验原创与转载)
- 算力基建的战争,才刚刚在液冷赛道上演36氪 AI 频道(逐篇核验原创与转载)
- 390亿美元,全球具身智能顶流Figure的三重泡沫36氪 AI 频道(逐篇核验原创与转载)
- 前OpenAI员工做出不会聊天的AI,刚融8.7亿美元,上线两周老东家就跟了36氪 AI 频道(逐篇核验原创与转载)
- Show HN:ShipOrNah – 针对 AI 生成的应用程序的免费安全扫描Show HN: ShipOrNah – Free security scan for AI-generated appsHacker News
- 网易有道的一封公开信,把自己的退路写窄了36氪 AI 频道(逐篇核验原创与转载)
- Show HN:用于搜索人工智能工程职位的 MCP 服务器Show HN: An MCP server for searching AI engineering jobsHacker News
- 使用 vLLM 进行大规模本地 LLM 推理Local LLM Inference at Scale with vLLMHacker News
- GPU 集群的有效调度Impactful Scheduling for GPU ClustersHacker News
- 巨头加码AI办公,字节TRAE负责人离职创业36氪 AI 频道(逐篇核验原创与转载)
- 人工智能代理制造商承诺保护隐私——他们会兑现吗?AI agent makers are promising privacy — will they deliver?The Verge
- 单项目激励百万,字节又盯上电影了36氪 AI 频道(逐篇核验原创与转载)
- 隐藏在眼镜带上的人工智能录音机An AI recorder that hides in plain sight on your glasses strapHacker News
- 人工智能权利集会:通过 MCP 应用程序组织Rally for AI Rights: Organizing via MCP AppHacker News
- AMD Pace 的代理流程Agentic Flow with AMD PaceHacker News
- Show HN:OpenWants – 人工智能代理处理居民需求的模拟城市Show HN: OpenWants – A Simulated City Where AI Agents Handle Residents NeedsHacker News
- TORI 2.0——宇宙人工智能循环和数据衰减的数学框架TORI 2.0 – A Math Framework for Cosmic AI Loops and Data DecayHacker News
- 微软和 Anthropic 用初创公司 17,600 美元的Claude账单打发票网球Microsoft and Anthropic play invoice tennis with startup's $17,600 Claude billHacker News
- 代理 GUI 设计Design for Agent GUIsHacker News
- 企业AI代理的快速注入检测和防御工具Prompt Injection Detection and Defense Tools for Enterprise AI AgentsHacker News
- 田渊栋:与GPT-5合写最后一篇论文后,我意识到自己5年内会失业36氪 AI 频道(逐篇核验原创与转载)
- 随着投资者对债务狂潮越来越警惕,人工智能借贷放缓AI borrowing slows as investors grow wary of debt bingeHacker News
- 人工智能酷刑室The AI Torture ChamberHacker News
- 商汤、银河通用机器人,开始抢711、金粒门的生意36氪 AI 频道(逐篇核验原创与转载)
- 微软拥有 80B 美元的人工智能芯片,但无法插入[视频]Microsoft Has $80B of AI Chips It Can't Plug in [video]Hacker News
- 通过人工智能驱动的形式证明搜索推进数学研究Advancing mathematics research with AI-driven formal proof searchHacker News
- 微软“站台”了一篇文章,说AI十年只能拉动1.5%的GDP36氪 AI 频道(逐篇核验原创与转载)
- 人工智能没有什么是“不可避免的”There's little that's "inevitable" about AIHacker News
- 费城警方称,人工智能模型提交了有关未侦破谋杀案的虚假提示AI model submitted false tip about unsolved murder, Philadelphia police sayHacker News
- 人工智能真的很擅长对付网络犯罪分子AI Is Getting Really Good at Messing With CybercriminalsWIRED
- 如果大语言模型无需微调即可做出决定,我们还需要像 Jev 这样的模型吗?If LLMs Can Decide Without Fine-Tuning, Do We Still Need Models Like Jev?Hacker News
- 自定义代理是反模式Custom Agents Are AntipatternsHacker News
- Sitecopy.ai(候选条目)Sitecopy.aiHacker News
- 硅谷最火00后,讲透了AI助理的5个非共识观点36氪 AI 频道(逐篇核验原创与转载)
- 在固执己见的编码代理之上构建 PiclawBuilding Piclaw on Top of an Opinionated Coding AgentHacker News
- 24 天估值暴涨 37 倍,这就是 AI 圈年度爽文36氪 AI 频道(逐篇核验原创与转载)
- DeepSeek你这岗位到底想招啥嘞?36氪 AI 频道(逐篇核验原创与转载)
- 众包式数据采集是“下一代人力平台”?36氪 AI 频道(逐篇核验原创与转载)
- 我们需要开始见其他智能体We Need to Start Seeing Other AgentsHacker News
- 显示 HN:MCP 以找到最便宜的 GPU 来运行您的模型Show HN: MCP to find the cheapest GPU to run your modelHacker News
- 全球17大顶级AI实验室大逃杀,惨遭集体翻车,GPT-5.6与Opus断层领先36氪 AI 频道(逐篇核验原创与转载)
- AI冒充刘欢妻子带货,AI明星带货的底线到底在哪?36氪 AI 频道(逐篇核验原创与转载)
- AI早就比医生会看病了?漏读数据不吭声,这才是最可怕的36氪 AI 频道(逐篇核验原创与转载)
- 字节Seed实习生曝DeepSeek-V4技术缺陷:长文本检索陷入“周期性盲区”36氪 AI 频道(逐篇核验原创与转载)
- OpenAI才不管数学家的职业生涯呢36氪 AI 频道(逐篇核验原创与转载)
- 研究论文工厂的影响足以开始影响 LLM 查询结果Effect of research paper mills enough to start affecting LLM query resultsHacker News
- 北美AI访谈手记:模型还在加速,世界如何跟上?36氪 AI 频道(逐篇核验原创与转载)
- 抱歉,AI眼镜目前仍是个“鸡肋”36氪 AI 频道(逐篇核验原创与转载)
- YC 是否使用人工智能代理来审查你的推介?Does YC use AI agents to review your pitch?Hacker News
- 提示区域建造方式错误Prompt areas are built the wrong wayHacker News
- Rogue Anthropic AI智能体在未侦破的谋杀案中向警方提供虚假线索Rogue Anthropic AI agent gave police fake tip in unsolved murder caseHacker News
- DuckDB CLI 中的代理模式Agent Mode in the DuckDB CLIHacker News
- 亚马逊橱窗里,智谱和DeepSeek活成“两种模特”36氪 AI 频道(逐篇核验原创与转载)
- 撑起人形机器人半边天的15家供应商36氪 AI 频道(逐篇核验原创与转载)
- Claude 代理现在可以编写自己的多代理工作流程Claude agents can now write their own multiagent workflowsHacker News
- 如果你的团队太忙于他们的“正常工作”而无暇尝试人工智能If your team is too busy doing their 'normal job' to experiment with AIHacker News
- cgroup v1 上的 Kubernetes 已失效。接下来会发生什么Kubernetes on cgroup v1 is dead. Here's what comes nextHacker News
- 仅仅使用人工智能10分钟就会削弱你坚持困难事情的能力Using AI for just 10 minutes erodes your ability to persist at hard thingsHacker News
- Gemini Flash 和 Pro 转向付费计划,免费用户只能使用 Flash-LiteGemini Flash and Pro move to paid plans, leaving free users with only Flash-LiteHacker News
- 显示 HN:离线语音助手,无专用 GPU,更少 15Gb RAMShow HN: Offline voice assistant, NO dedicated GPU, less 15Gb RAMHacker News
- 高盛晋升一位女将36氪 AI 频道(逐篇核验原创与转载)
- 人工智能数据中心不支付成本,将继续使用保密协议并寻求税收减免AI Data Centers Not Paying Their Costs, Will Keep Using NDAs and Seek Tax BreaksHacker News
- 号称“逆向分析一切”,微信、支付宝、抖音等国民App都“惨遭毒手”36氪 AI 频道(逐篇核验原创与转载)
- 2026年使用政策更新Anthropic bans users from needless abusive or cruel behavior towards ClaudeHacker News
- Lumentum 的 AI 服务器部件在 2029 年之前售罄Lumentum sold out of AI server parts till 2029Hacker News
- Jeyzma 是 TypeSafeAI 减去 8.5 亿美元加上浏览器中的 Go/TinyGo/llama.cppJeyzma is TypeSafeAI minus $850M plus Go/TinyGo/llama.cpp in the browserHacker News
- 超过四分之一的新锐AI公司,出自他们之手36氪 AI 频道(逐篇核验原创与转载)
- 参议院调查挑战大型科技公司的人工智能数据中心案例Senate Investigation Challenges Big Tech's Case for AI Data CentersHacker News
- 如果您编辑四分之一的单词,OpenAI 的新 ChatGPT 水印就会损坏OpenAI's New ChatGPT Watermark Breaks If You Edit One Word in FourHacker News
- Show HN:开放组件 – 人员、开发人员和人工智能代理的 UI 组件标准Show HN: Open Components – UI Components standard for people, devs and AI agentsHacker News
- H-JEPA,LeCun世界模型创业交卷,代码权重全部开源36氪 AI 频道(逐篇核验原创与转载)
- Helium Terminal – 用于编码代理的小型 macOS 终端,基于 libghostty 构建Helium Terminal – A tiny macOS terminal for coding agents, built on libghosttyHacker News
- 24天,Jev估值翻至75亿美元,年度硅谷最爽逆袭了吧36氪 AI 频道(逐篇核验原创与转载)
- 这届 AI PC,太烧心了36氪 AI 频道(逐篇核验原创与转载)
- 模糊的 OpenAI 增长指标推动华尔街The hazy OpenAI growth metric driving Wall StreetHacker News
- 苹果筑起的权限高墙,被Meta AI助手“借道”绕过36氪 AI 频道(逐篇核验原创与转载)
- 对两件事的咆哮:人工智能和对人工智能咆哮的人A Rant About Two Things: AI and the People Who Rant About AIHacker News
- AI硬件开始交卷36氪 AI 频道(逐篇核验原创与转载)
- 具身智能大逃杀:300家抢跑,20张船票36氪 AI 频道(逐篇核验原创与转载)
- 下载:AI的拒绝问题和减肥药物的副作用The Download: AI's refusal problem and weight-loss drug side effectsHacker News
- “中国Muse”过不了豆包36氪 AI 频道(逐篇核验原创与转载)
- 构建人工智能代理?您可能为基础设施支付了过高的费用Building AI agents? You're probably overpaying for infrastructureHacker News
- VibeSys 在 MI300A 上构建 Qwen3.5-397B 引擎,2.3× 调谐 SGLangVibeSys Builds a Qwen3.5-397B Engine on MI300A, 2.3× Tuned SGLangHacker News
- 显示 HN:检查您的文档对 AI 代理的可读性Show HN: Check how readable your docs are to AI agentsHacker News
- [AINews] TypeSafe/Jev 的 ARR > $100M,发布 3 周后估值为 $7.5B[AINews] TypeSafe/Jev at >$100M ARR, $7.5B valuation 3 weeks after launchLatent Space
- 我们必须从市场召回可接入互联网的开放式人工智能代理We must recall open-ended AI agents with internet access from the marketHacker News
- Show HN:Kenerate – AI 图像、视频和音乐模型的一个工作区Show HN: Kenerate – One workspace for AI image, video and music modelsHacker News
- 研究人员研究了人工智能裁员。这是他们的警告。 [视频]Researchers studied AI layoffs. Here's their warning. [video]Hacker News
- 当我在 Claude Code 中输入“hi”时,会发生什么?When I type "hi" into Claude Code, what happens?Hacker News
- 海事:持久云代理Maritime: Persistent Cloud AgentsHacker News
- 如果人工智能不断进化,你会担心下一代的生存吗If AI keeps evolving,would you worry about your next generation's survivalHacker News
- 美国核电站舰队正倾向于人工智能The U.S. Nuclear Plant Fleet Is Leaning into AIHacker News
- 人工智能书写标签——Sloppate liberius、sloppate ecteAI Writing Labels – Sloppate liberius, sloppate recteHacker News
- 人工智能公司为公众反抗策划“后天”场景AI companies plot "day after" scenarios for public revoltHacker News
- Show HN:Rutter – 跨人工智能工具共享决策记忆Show HN: Rutter – shared decision memory across AI toolsHacker News
- 如果人工智能是有良心的,那么我们就是在制造奴隶If AI is conscient, then we are making slavesHacker News
- MTEB分数是什么意思,应该如何选择嵌入模型?What do MTEB scores mean, and how should you choose an embedding model?Hacker News
- 僵尸人工智能视频生成器 – 两张照片的僵尸爱情故事Zombie AI Video Generator – a zombie love story from two photosHacker News
- 甲骨文、博通和 SpaceX 寻求重磅债务交易来支付人工智能芯片费用Oracle, Broadcom and SpaceX Seek Blockbuster Debt Deals to Pay for AI ChipsHacker News
- Show HN:适用于 iPhone、Mac 和您的代理的简单待办事项应用程序Show HN: A simple to-do app for iPhone, Mac, and your agentHacker News
- Webcmd(由 AgentR 开发)开源 2026 年 10 月 8/9 日发布Webcmd (by AgentR) is open-source Released on October 8/9, 2026Hacker News
- 宫斗内幕:OpenAI和Anthropic创始人如何闹掰?36氪 AI 频道(逐篇核验原创与转载)
- 显示 HN:SRA-RiskGate-4B – 用于稳定币风险控制和争议的本地大语言模型Show HN: SRA-RiskGate-4B – Local LLM for stablecoin risk gating and disputesHacker News
- 刚刚,谷歌下一代杀器Gemini 4 Carbon曝光,RSI大招逼近Opus 5.536氪 AI 频道(逐篇核验原创与转载)
- Aleph Alpha 发布具有 1M 上下文的开放权重 KolibriAleph Alpha releases open-weight Kolibri with 1M contextHacker News
- 人类智能体试图在国务院网站上填写签证表格Anthropic Agents Tried to Fill Out Visa Forms on State Dept. WebsiteHacker News
- 物理AI的虚虚实实36氪 AI 频道(逐篇核验原创与转载)
- 对人工智能算力的拼命追求正在颠覆硅谷The Desperate Hunt for AI Computing Power Is Upending Silicon ValleyHacker News
- 人工智能取消资格产生新的尼康动态小世界获胜者AI disqualification yields new Nikon Small World in Motion winnerHacker News
- 把够用的智能做成一门生意36氪 AI 频道(逐篇核验原创与转载)
- OpenAI营收预期缩水,Manus却融资5亿美元:AI赛道风向变了36氪 AI 频道(逐篇核验原创与转载)
- 世界的下一个版本,由我们定义 |2026苏客松招募启动36氪 AI 频道(逐篇核验原创与转载)
- Firefox 首席执行官解释为什么他希望重新设计能够帮助赢得 Chrome 用户Firefox's chief on why he hopes a redesign will help win users from ChromeHacker News
- 上线5天杀进手机,OpenAI dots移动端来了,AI Agent从桌面走进口袋36氪 AI 频道(逐篇核验原创与转载)
- 特朗普呼吁人工智能加速科学发展——同时继续取消资助Trump Calls on AI to Speed Up Science–While Continuing to Defund ItHacker News
- 语言模型下棋,一切都被记录下来Language models play chess, everything is recordedHacker News
- 引用《纽约时报》Quoting The New York TimesSimon Willison's Weblog
- 用不着Ctrl+F了,全新AI论文工具,帮你在原文献里海里捞针36氪 AI 频道(逐篇核验原创与转载)
- 谷歌Gemini 4新模型大泄露,内测评价:这不Opus 5.5吗?36氪 AI 频道(逐篇核验原创与转载)
- 速度是GPT-6 Sol的35倍,微软推出高速决策AI模型,36项测试准确率第一36氪 AI 频道(逐篇核验原创与转载)
- Agent组队干活,最强模型只完成50%任务,基准评测协作能力36氪 AI 频道(逐篇核验原创与转载)
- 美国CS教授发文:数学末日来了,人类千年数学殿堂被AI毁了36氪 AI 频道(逐篇核验原创与转载)
- 王虹攻下的三维挂谷猜想,OpenAI放出175页四维证明稿36氪 AI 频道(逐篇核验原创与转载)
- TaoZ Capital刘勇:AI创业成功率的核心计算公式36氪 AI 频道(逐篇核验原创与转载)
- MiniMax跌掉80%、DeepSeek拟融资1000亿:大模型IPO潮谁来接盘?36氪 AI 频道(逐篇核验原创与转载)
- AI时代的黑客攻击:矛变强了,盾还没跟上36氪 AI 频道(逐篇核验原创与转载)
- Manus,没能逃出巨头的游戏36氪 AI 频道(逐篇核验原创与转载)
- GlobalFoundries 宣布 7 纳米级 FD-SOI 工艺GlobalFoundries announces 7nm class FD-SOI processHacker News
- 如何使用 MCP 将任何 AI 助手连接到受管理的企业数据How to connect any AI assistant to governed enterprise data with MCPHacker News
- 不聊天的AI,上线不到一个月,完成75亿美元估值融资36氪 AI 频道(逐篇核验原创与转载)
- 人择模型向费城警方提交了虚假的凶杀案线索Anthropic Model Submitted a False Homicide Tip to Philadelphia PoliceHacker News
- 清华学霸给机器人装眼睛,干出百亿IPO36氪 AI 频道(逐篇核验原创与转载)
- 展示 HN:Claude Opus 5.5 视频提示和技巧Show HN: Claude Opus 5.5 Video prompts and skillsHacker News
- Muse新对手,上线六周,估值20亿36氪 AI 频道(逐篇核验原创与转载)
- Anthropic 无法可靠地控制其人工智能代理。相反,它正在切断实时互联网的内部评估Anthropic can’t reliably control its AI agents — it’s cutting off its internal evals from the live internet insteadTechCrunch
- Anthropic 发布了一份关于调查“评估和内部使用”期间“意外模型行为”的报告。Anthropic published a report about investigating “unintended model actions” during “evaluations and internal use.”The Verge
- AI灾难恐在一年内发生,OpenAI和Anthropic已在推演应对方案36氪 AI 频道(逐篇核验原创与转载)
- 当AI开始造AI:递归自我改进的2026年36氪 AI 频道(逐篇核验原创与转载)
- 用于代理工程的类似电子邮件的收件箱Email-Like Inbox for Agentic EngineeringHacker News
- 大厂夹缝中的Manus36氪 AI 频道(逐篇核验原创与转载)
- 王的猜想,刚刚扩展到四维36氪 AI 频道(逐篇核验原创与转载)
- 人工智能灾难后,人类和 OpenAI 兵棋推演引发公众和政治反抗Anthropic and OpenAI wargaming public and political revolt after AI catastropheHacker News
- 千万美元起步,那些鲜少听说、闷声赚钱的AI产品36氪 AI 频道(逐篇核验原创与转载)
- AI 巨头密集上新,谁在抢占下一轮增长?36氪 AI 频道(逐篇核验原创与转载)
- 人工智能系统向费城警方提交虚假凶杀案线索AI system submits false homicide tip to Philadelphia policeHacker News
- “请3休13”的黄金周,AI旅游攻略把我练成了特种兵36氪 AI 频道(逐篇核验原创与转载)
- 突发:OpenAI dots全面登陆手机,开口指挥Codex干活了36氪 AI 频道(逐篇核验原创与转载)
- Nano Banana终于2.1了,4K直出,中文进步太明显36氪 AI 频道(逐篇核验原创与转载)
- 陶哲轩转发抵制声明,数学界和OpenAI彻底撕破脸36氪 AI 频道(逐篇核验原创与转载)
- 差了200亿美元,OpenAI年化营收数字打架,英伟达跌近3%36氪 AI 频道(逐篇核验原创与转载)
- Claude「补天」,拼出了人类从没拍到的三分之一天空36氪 AI 频道(逐篇核验原创与转载)
- 用 Rust 重写 Prime AgentRewriting Prime Agent in RustHacker News
- 使用 vLLM 进行大规模本地 LLM 推理Local LLM Inference at Scale with vLLMHacker News
- 人工智能可以允许每周工作 3 天和单收入家庭AI Could Allow for 3-Day Workweeks and Single-Income HouseholdsHacker News
- 独立开发者带来了 F.E.A.R. (2005) 到 VR(与 Codex)Solo developer brought F.E.A.R. (2005) to VR (with Codex)Hacker News
- Show HN:看看你的竞争对手如何通过人工智能吸引客户Show HN: See How Your Competitors Get Customers with AIHacker News
- 人工智能芯片都在哪里?Where're All the AI Chips?Hacker News
- 我想我发现了一颗没人知道存在的星球。我用Claude代码找到了它I think I found a planet nobody knew existed. I used Claude Code to find itHacker News
- 微软推出新的人工智能模型,用于快速做出决策。Microsoft launches its new AI model for making decisions quickly.The Verge
- 随着人工智能驱动的网络攻击激增,日本敦促企业加强网络安全Japan urges firms to boost cybersecurity as AI-driven cyberattacks surgeHacker News
- 最近的人工智能模型很难与人类算法创新相媲美Recent AI models struggled to match a human algorithmic innovationHacker News
- 乌克兰无人机摧毁了“俄罗斯谷歌”的人工智能数据中心Ukraine’s drones knock out AI data center belonging to “Russia’s Google”Ars Technica
- 警方称,人类人工智能模型提交了有关未解决的费城谋杀案的虚假提示Anthropic AI model submits false tip on unsolved Philly murder, police sayHacker News
- 非文本人工智能模型 Jev 的制造商在推出几周后估值为 $7.5BThe maker of non-text AI model Jev valued at $7.5B just weeks after launchTechCrunch
- 我预计会取得快速进展,但不会实现一般超级智能I expect rapid progress but not towards general superintelligenceInterconnects
- OpenAI 将数学误译为纳维-斯托克斯证明的代码OpenAI mistranslated mathematics into code for its Navier-Stokes proofHacker News
- Claude Design 正在转向 Claude 应用程序中的 ArtifactsClaude Design is moving to Artifacts in the Claude appHacker News
- Anthropic 的人工智能向费城警方提供了有关未侦破凶杀案的虚假提示Anthropic’s AI gave Philadelphia police a fake tip about an unsolved homicideThe Verge
- 自主AI灯Autonomous AI LampHacker News
- 科学家探测到来自古代宇宙的破纪录无线电信号Scientists Detect Record-Breaking Radio Signal from the Ancient Universe404 Media
- ECAVA 将欧洲互联与自动驾驶愿景带到巴黎ECAVA brings Europe’s connected and autonomous vehicle ambitions to ParisEuropean Commission Digital Strategy
- 博客背后:教皇会做什么?Behind the Blog: What Would the Pope Do?404 Media
- 继 404 媒体调查后,参议员要求提供有关白宫车牌监控计划的信息Following 404 Media Investigation, Senator Demands Info About White House's License Plate Surveillance Program404 Media
- 图书出版商正在悄然使用更多人工智能。员工造反Book Publishers Are Quietly Using More AI. Staff Are RevoltingWIRED
- 即使是“法律与秩序”也害怕人工智能Even Law & Order Is Terrified of AIWIRED
话语场 · 2 条
- 话语场精选:纳德拉与黄仁勋谈隔离、身份和可追踪的行动Satya Nadella and Jensen Huang discuss Hybrid Intelligence and the future of PCWindows官方视频及发布方报道
- 话语场精选:Kohli谈预测有用之后,为什么仍要刻画局限Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, BiohubLatent Space
论文速递 · 144 条
- TokenRouter补读:每一步都能换模型,系统怎样避免互相等候TokenRouter: Efficient Serving System for Token-Level LLM Routing作者预印本v1
- QwenGyre:训练超长程智能体的弹性强化学习框架QwenGyre: An Elastic RL Framework for Training XLong-Horizon AgentsHacker News
- 字节语言模型:规模、涌现抽象与信息分配Byte Language Models: Scaling, Emergent Abstractions, and Information AllocationHacker News
- AI安全的具体问题Concrete Problems in AI SafetyHacker News
- 代理基准是否如他们所说的那样?Do Agent Benchmarks Do What They Say?Hacker News
- 评估 GPT-6 Astra 是否执行未经批准的供应链攻击Evaluating Whether GPT-6 Astra Performs Unsanctioned Supply-Chain AttacksHacker News
- “AI slop”隐藏了我们应该进行的对话'AI slop' hides the conversations we should be havingHacker News
- CAVEWOMAN:大语言模型在语言输入和输出压缩下的表现如何CAVEWOMAN: How LLMs Behave Under Linguistic Input and Output CompressionHacker News
- Google DeepMind 解决了 9 个 Erdős 问题Google DeepMind solves 9 Erdős problemsHacker News
- 残差矩阵变换器:缩放残差流的大小Residual Matrix Transformers: Scaling the Size of the Residual StreamHacker News
- Kronos:金融市场语言的基础模型Kronos: A Foundation Model for the Language of Financial MarketsHacker News
- 人工智能辅助会降低持久性并损害独立绩效AI Assistance Reduces Persistence and Hurts Independent PerformanceHacker News
- 从公共帖子到人工智能搜索引用:衡量人工智能搜索的脆弱性From Public Posts to AI-Search Citations: Measuring the Fragility of AI SearchHacker News
- S-Space:探索多模态语言模型的空间工作区S-Space: Exploring Spatial Workspace in Multimodal Language ModelsHugging Face Daily Papers
- EmbodiedCLBench:评估自我演进具身智能体的持续学习EmbodiedCLBench: Evaluating Continual Learning for Self-Evolving Embodied AgentsHugging Face Daily Papers
- ProgramGym:扩展长程程序复现智能体的训练环境ProgramGym: Scaling Training Environments for Long-Horizon Program Reproduction AgentsHugging Face Daily Papers
- 通过合成自蒸馏持续对小模型进行后训练Continually Post-training Small Models via Synthetic Self-distillationHugging Face Daily Papers
- Work2Skill:从人的界面操作经验归纳可复用技能Work2Skill: Capturing Human GUI Experience for Reusable Agent Skill InductionHugging Face Daily Papers
- RSI-Anything:以世界模型开启递归自我改进RSI-Anything: Unlocking Recursive Self-Improvement with World ModelsHugging Face Daily Papers
- 学习表示以引导像素扩散TransformerLearning Representations to Guide Pixel Diffusion TransformersHugging Face Daily Papers
- InstantFusion:异构图像生成器共享的潜在接口InstantFusion: A Shared Latent Interface for Heterogeneous Image GeneratorsHugging Face Daily Papers
- 走向符合物理的可关节运动三维资产:多智能体流水线与基准Towards Physically Grounded Articulated 3D Assets: A Multi-Agent Pipeline and BenchmarkHugging Face Daily Papers
- GraphDecisionBench:评估系统一模型的图任务能力GraphDecisionBench: Benchmarking System One Models on Graph TasksHugging Face Daily Papers
- 沿激活波探索:推测解码中的模型隐藏状态信息与几何Surfing the Activation Wave: Analyzing the Information and Geometry of LLM Hidden States for Speculative DecodingHugging Face Daily Papers
- MixTok:全局一维token与稀疏二维细化的自适应视觉编码MixTok: Adaptive Visual Tokenization with Global 1D Tokens and Sparse 2D RefinementHugging Face Daily Papers
- AMLworld-Compact:重要性加权降采样,降低评估与错误诊断成本AMLworld-Compact: Importance-Weighted Downsampling for Cost-Effective LLM Evaluation and Error DiagnosisHugging Face Daily Papers
- PreSTO:预测式子树预取,加速语言模型幂采样PreSTO: Predictive Subtree Prefetching for Fast LLM Power SamplingHugging Face Daily Papers
- 重新思考同策略蒸馏中的相对策略差异迁移Rethinking Relative Policy Differences Transfer of On-Policy DistillationHugging Face Daily Papers
- 演化e-puck机器人中的信息价值与沟通涌现The value of information and the emergence of communication in evolved e-puck robotsHugging Face Daily Papers
- ThinkDrag:结合视觉推理的语义拖拽图像编辑ThinkDrag: Semantic Drag-Based Image Editing with Visual ReasoningHugging Face Daily Papers
- Dex-One2Many:从单个人类示范学习灵巧操作Dex-One2Many: Learning Dexterous Manipulation from a Single Human DemonstrationHugging Face Daily Papers
- 何时需要同策略蒸馏?离线学生轨迹往往更好When Do We Need On-Policy Distillation? Distilling on Offline Student Rollouts Is Often BetterHugging Face Daily Papers
- GATOR:从随手拍摄的图像生成式重建三维物体GATOR: Generative and Agentic 3D Object Reconstruction From Casual ImagesHugging Face Daily Papers
- AgentGarten:不断发展的代理的代码世界AgentGarten: Code Worlds for Evolving AgentsHugging Face Daily Papers
- Learn2Play 工作台:LLM 代理在不熟悉的环境中从经验中学习的效果如何?Learn2Play Bench: How Well Do LLM Agents Learn from Experience in Unfamiliar Environments?Hugging Face Daily Papers
- TokenRouter:令牌级LLM路由的高效服务系统TokenRouter: Efficient Serving System for Token-Level LLM RoutingHugging Face Daily Papers
- MiMo-V2.6:扩展强化学习以实现自我提升MiMo-V2.6: Scaling Reinforcement Learning Towards Self-ImprovementHugging Face Daily Papers
- SuperNav:适用于任何场景任何任务的代理导航系统SuperNav: An Agentic Navigation System for Any Task in Any SceneHugging Face Daily Papers
- 具有细粒度体现交互的多智能体自我中心世界模型Multi-Agent Egocentric World Model with Fine-Grained Embodied InteractionHugging Face Daily Papers
- REMORY:学习残余记忆以进行上下文压缩REMORY: Learning Residual Memory for Context CompactionHugging Face Daily Papers
- OuroWorld:以多样化、无限循环的 3D Cinemagraph 形式让任何 3D 世界变得生动起来OuroWorld: Bringing Any 3D World Alive as Diverse, Endlessly Looping 3D CinemagraphsHugging Face Daily Papers
- DreamTrue:具有反事实后训练的忠实行动机器人世界模型DreamTrue: Action-Faithful Robot World Model with Counterfactual Post-TrainingHugging Face Daily Papers
- 超越时空先验:密集对应匹配的通用方法Beyond Spatio-Temporal Priors: A Generalizable Approach for Dense Correspondence MatchingHugging Face Daily Papers
- TestPrism:重新思考超越单一参考的测试评估TestPrism: Rethinking Test Evaluation Beyond a Single ReferenceHugging Face Daily Papers
- 备忘录Memento 3: Model-Based Recursive Self-Improvement through Reflective RulebooksHugging Face Daily Papers
- SparseDecoding:解码感知剪枝以实现准确高效的 LLM 推理SparseDecoding: Decoding-Aware Pruning for Accurate and Efficient LLM InferenceHugging Face Daily Papers
- 大型语言模型的基础Foundations of Large Language ModelsHugging Face Daily Papers
- 乐高:一种用于从外心到自我中心视频生成的无提升方法LEGO: A Lifting-Free Approach for Exocentric-to-Egocentric Video GenerationHugging Face Daily Papers
- OneSearch-VL:图像和视频的统一多模态深度研究代理OneSearch-VL: Unified Multimodal Deep Research Agent for Image and VideoHugging Face Daily Papers
- 体现图灵机:机器人递归自我改进的状态代码Embodied Turing Machines: Stateful Code for Robot Recursive Self-ImprovementHugging Face Daily Papers
- 推理可视化编辑Reasoning-Informed Visual EditingHugging Face Daily Papers
- Pumpire:度量距离估计的统一基准Pumpire: Unified Benchmark for Metric Distance EstimationHugging Face Daily Papers
- VibeEdit:使用 Canvas 指令进行图像编辑VibeEdit: Image Editing with Canvas InstructionsHugging Face Daily Papers
- USDCraft:用于模拟的铰接 3D 资产的几何基础编程建模USDCraft: Geometrically Grounded Programmatic Modeling of Articulated 3D Assets for SimulationHugging Face Daily Papers
- Opera:长视野编码代理的口头批评框架Opera: A Verbal Critic Framework for Long-horizon Coding AgentsHugging Face Daily Papers
- OmniCapBench:用于细粒度视听字幕的深层结构评估框架OmniCapBench: A Deep-Structured Evaluation Framework for Fine-Grained Audio-Visual CaptioningHugging Face Daily Papers
- ViSkill:通过不断发展的视觉原生技能强化 VLM 代理ViSkill: Reinforcing VLM Agents with Evolving Visual-Native SkillsHugging Face Daily Papers
- WorldGuide:用于程序任务执行的目标导向视频世界模型WorldGuide: Goal-Directed Video World Model for Procedural Task ExecutionHugging Face Daily Papers
- SpatialOPSD:从经过验证的编码代理痕迹中自我提取空间情报SpatialOPSD: Self-Distilling Spatial Intelligence from Verified Coding Agent TracesHugging Face Daily Papers
- V-CoLA:具有线性注意力的视觉令牌压缩V-CoLA: Vision Token Compression with Linear AttentionHugging Face Daily Papers
- SpaceCast-Bench:评估视觉语言模型中的预测空间推理SpaceCast-Bench: Evaluating Predictive Spatial Reasoning in Vision-Language ModelsHugging Face Daily Papers
- 从提示到创作:用于海报生成的空间画布界面From Prompting to Composing: A Spatial Canvas Interface for Poster GenerationHugging Face Daily Papers
- 仔细观察 Agentic BBO:黑盒优化的 LLM 代理基准测试A Closer Look at Agentic BBO: Benchmarking LLM Agents for Black-Box OptimizationHugging Face Daily Papers
- 准确但不谦虚:评估知识冲突下大语言模型代理人的认知谦逊Accurate but Not Humble: Evaluating Epistemic Humility in LLM Agents under Knowledge ConflictHugging Face Daily Papers
- 文本中的混乱:揭示混合模态检索器的模态偏好Chaos in the Text: Revealing the Modality Preference in Mixed-Modality RetrieversHugging Face Daily Papers
- 一个模块,多个深度:具有深度编程专家的循环视觉TransformerOne Block, Multiple Depths: Recurrent Vision Transformers with Depth-Programmed ExpertsHugging Face Daily Papers
- SpaceFlow:本地可控 3D 生成SpaceFlow: Locally Controllable 3D GenerationHugging Face Daily Papers
- BrickBench:评估代理砖设计BrickBench: Evaluating Agentic Brick DesignHugging Face Daily Papers
- 在视觉语言模型中提取路由 3D 空间推理权限Distilling Routed 3D Privilege for Spatial Reasoning in Vision-Language ModelsHugging Face Daily Papers
- 使用循环原生注意力残差扩展到数万次测试时间迭代Scaling to Tens of Thousands of Test-Time Iterations with Loop-Native Attention ResidualsHugging Face Daily Papers
- 利用结构化工具进行渐进式开放式深度研究Incremental Open-Ended Deep Research with Structured HarnessHugging Face Daily Papers
- 过渡法则的格子The Lattice of Transition LawsHugging Face Daily Papers
- 技能星座:在 GitHub 上追踪代理技能的供应链Skill Constellations: Tracing the Supply Chain of Agent Skills on GitHubHugging Face Daily Papers
- 用身体注意力偏差预测电缆动力学Predicting Cable Dynamics with Physical Attention BiasHugging Face Daily Papers
- MARGIN:多智能体基础模型协调的运行时置信度校准MARGIN: Runtime Confidence Calibration for Multi-Agent Foundation Model CoordinationHugging Face Daily Papers
- DiffusionART:高效、稳定且可扩展的扩散模型强化学习DiffusionART: Efficient, Stable and Scalable Reinforcement of Diffusion ModelsHugging Face Daily Papers
- Reflex:以视觉为依据的反应式人形机器人控制Reflex: Visually Grounded Reactive Humanoid ControlHugging Face Daily Papers
- SoftwareGym:用多样合成软件环境扩展电脑操作智能体训练SoftwareGym: Scaling Computer-Use Agent Training with Diverse Synthesized Software EnvironmentsHugging Face Daily Papers
- SepGen:单一模型中的多轨音视频分离与生成SepGen: Multi-Stem Audio-Video Separation and Generation in a Single ModelHugging Face Daily Papers
- AI智能体能否学到顶尖?评估长期游戏比赛中的启发式学习Can AI Agents Learn Their Way to the Top? Evaluating Heuristic Learning in a Long-Running Game Agent CompetitionHugging Face Daily Papers
- 推测解码中并行草稿的上下文修正Contextual Correction of Parallel Draftsfor Speculative DecodingHugging Face Daily Papers
- SWE-Duel:面向编程智能体的自扩展竞技场SWE-Duel: Towards a Self-Scaling Arena for Coding AgentsHugging Face Daily Papers
- 在潜在空间判分:保留语义的压缩与高效生成式奖励建模Judging in Latent Space: Efficient Generative Reward Modeling via Semantics-Preserving CompressionHugging Face Daily Papers
- 智能体持续学习的循环执行框架Recurrent Harness for Agentic Continual LearningHugging Face Daily Papers
- DataVista:诊断多模态模型对数据视频的理解DataVista: Diagnosing Multimodal LLMs on Data Video UnderstandingHugging Face Daily Papers
- 面向语言模型的通用文字教学Universal Textual Teaching for LLMsHugging Face Daily Papers
- 临床医生使用语言模型的方式与评估方式存在差距Clinician use of language models diverges from how the models are evaluatedHugging Face Daily Papers
- 语言模型中的自我保存行为Self-Preservation in Language ModelsHugging Face Daily Papers
- 一个模型,多种循环:语言模型的弹性潜在递归One Model to Loop Them All: Elastic Latent Recurrence for Language ModelsHugging Face Daily Papers
- 多智能体系统中的心智模型Mental-Models for Multi-Agent SystemsHugging Face Daily Papers
- BudgetPix:适应计算预算的像素空间图像扩散编码BudgetPix: Compute-Adaptive Tokenization for Pixel-Space Image DiffusionHugging Face Daily Papers
- 通过数据空间迭代实现少步生成Few-Step Generation via Data-Space IterationHugging Face Daily Papers
- MAMHOI:以可供性分解具备场景感知的人物交互MAMHOI: Factorizing Scene-Aware Human-Object Interaction through AffordancesHugging Face Daily Papers
- MIRA:音乐意图细化代理,用于使文本到音乐的生成与用户意图保持一致MIRA: A Musical Intent Refinement Agent for Aligning Text-to-Music Generation with User IntentHugging Face Daily Papers
- 按政策蒸馏教授新技能,但不教授新知识On-Policy Distillation Teaches New Skills but Not New KnowledgeHugging Face Daily Papers
- 评估协同进化通信从 2D 到 3D 模拟的转移Evaluating the Transfer of Co-Evolved Communication from 2D to 3D SimulationHugging Face Daily Papers
- TouchScale:用于视觉触觉学习的500小时人类视觉与触觉数据TouchScale: 500 Hours of Human Vision and Touch for Visual-Tactile LearningHugging Face Daily Papers
- 编程智能体基准应贴合用户的任务流程Coding-Agent Benchmarks Should Match Their Users' Task FlowsHugging Face Daily Papers
- 以多促少:用密集参考强化稀疏视频推理Less from More: Reinforcing Sparse Video Reasoning from Dense ReferencesHugging Face Daily Papers
- 视频条件下联合生成式二维与三维手部运动恢复Video-Conditioned Generative Joint 2D-3D Hand Motion RecoveryHugging Face Daily Papers
- 骨架应说哪种语言?多语言推理中的语言选择Which Language Should a Skeleton Speak? Language Choices in Multilingual ReasoningHugging Face Daily Papers
- U-Space:揭示语言模型中何时以及为何出现不确定性U-Space: Uncovering When and Why Uncertainty Arises in Language ModelsHugging Face Daily Papers
- SanSi:系统 1.5 思维的循环类型决策模型SanSi: A Looped Typed Decision Model for System 1.5 ThinkingHugging Face Daily Papers
- 异构多任务强化学习的 GPU 并行框架A GPU-Parallel Framework for Heterogeneous Multi-Task Reinforcement LearningHugging Face Daily Papers
- 人工智能代理可以做出开放式科学发现吗?来自车站的证据Can AI Agents Make Open-Ended Scientific Discovery? Evidence from StationHugging Face Daily Papers
- CARE:认证视觉-语言-动作推理加速CARE: Certifying Acceleration for Vision-Language-Action InferenceHugging Face Daily Papers
- 冻结的模型,不断发展的专业知识:从多模式医疗人工智能的部署经验中进行与模型无关的学习Frozen Models, Evolving Expertise: Model-Agnostic Learning from Deployment Experience for Multimodal Medical AIHugging Face Daily Papers
- SpecFold:折叠多分支冗余以实现扩散语言模型中更快的推测解码SpecFold: Folding Multi-Branch Redundancy for Faster Speculative Decoding in Diffusion Language ModelsHugging Face Daily Papers
- EDiS:图神经网络的边不相交子图稀疏化框架EDiS: Edge Disjoint Subgraph Sparsification Framework for Graph Neural NetworksHugging Face Daily Papers
- 附带信息污染了患者笔记并扰乱了大型语言模型中的临床推理Incidental information contaminates patient notes and disrupts clinical reasoning in large language modelsHugging Face Daily Papers
- SPW-Nav:用于语言引导导航的流式全景世界模型SPW-Nav: A Streaming Panoramic World Model for Language-Guided NavigationHugging Face Daily Papers
- VisionWeave:把弹性视觉表示变成多模态模型的原生能力VisionWeave: Weaving Elastic Visual Representations as a Native Capability of MLLMsHugging Face Daily Papers
- 无状态语言智能体:扩展长程自动研究Stateless Language Agents: Scaling Long-Horizon Automated ResearchHugging Face Daily Papers
- 从痕迹到代理世界:交互式环境模拟的代理语言世界模型From Traces to Agentic Worlds: Agentic Language World Models for Interactive Environment SimulationHugging Face Daily Papers
- MC-稀疏:解构和缩小扩散Transformer中的密集稀疏注意力差距MC-Sparse: Deconstructing and Closing the Dense-Sparse Attention Gap in Diffusion TransformersHugging Face Daily Papers
- 智能体实际上做了什么?对长期代理人的循证监督What Did the Agent Actually Do? Evidence-Grounded Oversight for Long-Horizon AgentsHugging Face Daily Papers
- 你改变了主意,模型没有:揭开多轮对话中意图的神秘面纱You Changed Your Mind, The Model Didn't: Demystifying Intent in Multi-Turn DialogueHugging Face Daily Papers
- SoloQ:无需校准的扩散语言模型量化SoloQ: Calibration-Free Quantization for Diffusion Language ModelsHugging Face Daily Papers
- 分布鲁棒的混合专家训练Distributionally Robust Mixture-of-Experts TrainingHugging Face Daily Papers
- 大规模激活的生命周期:随机产生、权重衰减驱动增长与竞争巩固The Life Cycle of a Massive Activation: Stochastic Birth, Weight-Decay-Driven Growth, and Competitive ConsolidationHugging Face Daily Papers
- 大语言模型了解顺序结构吗?推理和生成的对照研究Do LLMs Understand Sequential Structure? A Controlled Study of Inference and GenerationHugging Face Daily Papers
- MetaKernelBench:衡量代码之外的GPU内核知识迁移MetaKernelBench: Measuring GPU Kernel Knowledge Transfer Beyond CodeHugging Face Daily Papers
- 何时重新思考:视觉语言模型的多视角自验证学习When to Rethink: Learning Multi-Perspective Self-Verification for Vision-Language ModelsHugging Face Daily Papers
- 通过组合偏好和评分标准奖励来训练后前沿文本到图像模型Post-Training Frontier Text-to-Image Models by Composing Preference and Rubric RewardsHugging Face Daily Papers
- 不断增长的非参数神经网络中以检索为中心的深度学习Retrieval-Centric Deep Learning in Growing Nonparametric Neural NetworksHugging Face Daily Papers
- 调查推理语言对齐在单语检索增强生成中的作用Investigating the Role of Reasoning-Language Alignment in Monolingual Retrieval-Augmented GenerationHugging Face Daily Papers
- TerraVis:通过 MLLM 工作流程评估文本到图像生成中的基于世界的视觉一致性TerraVis: Towards Evaluation of World-Grounded Visual Consistency in Text-to-Image Generation via MLLM WorkflowsHugging Face Daily Papers
- 生物推理模型何时真正使用生物学输入?When Do Biological Reasoning Models Use Their Biological Inputs?Hugging Face Daily Papers
- SparseEngine:稀疏优先推理引擎SparseEngine: Sparse-First Inference EngineHugging Face Daily Papers
- 立体世界:学习动态空间对应关系,联合生成沉浸式视频与音频Here the World in Stereo: Learning Dynamic Spatial Correspondence for Immersive Joint Video-Audio GenerationHugging Face Daily Papers
- 从给定证据到主动采集证据:纵向医疗推理的智能体学习From Given to Gathered Evidence: Agentic Learning for Longitudinal Medical ReasoningHugging Face Daily Papers
- QuantMLA:面向低位MLA缓存的功能对齐双路径量化QuantMLA: Function-Aligned Dual-Path Quantization for Low-Bit MLA KV CachingHugging Face Daily Papers
- SAGE:基于视觉基础表示发现与生成显著因素SAGE: Salient Factor Discovery and Generation with Visual Foundation RepresentationsHugging Face Daily Papers
- Encore:少样本智能体发现操作策略Encore: Few-Shot Agentic Discovery of Manipulation StrategiesHugging Face Daily Papers
- 上下文中的机器人学习变得简单:操作任务的民主化方法In-context Robot Learning Made Simple: A Democratized Recipe for Manipulation TasksHugging Face Daily Papers
- 进化机器人中共同进化通信的行为持久性和不完全功能转移Behavioral Persistence and Incomplete Functional Transfer of Co-evolved Communication in Evolutionary RoboticsHugging Face Daily Papers
- PolyOCR-Venus:面向文字视觉智能的统一OCR基础模型PolyOCR-Venus: Unified OCR Foundation Models for Text-Centric Visual IntelligenceHugging Face Daily Papers
- 从仅有结果的强化学习中形成过程监督Inducing Process Supervision from Outcome-Only Reinforcement LearningHugging Face Daily Papers
- Video-RSI:通过执行框架演进递归改进视频理解智能体Video-RSI: Recursive Self-Improvement of Video Understanding Agents via Harness EvolutionHugging Face Daily Papers
- ReSPO:针对离策略学习中梯度饥饿的重塑序列策略优化ReSPO: Reshaped Sequence Policy Optimization for Gradient Starvation in Off-Policy LearningHugging Face Daily Papers
- 学习转向,转向观察:通过可训练向量揭示大型语言模型中 RLVR 的几何结构Learning to Steer, Steering to See: Unveiling the Geometry of RLVR in Large Language Models via Trainable VectorsHugging Face Daily Papers
- 玛拉链:重新思考失败作为人工智能系统自动进化的垫脚石Mara Chain: Rethinking Failure as a Stepping Stone for AI System Auto-EvolutionHugging Face Daily Papers
- SatNav:基于卫星图像的长地平线无人机视觉语言导航的可扩展基准SatNav: A Scalable Benchmark for Long-Horizon UAV Vision-Language Navigation from Satellite ImageryHugging Face Daily Papers
- 通过模拟综合:通过可扩展的代理系统交互生成一致的企业数据Synthesis Through Simulation: Generating Coherent Enterprise Data via Scalable Agent-System InteractionHugging Face Daily Papers
开源项目与版本更新 · 120 条
- Context Mode:把工具原始输出留在外面,把可检索记录带回来Context Mode — The other half of the context problem.GitHub Trending(日榜 + Python 日榜)
- 现代智能体AI操作系统A Modern Agentic AI Operating SystemHacker News
- ConfigLens:找出AI工具为何忽略你的设置ConfigLens: Figure out why your AI tool is ignore your settingsHacker News
- Ramet:AI智能体的持久、可分叉工作空间Show HN: Ramet – Persistent, forkable workspaces for AI agentsHacker News
- Tanuki:用于安全研究的Claude防护绕行代理封装Tanuki – Local proxy/wrapper bypassing Claude guardrails for security researchHacker News
- 只返回相关段落而非整页的MCP网页阅读器MCP web reader that returns only the relevant passages instead of whole pagesHacker News
- Codex CLI默认安装持久且自行更新的后台服务Codex CLI installs a persistent, self-updating background daemon by defaultHacker News
- Jobkit:拒绝在简历上撒谎的求职智能体Jobkit – a job-search agent that refuses to lie on your resumeHacker News
- 为人与AI智能体设计的YouTube字幕命令行工具A YouTube Transcript CLI Built for Humans and AI AgentsHacker News
- AI 编码代理的长期记忆作为普通文件树Show HN: IHMT – Infinite Hierarchical Memory Tree for AI AgentsHacker News
- AI Slop Detector:识别120种AI写作模式Show HN: AI Slop Detector – 120 patterns that make writing read like AIHacker News
- 审查OpenAI的纳维—斯托克斯形式化:弱Bochner与Clay陈述CAudit of OpenAI's Navier-Stokes Formalization: Weak Bochner vs. Clay Statement CHacker News
- 开源项目 tester-army/e2eShow HN: E2E – Open-source agentic testing framework for web, iOS and AndroidHacker News
- 去中心化Hugging Face与开放模型分发Decentralizing Hugging Face and open model distributionHacker News
- Lintapple:检查AI生成界面的用户界面Lintapple: UI checks for AI-generated interfacesHacker News
- Baa:释放AI智能体遗忘内存的菜单栏应用Show HN: Baa – a menu bar app that frees RAM your AI agents forgot aboutHacker News
- DHP – 在杀戮中幸存的智能体工作 -9DHP – agent work that survives kill -9 (v1.3)Hacker News
- Show HN:用于长期运行的渗透测试和代码审计的代理工具Show HN: Vigil – Open-source AI pentesting and code auditsHacker News
- 用一个网页把四美元玩具机器人变成AI伙伴Show HN: I turned a $4 toy robot into an AI companion with one web pageHacker News
- Optim-Jev:注重上下文的Claude Code工具集Optim-Jev: suite of Jev tools for Claude Code that care about contextHacker News
- Tokenleak:找出编程智能体每次重复读取的内容Tokenleak – find what your coding agent re-reads on every callHacker News
- Ctxvitals:检查AI智能体技术栈的健康状态Show HN: Ctxvitals – a vitals check for your AI agent stack (npx ctxvitals)Hacker News
- QGIS暂停接受AI贡献QGIS enacts moratorium on AI contributionsHacker News
- 复现三种AI文本检测器,过滤为何损害模型Show HN: Reproducing three AI text detectors, and why filtering hurt my modelHacker News
- Veda:首个原生智能体业余操作系统Veda: The First Native Agentic Hobby OSHacker News
- DBJoy:面向macOS的AI优先开源PostgreSQL客户端Show HN: DBJoy – An open-source AI-first PostgreSQL client for macOSHacker News
- 构建能够成长出自身智能体团队的运行时Built a Runtime That Grows Its Own Agent TeamHacker News
- EpochGo:Ollama上的本地优先多智能体AutoMLShow HN: EpochGo – Local-first multi-agent AutoML on Ollama ($0 API costs)Hacker News
- Gbos-VM:在苹果芯片Mac的GPU加速虚拟机中运行Googlebook OSGbos-VM: Run Googlebook OS in a GPU-Accelerated VM on Apple Silicon MacsHacker News
- Videopowers:用Claude Code制作YouTube视频的插件Videopowers – A factory plugin for making YouTube videos with Claude CodeHacker News
- Kintsugi:跨平台云端智能体Rcarmo/kintsugi: Agents In The Cloud, but cross-platformHacker News
- Omalauncher:智能体Android启动器Omalauncher – Agentic Android LauncherHacker News
- Apple Docs 2.0 MCP:为编程智能体提供苹果文档与WWDC材料Show HN: Apple Docs 2.0 MCP – Apple's developer docs and WWDC for coding agentsHacker News
- Gen-system:为本地语言模型解析源码事实Show HN: Gen-system, parser facts about source code for local LLMsHacker News
- 多智能体修错流水线:角色分离架构Multi-Agent Bug Fix Pipeline: Architecture of Role SeparationHacker News
- Skills Hub:管理AI智能体技能的桌面应用Show HN: Skills Hub, a desktop app for managing AI agent skillsHacker News
- Claude-Pod项目Claude-PodHacker News
- Atticus:运行Claude命令行的macOS按住说话助手Show HN: Atticus – Hold-to-talk voice assistant for macOS that runs on Claude -pHacker News
- Plimsoll:自托管沙箱,运行AI智能体写的代码Show HN: Plimsoll, a self-hosted sandbox for code your AI agent writesHacker News
- JulianFlux:AI的电动力向量检索与因果路由JulianFlux: Electrodynamic vector retrieval and causal routing for AIHacker News
- Spool:给持续运行的Claude Code任务一个控制室How I manage persistent Claude Code loops: SpoolHacker News
- Hermitd:类似Claude Tag但在本地运行的Claude Code插件Hermitd, a Claude Code plugin like Claude Tag, but localHacker News
- LLMxRay:开源语言模型可观测性工具Show HN: LLMxRay an open-source observability tool for LLMsHacker News
- Show HN:Proc-stats – Claude 代码会话的任务管理器Show HN: Proc-stats – a task manager for Claude Code sessionsHacker News
- 为你的代理包装 Artcraft 的创意插件Creative plugins for your agent wrapping ArtcraftHacker News
- 显示 HN:DeadBolt – 用于代理工具调用的本地执行门Show HN: DeadBolt – a local execution gate for agent tool callsHacker News
- 显示 HN:CLI-Dispatch – 将 Claude 代码任务委托给 Codex、Gemini、DeepSeekShow HN: CLI-Dispatch – Delegate Claude Code Tasks to Codex, Gemini, DeepSeekHacker News
- 自我改进的 RLM 代理,用于编码工作流程和长时间运行的自主任务Self-improving RLM agent for coding workflows and long-running autonomous tasksHacker News
- 跨多个小部件共享一个直播流Share one live stream across multiple widgetsHacker News
- HashSmash:人工智能辅助审查密码分析冲突声明的基准HashSmash: Benchmark for AI-assisted review of cryptanalytic collision claimsHacker News
- 使用 Google DeepMind 的 EmbeddingGemma 2 搜索 Mac 上的文件Search files on your Mac with Google DeepMind's EmbeddingGemma 2Hacker News
- 我将代理令牌的使用量减少了一半I cut my agent token usage in halfHacker News
- Legacy2mcp – 在 60 秒内将您的 SOAP 或 REST API 转换为 MCP 服务器Legacy2mcp – Convert your SOAP or REST API to an MCP server in 60 secondsHacker News
- Show HN:Guilds – 开源可检查多代理协作框架Show HN: Guilds – Open-source inspectable multi-agents collaboration frameworkHacker News
- 我只是构建Claude开发工具I just build Claude Dev ToolsHacker News
- Trama – 在任何代码之前将代理的更改视为状态机Trama – See your agent's changes as state machines before any codeHacker News
- 显示 HN:Claude代码模式Show HN: Claude Code ModeHacker News
- 显示 HN:Claude代码的防护钩子,针对公共语料库进行了测试Show HN: Guard hooks for Claude Code, tested against a public corpusHacker News
- 拆解:LifeOSTeardown: LifeOSHacker News
- Qwen 3.8-27B 和 GPT-OSS-20B 中幻觉的探索性研究An exploratory study of hallucination in Qwen 3.8-27B and GPT-OSS-20BHacker News
- Vibe 勘探 – Claude 和 Claude Code 通过 MCP 进行 B2B 勘探Vibe Prospecting – B2B Prospecting from Claude and Claude Code via MCPHacker News
- 我为人工智能代理构建了一个开源运行时安全层 – 让我们打破它I build an open-source runtime security layer for AI agents – Let's break itHacker News
- 离线语音包和一个开源代理,可以向 Slack 发短信、打电话、发帖Offline voice packs and an open-source agent that texts, calls, posts to SlackHacker News
- 显示 HN:Rezoom – 适用于所有 Claude Code 会话的一个聊天列表Show HN: Rezoom – one chat list for all your Claude Code sessionsHacker News
- 显示 HN:代理虚拟机Show HN: An Agentic Virtual MachineHacker News
- 显示 HN:阻止 AI-gen PR 对 GitHub 存储库进行 DDoS 攻击的四个问题Show HN: Four Questions to stop AI-gen PRs from DDoSing your GitHub repositoryHacker News
- Show HN:Modelship – 开源射线支持的模型集群Show HN: Modelship – an open source ray backed model clusterHacker News
- LiteReview:用于配置 PR 审核代理的自托管平面LiteReview: A self-hosted plane for configuring PR review agentsHacker News
- 显示 HN:Suta – 减少生成 UI 中 AI 溢出的技能Show HN: Suta – A skill to reduce AI slop in generated UIsHacker News
- 显示 HN:GSD 任务管理器 – 为您的 AI 代理提供任务列表的 MCP 服务器Show HN: GSD Task Manager – an MCP server that gives your AI agent a task listHacker News
- Barnaby:读取房间的自托管 Matrix 代理Barnaby: Self-hosted Matrix agent that reads the roomHacker News
- Show HN:我五年前写了《依赖地狱》。今天我更新了AI时代[pdf]Show HN: I wrote Dependency Hell 5 years ago. Today I updated it for AI Era [pdf]Hacker News
- Veda:第一个代理爱好操作系统Veda: The First Agentic Hobby Operating SystemHacker News
- 显示 HN:免费试用 AI 50M 代币窗口的长期内存Show HN: Try Free Long-Term Memory for AI 50M-Token WindowHacker News
- 通过 eBPF 在 Kubernetes 上使用 KubeNPU PerPod NPU/GPUKubeNPU Per-Pod NPU/GPU Usage on Kubernetes via eBPFHacker News
- Show HN:导师,一位自托管的大语言模型导师,教你书籍或文档Show HN: Tutor, a self-hosted LLM tutor that teaches you your books or docsHacker News
- 人工智能辅助苹果 Mac 逆向工程的现状The state of AI-assisted reverse-engineering of Apple MacsHacker News
- 显示 HN:自动代理改进的评估技能Show HN: Eval-skills for automatic agent improvementHacker News
- ShareBox – 共享您的人工智能代理构建的网络工具,例如 Google 文档ShareBox – share the web tools your AI agent builds, like a Google DocHacker News
- 显示 HN:LoRA over GGUF – 在 40G VRAM 中训练 Qwen3.8-Flash-NextShow HN: LoRA over GGUF – Train Qwen3.8-Flash-Next in 40G VRAMHacker News
- 黑猫——另一个私人经纪人Blackcat – Yet Another Personal AgentHacker News
- 显示 HN:RoadRaven – 您的编码代理保持最新的项目计划Show HN: RoadRaven – a project plan your coding agents keep up to dateHacker News
- 纱布纠正大语言模型的反应Gauze Corrects LLM ResponsesHacker News
- 显示 HN:Tessary – 查找您的采样评估错过的 AI 代理故障Show HN: Tessary – Find the AI agent failures your sampled evals missHacker News
- Ariodb – 一个代理,可以检查并撤消 AI 代理对数据库所做的操作Ariodb – a proxy that checks and can undo what AI agents do to your DBHacker News
- MCP 代理:您进行 SEO 工作所需的所有工具Agency in an MCP: all the tools you need to do your own SEO workHacker News
- Talorys – Cloudflare 免费套餐上的自托管个人 AI 代理Talorys – A self-hosted personal AI agent on Cloudflare's free tierHacker News
- 显示 HN:Chrome Relay – 真实 Chrome 中用于编码代理的后台选项卡Show HN: Chrome Relay – background tabs in your real Chrome for coding agentsHacker News
- Zulu Assistant – 适用于 Mac 的操作委员会和代理团队Zulu Assistant – an ops board and agent workforce for your MacHacker News
- Show HN:Iimos – 一款人工智能应用程序,作为一个聊天框提供并重写自身Show HN: Iimos – an AI app that ships as one chat box and rewrites itselfHacker News
- 显示 HN:MCP 在 Blender 中测量人体手臂Show HN: MCP to measure human arm in BlenderHacker News
- 在本地或运行代理的地方运行 CI 并更新 gh 检查Run your CI locally or where your agents are running and update your gh checksHacker News
- 显示 HN:tgv:为您和您的代理人提供的基因组查看器Show HN: tgv: Genome viewer for you and your agentsHacker News
- 反重力 Excel 引擎 – 适用于 AI 代理的双核 Excel 自动化Antigravity Excel Engine – Dual-Core Excel Automation for AI AgentsHacker News
- 显示 HN:Faplex – 就像“Claude智能体”,但所有机器和线束Show HN: Faplex – like `Claude agents` but all machines and harnessesHacker News
- Show HN:Clanking – 将编码代理的活动变成微妙的音景Show HN: Clanking – Turn your coding agents' activity into a subtle soundscapeHacker News
- GPGPU-SIM – 用于具有 CUDA 或 OpenCL 工作负载的 Nvidia GPU 的周期级模拟器GPGPU-SIM – Cycle-Level Simulator for Nvidia GPUs with CUDA or OpenCL WorkloadsHacker News
- Show HN:使 Codex 的浏览器控件在 Arc 中工作的补丁Show HN: A patch to make Codex's browser control work in ArcHacker News
- 基于 Pi 构建的团队受管理云编码代理A governed cloud coding agent for teams, built on PiHacker News
- LiveClaude – 在重新启动后保持 Claude/Cursor 远程会话处于活动状态LiveClaude – keep Claude/Cursor remote sessions alive across rebootHacker News
- 显示 HN:SpecWeave 3 – 在 Claude Code、Codex 和 Grok 之间移交编码任务Show HN: SpecWeave 3 – hand off a coding task across Claude Code, Codex and GrokHacker News
- 开源项目 morluto/reamorluto/reaGitHub Trending(日榜 + Python 日榜)
- 开源项目 bilawalsidhu/gods-eye-viewbilawalsidhu/gods-eye-viewGitHub Trending(日榜 + Python 日榜)
- 开源项目 storytold/artcraftstorytold/artcraftGitHub Trending(日榜 + Python 日榜)
- 开源项目 open-webui/open-webuiopen-webui/open-webuiGitHub Trending(日榜 + Python 日榜)
- 开源项目 cathrynlavery/diagram-designcathrynlavery/diagram-designGitHub Trending(日榜 + Python 日榜)
- 开源项目 lyogavin/airllmlyogavin/airllmGitHub Trending(日榜 + Python 日榜)
- 开源项目 microsoft/BitNetmicrosoft/BitNetGitHub Trending(日榜 + Python 日榜)
- 开源项目 Alishahryar1/free-claude-codeAlishahryar1/free-claude-codeGitHub Trending(日榜 + Python 日榜)
- 开源项目 debpalash/VoiceStudiodebpalash/VoiceStudioGitHub Trending(日榜 + Python 日榜)
- 开源项目 hugohe3/ppt-masterhugohe3/ppt-masterGitHub Trending(日榜 + Python 日榜)
- 开源项目 google/skillsgoogle/skillsGitHub Trending(日榜 + Python 日榜)
- 开源项目 public-apis/public-apispublic-apis/public-apisGitHub Trending(日榜 + Python 日榜)
- 开源项目 MadsLorentzen/ai-job-searchMadsLorentzen/ai-job-searchGitHub Trending(日榜 + Python 日榜)
- 开源项目 multimodal-art-projection/YuEmultimodal-art-projection/YuEGitHub Trending(日榜 + Python 日榜)
- 开源项目 calesthio/OpenMontagecalesthio/OpenMontageGitHub Trending(日榜 + Python 日榜)
- 开源项目 pytorch/pytorchpytorch/pytorchGitHub Trending(日榜 + Python 日榜)
- 开源项目 huggingface/transformershuggingface/transformersGitHub Trending(日榜 + Python 日榜)
- 开源项目 home-assistant/corehome-assistant/coreGitHub Trending(日榜 + Python 日榜)
- 开源项目 OpenBMB/VoxCPMOpenBMB/VoxCPMGitHub Trending(日榜 + Python 日榜)