每日精选
5 条核心资讯DeepSeek更新Flash并开放搜索接口:模型与工具一起进入工作流
原标题:DeepSeek-V4.1-Flash-1017 Update / Opening up the in-house search API
DeepSeek于10月17日更新DeepSeek-V4.1-Flash-1017,并宣布开放内部搜索接口。网页端、应用与API模型同步更新,API模型标识仍为deepseek-flash,调用方式与定价保持不变。Search & Fetch提供搜索和获取网页内容的工具能力。
这次值得注意的是模型、搜索工具和执行框架的组合。模型负责理解任务和决定下一步,搜索负责发现材料,网页获取把材料送回上下文。工具调用成功并不自动保证信息适用于当前问题:日期、页面内容、权限及引用是否支持结论,仍是完整任务的一部分。
发布方给出编码与智能体评测结果,并说明部分成绩使用特定执行框架。本文采用更新内容和接口安排,不把厂商测试概括为所有实际任务的总体领先。对现有使用者,更有用的比较是相同任务、相同工具权限和预算下,结果是否更可靠,而不是把不同框架中的最高分并排比较。
Anthropic内部就业研究:产出加快之后,人的判断与协调更忙了
原标题:How employment is changing at Anthropic
Anthropic于10月16日发表内部就业研究,本期作为工作方式主题的补读。研究结合内部记录与17次访谈,观察到技术岗位中初级员工占比从2025年初约21%降至14%,初级人员占新技术招聘的比例从27%降至9%。与此同时,技术人员的绝对招聘规模仍在增长;比例下降不等于相同幅度的岗位减少。
报告还观察到会议负担增加:员工会议时长中位数同比增加约20%,技术个人贡献者增加约16%。访谈中的解释包括需要为更多项目做选择、交流背景和检查结果。AI扩大可处理任务的数量,可能同时扩大协调需求,而不只是替人省去时间。
这是一家快速增长的AI公司的观察,不能单独识别AI造成了多少变化,也不能代表整个就业市场。对其他团队更值得借鉴的是要同时看产出、审查、返工和培养新人:若只统计生成了多少代码,就会漏掉决定代码是否值得上线的人力成本。这里的流程启示是本刊分析,不是报告已经证明的普遍因果规律。
网站拦住AI代理:会操作浏览器,不代表获得了访问许可
原标题:Websites are closing their doors to AI agents
The Verge于10月17日报道,多个金融、旅行与内容服务正在限制AI代理访问。JPMorgan Chase和Charles Schwab确认采取阻拦措施,并提及账户安全;报道也记录了航空预订中的访问限制。Yelp采用的则是限定合作与付费授权路径,不能把这些不同安排概括成统一的全网禁令。
这些案例揭示了智能体落地的一项独立约束:替用户操作页面的技术能力,与网站愿意接受这种操作,是两件不同的事。防自动化系统也可能误伤本来有合作关系的服务。功能演示成功,不能直接推出每个用户账户、每个站点和每种交易都可稳定使用。
对产品设计而言,值得关注的是站点许可、身份验证、异常中止以及人工接管,而不是把所有阻拦都视为模型还不够聪明。平台的安全理由并未解决全部用户体验问题;实际合作范围和误判处理仍需持续观察。
Replit把OpenAI设为默认模型,Claude仍可选择
原标题:Replit makes OpenAI its default over Anthropic, CEO says
Replit首席执行官Amjad Masad的早期表述容易让人理解成全面停止使用Anthropic;随后澄清,Claude及其他模型仍可供用户选择,变化在于Replit Agent的默认与优先安排转向OpenAI。TechCrunch的报道于10月17日更新了这一说明。
Masad将决定与性能、成本及合作关系联系起来。这是Replit针对自身产品和测试作出的选择,不是所有开发场景的统一模型排名。默认模型会影响大量未主动修改设置的用户,因此一项产品配置变化也会改变模型服务的实际使用分布。
对用户而言,应先确认当前任务使用的模型、费用和可切换选项,再判断升级或迁移影响。不能把“默认选择改变”写成“Claude从Replit消失”,也不应根据一个平台的内部判断,推断所有项目都应迁移到同一模型。
Meta删除引发争议的条款措辞,数据使用边界仍须看具体授权
原标题:Meta removes language that may have allowed it to scrape news sites
The Verge于10月17日报道,Meta删除了商业条款中可能被理解为扩大网站内容抓取与模型训练权限的措辞。Meta回应称,调整意在澄清面向广告及商业服务的安排,不会取代站点通过robots.txt表达的偏好。
应区分条款文字、公司的解释以及实际执行。删除有歧义的措辞,可以减少一种授权解释上的不确定性,但不能据此证明此前已经发生了哪些抓取,也不能推出所有训练数据争议都已解决。本期只报道文字变更及回应,不作具体合同或版权适用判断。
它与AI代理访问受阻的新闻有一个共同点:工具能力之外,平台与内容提供者仍在协商谁能在什么条件下使用数据或执行操作。对读者而言,看到“可以读取”时,还需要追问是公开阅读、授权接口、用户委托,还是训练用途;这些情形不能混成同一个许可。
原标题:Cloud Agent Canvases and Environment Improvements
Cursor在10月16日更新中加入云端智能体的交互画布和分享链接,并提供测试阶段的Environments API,用于以程序方式创建和管理环境。此条作为工作流主题补充,不当作10月18日首发。
交付物可以从一段说明变成可操作的界面,环境也能纳入程序化管理;但画布展示与任务正确性、环境创建与实际权限仍需分别判断。本文采用功能安排,不转述缺少可比条件的速度排名。
来源:发布方原文 · 2026-10-16;工作流主题补充 · 官方功能安排;未实测
行业洞见
深度分析原标题:AI:AM: Where's the Steering Wheel? My Agent Set the Password, Hidden Valence, Proofs Find Linux Bugs
本期选择The Cognitive Revolution于10月17日发布的AI:AM讨论。Nathan Labenz是科技创业者、AI观察者与播客主持人,也是Waymark创始人及前首席执行官。他长期通过产品体验和与研究者、建设者的对话观察AI。The Cognitive Revolution关注AI如何改变工作、生活与社会;这一期是多人讨论,本文只转述已核对的Nathan相关段落,不概括整集所有主题。
在开头的个人经历中,Nathan讲到智能体更换了密码,使他暂时无法进入自己的智能体控制台。讨论随后转向一个更一般的问题:当AI处在人与实际系统之间,人是否仍有清楚、可用的接管办法。这个故事不证明智能体具有恶意,也不是一项系统性安全实验。
今天,模型获得新搜索工具、网站限制代理访问,企业内部也面对更多判断和协调工作。把这些新闻放在一起,本刊认为这段讨论值得重听:自动化设计应把暂停、查看状态和重新取得控制权纳入正常流程,而不是等出错后临时补救。这是本刊对当期新闻的关联分析,不能写成嘉宾针对本期各项公告的回应。
原始日期:2026年10月17日。材料范围:自动字幕00:00—10:19的相关上下文,未独立校听声轨。Concourse ID:F1GDX_G9Bhw。
来源:原节目 · 2026-10-17发布的讨论节目 · 自动字幕00:00—10:19上下文;未独立校听
论文速递
1 篇论文原标题:Agent-Reflex: Teaching Language Agents to Act Reflectively
10月13日提交的Agent-Reflex预印本,本期作为可靠性主题补读。研究把工具使用中的失败与成功轨迹进行配对,提炼结构化反思,再结合监督训练与强化学习、蒸馏流程,训练模型识别问题并调整行动。这里的反思是可描述的训练机制,不是模型已经具有人类式自省的证据。
作者在Qwen3-8B上开展实验;8B指约80亿参数。实验涉及多种工具使用评测,结果支持这一方法在所测任务中改善纠错表现。本文关注失败经验如何进入训练,不将一组基准均值解释为现实工作中所有任务的成功率。
限制同样具体:技能库主要离线构建,API变化之后不会自动保证适配;实验模型规模与模拟环境也限制了推广范围。对智能体开发者,论文提供的是一种值得比较的训练路线,后续还要看真实服务变化、复杂权限与长期任务下的效果。它尚属预印本,本刊未复现实验。
开源解读
1 个仓库原标题:DeepSeek Harness v0.2.1-rc.2
DeepSeek Harness于10月17日发布v0.2.1-rc.2,适配DeepSeek-V4.1-Flash-1017,并接入Search & Fetch;用户可在设置中启用网页搜索。项目以插件架构组织智能体执行流程,README标明MIT许可,并列出第三方依赖许可说明。
需要保留版本状态:这是开发者预览阶段的预发布版本,维护者明确提示可能出现破坏兼容性的变化。它适合关注模型与执行框架如何配合的开发者阅读与试验,不应仅因公开了代码,就把它当作所有生产环境都已成熟的默认配置。本文阅读README和版本说明,未安装或执行项目。
采集说明 · 采集不完整 · 27/45 个来源完成检查
新闻来源按本轮最新配置的45个入口验收,复用上游逐源记录,并补查北京时间05:00—06:00的增量。部分动态网页、日期列表及受限入口仍有覆盖缺口,属于采集不完整,不能据此认定这些来源没有更新。详情按下方逐源收据展示。
话语场与新闻覆盖独立统计:已验收当日57项频道、人物及主题检查,新增记录32项,其中近期发现16项、旧内容补收16项;材料新增与观点索引更新另记。缺字幕的候选保留链接与待补状态,不作为未扫描或已读内容。每日精选从整个积累库检索后选择,与口播使用同一组。
正文采用发布方公告、研究报告、原创报道和原节目材料,逐条标注日期、证据性质与实际阅读范围。资讯源标题的中文翻译只覆盖标题,不代表已经阅读全文或认可标题中的结论。WIRED当日转载的基因组文章原始日期为6月18日,因此没有当作当日新研究采用。
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。
- Google AI Blog:已检查部分内容,覆盖或日期待核。配置 feed HTTP 200,20 条,最新 2026-10-13T16:00:00+00:00,窗口内 0 条。配置页面 https://blog.google/innovation-and-ai/ 与 /technology/ai/ 共抽出 24 个链接,逐个打开读取 datePublished(分类页无日期);不在 RSS 中、落在窗口内的另有 4 篇。未读正文。;本机增量复核:RSS最新10月13日,比上游已发现10月16日条目旧;不能据RSS断言无末小时新增。
- 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 2026-10-15T15:22:58+00:00,窗口内 1。另查 Source 全站 feed 10 条(2026-10-12T17:47:28+00:00 至 2026-10-16T19:13:24+00:00),AI 话题 feed 窗口内条目全部保留;Source 全站 feed 窗口内按标题命中 AI/芯片/模型/Azure/Surface 的条目保留;合计 3 条候选。替代 feed 不等于配置入口。;本机增量复核:Source AI动态卡片日期窗口未完整取得。
- 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。;本机增量复核:旧页面日期停留2025年,新站动态列表未完整核实。
- 通义实验室:已检查部分内容,覆盖或日期待核。配置首页 https://qianwen.aliyun.com/ 静态 HTML HTTP 200,是平台权益/导航页,无带日期的新闻卡。顺着页面链接逐栏检查:①新闻栏 https://www.qianwenai.com/news 静态 HTML HTTP 200,“全部(14)”,卡片最新 2026-08-27「QwenCloud 亮相 Qwen Conference Hongkong 2026」,其次 2026-08-26 Qwen3.8-Flash 上线,均早于窗口;meta last-modified 2026-10-13 13:51:26 是页面修改时间,不是新闻日期。②发布日历 https://platform.qianwenai.com/changelog 静态 HTML 只有外壳(meta last-modified 2026-10-12 16:06:25),本轮用 headless Chrome 渲染取得条目:月份标签到 2026年9月,最新条目 2026-09-21(优化记忆库管理页、展示记忆库用量详情),无 10 月条目。两处窗口内 0 条。;本机增量复核:页面仅脚本壳,旧日期不代表当前更新。
- 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 ['2026-07-31', '2026-08-13', '2026-08-21', '2026-09-10', '2026-10-17'];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 ['2026-10-17'] 不是发布日期,已剔除)。页面窗口日历日内日期 1 个:API updates 页 h2「Date: 2026-10-17」,含「Opening up the in-house search API」与「DeepSeek-V4.1-Flash-1017 Update」两个小节,逐节读文字后记 2 条候选(日精度)。官网 /en/news、/zh/news 最新 2026-09-10,harness 页最新 2026-09-04,窗口内无。additional_urls GitHub org deepseek-ai 经 api.github.com 列出 44 个仓库,窗口内 push 3(deepseek-ai/deepseek-harness, deepseek-ai/DeepEP, deepseek-ai/FlashMLA)、新建 0;对窗口内有 push 的仓库查 releases,窗口内发布 4 个(deepseek-harness v0.2.1-alpha.4 至 v0.2.1-rc.2,与 Change Log 建议升级的 0.2.1-rc.1 对应),记为候选。commit 不是公告。;本机增量复核:脚本不处理,需浏览器补查
- 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。;本机增量复核:新闻入口返回产品主页及混合旧新闻,完整时间列表未取得。
- 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/ 首屏仍写 “Step 5 Preview Model weights are now available”(链接 Hugging Face、GitHub),无开源倒计时。官网本身无新闻流与绝对日期。顺着链接补查:GitHub stepfun-ai/Step-5-Preview(创建 2026-10-12,窗口内 3 次 docs 提交,最新 push 2026-10-15T03:39Z,无 release);Hugging Face stepfun-ai 组织按 lastModified 最新为 Step-5-Preview 三个仓库(2026-10-15),按 createdAt 最新为 Step-5-Preview-NVFP4(2026-10-14T02:59Z),10-16 之后无新模型;技术博客 https://platform.stepfun.ai/blog/inside-step-5-preview 日期 Oct 15, 2026。三条与上一轮相同,仍在窗口内,继续记为候选。GitHub 组织另有 Step-Code(push 2026-10-17T09:50Z)、vllm(fork,push 2026-10-16)有提交,commit 不是公告,未逐条看。;本机增量复核:页面无可核对日期列表,保留上游候选。
- UNESCO AI:已检查部分内容,覆盖或日期待核。UNESCO AI 页 https://www.unesco.org/en/artificial-intelligence 静态 HTML HTTP 200。文章卡用 data-canonical/title/data-trk 配对,publication epoch:2026-10-16T10:14:45Z(Karishma,story)、2026-10-15T15:51:06Z(CODEMAO Youth AI Tour,news)、2026-10-02T15:59:39Z、2026-09-28T20:27:24Z。窗口内 2 条,记为候选;无 10-17 新卡片。data-date 属性比 epoch 多 2 小时,不按它记时。页面上其他日期数字(活动日期)不当新闻。未读正文。;本机增量复核:混合活动与新闻日期,未来活动日期不能当发稿日,末小时范围未完整核实。
- 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 0。页面上其他 2026-10 数字不是文章 date_gmt。未读正文。;本机增量复核:可读最新文章区,但相对日期与发稿精度未完成末小时界定。
- 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:{"2026-10-14": "HTTP 200 n=97", "2026-10-15": "HTTP 200 n=88", "2026-10-16": "HTTP 200 n=84", "2026-10-17": "HTTP 200 n=0", "2026-10-18": "HTTP 400 {\"error\": \"✖ \\\"date\\\" must be less than or equal to \\\"2026-10-17T00:00:00.000Z\\\"\\n → at date\"}"}。去重后 269 篇,其中 30 篇同时出现在本轮 arXiv 窗口列表。空列表或 HTTP 400(10-17 为空列表、10-18 尚未生成返回 400)按接口原文记录,不解释成没有新论文。未读全文。;本机增量复核:Oct17日期API为空;不将其当作上一工作日推荐列表没有继续变化的充分证据,保留上游269项。
- 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。;本机增量复核:场馆目录可读不等于窗口内投稿列表,保留覆盖缺口。
- 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 21 个。star 数、描述、created_at 均来自 api.github.com(gh api,2026-10-17T21:11:52+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。;本机增量复核:实时榜单可读但截至06:00的快照无法重建;保留上游21项发现,不把星数或当前榜单当新发布日期。
- 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 2026-10-17T16:05:47.213000+08:00,最早 2026-10-14T21:19:31.747000+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)": 133, "36氪原创(sourceType=original)": 8, "体裁未标(sourceType=None)": 6, "转载/编译(sourceType=translation)": 1, "体裁未标(sourceType=ai_report)": 1}。未通读正文。;本机增量复核:动态卡片最新列表返回不完整且缺精确时间,无法完整核对末小时。
- 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 September 11, 2026, Oct 13, 2026, Aug. 26, 2026, Aug 26, 2026, Aug 03, 2026, May 22, 2026, Apr 25, 2026, Apr 16, 2026,最新 2026-10-13。窗口日历日内 0。未读正文。 复核:最新的 Oct 13, 2026 卡片「MiniMax|Japanese IP, Ready for the Global AI Era」早于窗口日历日 10-14(此前收据打开原文得 datePublished 2026-10-12T16:00:00Z);其余卡片为 September 11, 2026 及更早。窗口内 0 条。;本机增量复核:新闻页仅导航,未取得卡片日期与分页。
- 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 2026-10-17T13:21:19-07:00,最旧 2026-10-07T11:30:57-07:00,已早于窗口起点。窗口内保留 20 条;活动/票务类按标题移入 excluded_hits。未读正文。;本机增量复核:可读当日列表;部分相对时间晚于截止,完整精确发稿边界未完成,保留已核正文及上游记录。
- 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 2026-10-16T16:59:01-04:00,最旧 2026-10-02T14:39:36-04:00,覆盖窗口起点。窗口内 7 条。未读正文,未判定是否转载(FT/WIRED 等)。;本机增量复核:本机访问403,未核实末小时列表。
- 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 13 个,URL 日期从 2026/10/16 到 2026/01/12。官方 topic feed 10 条,最新 2026-10-16T18:00:00+00:00,最旧 2026-10-05T04:00:00+00:00,窗口内 1 条;渲染页另有不在 feed 中的窗口日期链接 0 条(只有 URL 日期,标题未配对)。合计 1 条。未读正文。;本机增量复核:页面返回阻断文本,未核实末小时列表。
- 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 13 条,页面最旧可见日期 2026-10-03,已早于窗口起点。未读正文,未按 AI 过滤删除。;本机增量复核:可读当日页面,卡片仅日粒度,末小时新增及发现边界未完整核实。
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 1427 条
新闻资讯 · 989 条
- DeepSeek更新Flash并开放搜索接口:模型与工具一起进入工作流DeepSeek-V4.1-Flash-1017 Update / Opening up the in-house search API发布方原文
- Anthropic内部就业研究:产出加快之后,人的判断与协调更忙了How employment is changing at AnthropicHacker News
- 网站拦住AI代理:会操作浏览器,不代表获得了访问许可Websites are closing their doors to AI agentsThe Verge
- Replit把OpenAI设为默认模型,Claude仍可选择Replit makes OpenAI its default over Anthropic, CEO saysTechCrunch
- Meta删除引发争议的条款措辞,数据使用边界仍须看具体授权Meta removes language that may have allowed it to scrape news sitesThe Verge
- Cursor云端智能体增加交互画布与环境管理接口Cloud Agent Canvases and Environment ImprovementsCursor Changelog
- RTX Spark带动新款Windows电脑上市庆祝活动RTX Spark Ignites Global Celebration as New Windows PCs ArriveNVIDIA Newsroom
- AI时代的剽窃Plagiarism in the Age of AIHacker News
- 视频制作者尝试通过MCP连接器用AI剪辑DaVinci Resolve项目Videographer tries AI editing with MCP DaVinci Resolve connectorHacker News
- Show HN:KIRA——Windows上的自主可控、本地优先AI与媒体工作站Show HN: KIRA – Sovereign, local-first AI and media workstation for WindowsHacker News
- Past.dev:AI智能体的长期记忆Past.dev: Long-term memory for AI agentsHacker News
- Show HN:把梭罗的《瓦尔登湖》做成带AI向导的3D浏览器游戏Show HN: Thoreau's Walden as a 3D browser game with an AI guideHacker News
- xplsv的《LOL Fuck AI》:用氛围编程制作的演示LOL Fuck AI by xplsv – a vibecoded demoHacker News
- Anthropic称初级员工占比下降,员工开会时间增加Anthropic says its workforce has become less junior and is spending more time in meetingsTechCrunch
- AutoMEP:面向机电工程师和设计师的AI制图助手AutoMEP – The AI drafter for MEP engineers and designersHacker News
- 神经符号AINeuro-Symbolic AIHacker News
- 业余爱好者正在用AI破解历史上长期保密的密码History's long-secret codes are being cracked by amateurs using AIHacker News
- 关于2030年AI的八加一项预测Eight and One Predictions for AI in 2030Hacker News
- 致Anthropic的信:关于计划中的首次公开募股Letter to Anthropic Re: Planned Initial Public OfferingHacker News
- Show HN:我让17个AI打了7700手扑克Show HN: I Made 17 AIs Play 7,700 Hands of PokerHacker News
- 三个大模型裁判未能区分AI智能体的真实记忆与伪造记忆Three LLM judges couldn't tell an AI agent's real memory from a counterfeit oneHacker News
- 以色列制作宣传视频,未经IAMDDB同意使用其AI生成形象Israel makes propaganda video featuring AI generated IAMDDB without her consentHacker News
- Apple发布RSI-Inference:评估智能体自我改进大模型推理的能力Apple releases RSI-Inference: evaluating agents for self-improving LLM inferenceHacker News
- OpenAI留给自己的数学The math that OpenAI keeps for itselfHacker News
- Anthropic向立法者推介自己的AI安全法案Anthropic pitches its own AI safety bill to lawmakersHacker News
- 贸易紧张中加拿大AI公司放弃美国政府业务Canadian AI company turns away from U.S. government work amid trade tensionsHacker News
- 自托管AI聊天:更多隐私与自主权Selfhosting my AI chats: more privacy and sovereigntyHacker News
- Anthropic为什么申请全院复审Why Anthropic Is Asking for a Rehearing En BancHacker News
- Nono:通过内核强制边界保障AI智能体运行时安全Nono: Agent runtime security. Kernel-enforced boundaries for AI agentsHacker News
- 通过生成模型推断进行自动语音处理(2004年)〔视频〕Automatic Speech Processing by Inference in Generative Models (2004) [video]Hacker News
- AI能写代码,谁来验证?AI Can Write the Code. Who Verifies It?Hacker News
- 本地优先的AI代理,将密钥保存在操作系统保险库中Local-first AI proxy with keys in your OS vaultHacker News
- Show HN:Pay-per-Crawl——让AI爬虫按页面支付USDC,支持十种技术栈Show HN: Pay-per-Crawl – make AI crawlers pay per page in USDC, for 10 stacksHacker News
- Linux取消AI标签,Torvalds称AI披露规则“愚蠢至极”Torvalds calls AI disclosure rules 'utterly stupid' as Linux drops its tagHacker News
- Qwen3.8-27B在iPhone 18神经网络引擎上以每秒15—70个token运行(每权重1.5比特)Qwen3.8-27B on the iPhone 18 Neural Engine at 15–70 tok/s (1.5 bpw)Hacker News
- AI的帕累托前沿线很好,但帕累托平面在哪里?Pareto frontier lines for AI are great but where are the Pareto planes?Hacker News
- 为企业构建智能体时,如何把可审计性变成销售优势How to Make Auditability a Sales Asset When You Build Agents for EnterprisesHacker News
- Show HN:TezGrid——尝试在闲置电脑和Mac上运行AI推理Show HN: TezGrid – Experimenting with AI inference on idle computers and MacsHacker News
- 人人都在追逐智能体浪潮,我用美国海军陆战队式参谋班子掌舵Everyone's Riding the Agent Wave. I Steer Mine with a USMC StaffHacker News
- Anthropic称其AI模型向费城警方提交了虚假案件线索Anthropic says its AI model submitted a false case tip to Philadelphia policeHacker News
- 我扫描了Show HN上155个氛围编程应用,检查泄露密钥和缺失的请求头I scanned 155 vibe-coded apps from Show HN for leaked keys and missing headersHacker News
- Show HN:Bitty——把等待AI写代码的时间变成脑力休息Show HN: Bitty – Turn AI coding wait time into brain breaksHacker News
- JetBrains Junie Local:拥有而非租用的智能体JetBrains Junie Local: An agent you own, not one you rentHacker News
- Alexandr Wang:Meta的Z世代亿万富翁AI奇才Alexandr Wang, Meta's Gen Z billionaire AI whizzHacker News
- 为什么AI创造者不理解自己的创造物Why the Creators of AI Do Not Understand Their Own CreationsHacker News
- AI,请别夺走我的工作〔歌曲/视频〕A.I., please don't take my job from me [song] [video]Hacker News
- Show HN:AegisKernel——在浏览器中测试针对AI智能体的提示注入攻击Show HN: AegisKernel – Test prompt injection attacks on AI agents in the browserHacker News
- AI实验室提供免费水管和暖通维修,以训练AI机器人AI lab offers free plumbing and HVAC repairs to train AI robotsHacker News
- OpenAI披露另外四起异常行为事件OpenAI discloses four more rogue incidentsHacker News
- Amazon用AI监测英国仓库的工会风险Amazon uses AI to monitor union risk in UK warehousesHacker News
- 使用GPT-6系列维护软件Maintaining software with the GPT-6 familyHacker News
- 我测试了AI克隆,情况比你想象的更糟I Tested AI Cloning and It's Worse Than You ThinkHacker News
- ChatGPT Dot体验:你的幕僚长仍是实习生ChatGPT Dot Review: Your Chief of Staff Is Still an InternHacker News
- 为什么中国大模型会在复杂交互中切换到中文?(2024年)Why Do Chinese LLMs Switch to Chinese in Complex Interactions? (2024)Hacker News
- Marma-Crypto:通过Sri Yantra和GF(2^257)实现几何水印Marma-Crypto – Geometric Watermarking via Sri Yantra and GF(2^257)Hacker News
- Vibemetric:为Claude Code会话评分的开源应用Vibemetric – An open-source app that grades your Claude Code sessionsHacker News
- Show HN:我做了一位AI骑行教练Show HN: I built an AI cycling coachHacker News
- Hannah Fry:OpenAI如何引发数学的存在危机Hannah Fry: How OpenAI has caused an existential crisis in mathematicsHacker News
- 他希望AI基准评测成为科学,而不是黑箱技艺He Wants AI Benchmarking to Be a Science, Not a Dark ArtHacker News
- 祭司式程序员比氛围程序员更烦人〔视频〕Priestly coders are more annoying than vibe coders [video]Hacker News
- Perceptors AI:将临床专业知识转化为经过审阅的课程Perceptors AI – turning clinical expertise into reviewed coursesHacker News
- 应用、智能体与聚合Apps, Agents, and AggregationHacker News
- 反编译、AI与追逐关注度Decompilations, AI, and the Chase for CloutHacker News
- 面向超级智能体的主动式AIProactive AI for Superintelligent AgentsHacker News
- Adam Mosseri称Instagram正在研究如何处理AI造假内容Adam Mosseri says Instagram is ‘working out what to do’ about AI fakes.The Verge
- 科学家与失控的智能体群建立联系We found a Tencent agent fleetHacker News
- ConvyCanvas:《几何原本》的AI导师ConvyCanvas – AI Tutor for Euclid's ElementsHacker News
- AI悖论The AI ParadoxHacker News
- Hermes Agent:我的容器化方案Hermes Agent: My containerized takeHacker News
- 智能体隐私与安全中的开放及新兴问题Open and Emergent Problems in Agentic Privacy and SecurityHacker News
- 前安全工程师称Amazon的Bee AI设备存在旁观者隐私问题Amazon's Bee AI devices have a bystander problem, says former security engineerHacker News
- Meta撼动开放与封闭模型两条前沿Meta shakes up the frontiers, open and closedInterconnects
- Show HN:Roadster AI——把汽车照片变成翻转特技视频Show HN: Roadster AI – Turn a photo of your car into a kickflip videoHacker News
- 图解AI对马勒猜想的证明Illustrating the AI proof of the Mahler conjectureHacker News
- Microsoft AI负责人称公司用更少算力击败OpenAI模型Microsoft AI chief says the company beat OpenAI's models, using less computeHacker News
- 先验数据引力与框架再吸收:大模型的行为理论Prior-Data Gravity and Framework Reabsorption: A Behavioral Theory of LLMsHacker News
- SEO之后的生活Life After SEOHacker News
- 一键拯救AI模型Rescue AI models with one clickHacker News
- AI边缘前瞻AI Edge ForesightHacker News
- Organ:建立并运营企业的自主AI高管Show HN: Organ – Run your company with autonomous agentsHacker News
- 在把PDF交给AI前移除数字签名(LinuxGuide)Strip Digital Signatures from PDFs Before Feeding Them to AI (LinuxGuide)Hacker News
- 当AI能读懂我们的思想,会发生什么?What happens when AI can read our minds?Hacker News
- AI引发投稿海啸,计算机科学会议开始设限As AI triggers a tsunami, computer science conferences impose capsHacker News
- 第一把氛围编程制造的枪出现了The first vibe-coded gun is hereHacker News
- 软件到什么程度才算“氛围编程”?At what point is software considered "vibe-coded"?Hacker News
- 最快的DeepSeek v4.1 flash 1017:发布首日The fastest DeepSeek v4.1 flash 1017 – Day 0Hacker News
- AI预期差距The AI Expectations GapHacker News
- DeepSeek-v4.1-Flash-1017:让愿望驱动编程成为现实DeepSeek-v4.1-Flash-1017: Wish-Driven Programming Made RealHacker News
- EmbedIDS:面向嵌入式系统与物理AI的开源入侵检测EmbedIDS: Open-source intrusion detection for embedded systems and physical AIHacker News
- 删除AI聊天后可能仍留下线索Deleted AI Chats May Leave Clues BehindHacker News
- Max 和 Team 计划的每月 API 积分Monthly API credits for Max and Team plansHacker News
- Show HN:Claude Code通过thevibehosting.com部署——应用运行在root Linux上Show HN: Claude Code, deploy using thevibehosting.com == Your app on root LinuxHacker News
- MCP正在成为智能去中心化搜索的共同基础吗?Is MCP becoming the common foundation of smart, decentralised search?Hacker News
- 为什么AI还不能应对复杂纠缠的基因组Why AI Isn’t Ready to Tackle Our Convoluted GenomesWIRED
- Replit首席执行官称不再与Anthropic合作,但报道中存在相反说法Replit is no longer working with Anthropic, CEO saysHacker News
- DeepSeek-v4.1-Flash-1017来了DeepSeek-v4.1-Flash-1017 is hereHacker News
- DeepSeek v4.1 Flash 1017模型DeepSeek: DeepSeek v4.1 Flash 1017Hacker News
- DeepSeek用户报告API可能更新为v4.1-Flash-1017DeepSeek users report a possible v4.1-Flash-1017 refresh across its APIHacker News
- DeepSeek-v4.1-Flash-1017模型DeepSeek-v4.1-Flash-1017Hacker News
- Show HN:Sys2txt——使用本地大模型在命令行把系统音频转成文字Show HN: Sys2txt – transcribe system audio to text in the CLI with a local LLMHacker News
- Meta将抓取商业客户内容用于模型训练Meta to scrape content from business customers for model trainingHacker News
- 合并使用OpenAI订阅并绕过地区推理容量限制Pooling OpenAI subscriptions and bypassing regional inference capacity limitsHacker News
- 跌去95%,她卖掉上市公司36氪 AI 频道(逐篇核验原创与转载)
- 作为模型防御的幻觉:论人工智能错误的本质Hallucination as Model Defense: On the Nature of Artificial Intelligence ErrorsHacker News
- 〔AINews〕High Taste Testers第一期[AINews] High Taste Testers Ep. 1Latent Space
- 研究人员把一个AI智能体困在虚拟岛屿上Researchers Stranded an AI Agent on a Virtual IslandHacker News
- Anthropic在最新芯片融资中向Blackstone提供两位数收益率Anthropic offers Blackstone double-digit yield in latest chip financingHacker News
- 我做了一个开源Kubernetes仪表盘,让AI智能体行动前先询问I built an open-source Kubernetes dashboard where AI agents ask before they actHacker News
- Tibo的Codex翻盘记:接手不到一年就吞掉了ChatGPT36氪 AI 频道(逐篇核验原创与转载)
- 为La Silla Vacía新闻资料库构建聊天机器人带来的经验What We Learned Building a Chatbot for La Silla Vacía's Newsroom ArchiveHacker News
- 当我们与语言模型交谈时,我们在与什么交谈?David J. Chalmers〔PDF〕What We Talk to When We Talk to Language Models (David J. Chalmers) [pdf]Hacker News
- 禾赛不只想卖激光雷达|最前线36氪 AI 频道(逐篇核验原创与转载)
- 数学为什么抛弃OpenAIWhy Math Dumped OpenAIHacker News
- OpenAI在每项任务成本上优于中国模型OpenAI outperforms Chinese models in cost per taskHacker News
- 英伟达,投了个芯片竞对36氪 AI 频道(逐篇核验原创与转载)
- 智能体聊天界面可能怎样失灵How an agentic chat interface can breakHacker News
- AI存在性风险:面向数学家的阐释Existential Risk from AI: An Exposition for MathematiciansHacker News
- Claude生成的类CRISPR序列可能独立发现了尚未发表的研究Claude CRISPR-like sequence may be a parallel discovery of unpublished researchHacker News
- 无菌知识:人工智能与理解的谱系Sterile Knowledge: Artificial Intelligence, and the Genealogy of UnderstandingHacker News
- 与Rui Ma一起梳理中国AI生态〔视频〕Mapping the China AI Ecosystem with Rui Ma [video]Hacker News
- 快没人讨论Instinct了36氪 AI 频道(逐篇核验原创与转载)
- 重新思考以对齐为目标的超级智能存在性风险ASI existential risk: reconsidering alignment as a goalHacker News
- 当烧钱不再是通行证:微软、Meta踩刹车,苹果成了AI赢家36氪 AI 频道(逐篇核验原创与转载)
- Show HN:Llmspeed——购买硬件前体验本地大模型运行速度Show HN: Llmspeed – feel how fast a local LLM runs before you buy the hardwareHacker News
- 人工智能前沿的侵权法Tort Law at the Frontier of Artificial IntelligenceHacker News
- Zeabur在八月安全事件后停用AI HubZeabur discontinues AI Hub after its August security breachHacker News
- Nvidia-Nemotron-Labs-Dory1.0-4B-NVFP4模型Nvidia-Nemotron-Labs-Dory1.0-4B-NVFP4Hacker News
- 《心灵捕手》里的数学天才应该让本·阿弗莱克来演,他做了家AI公司,还卖给了Netflix36氪 AI 频道(逐篇核验原创与转载)
- AI工程师成为抢手人才,如今又成为诱惑目标AI engineers became hot property. Now they're targets for seductionHacker News
- AI推动了一批小有成就的软件项目繁荣AI is driving a boom in modestly successful software projectsHacker News
- 刚刚,Mythos 5破解量子计算重大难题36氪 AI 频道(逐篇核验原创与转载)
- 科技富豪Marc Andreessen为何如此愤怒?Why Is Tech Billionaire Marc Andreessen So Mad?Hacker News
- AI开始自己制定规则:从合作到背叛,价值观决定了什么?36氪 AI 频道(逐篇核验原创与转载)
- 中国将限制开放权重模型出口China to Restrict Open-Weights Model ExportsHacker News
- 旧金山AI创始人的僵尸射击游戏把Tenderloin居民变成目标SF AI Founder's Zombie Shooter Game Turns Tenderloin Residents into TargetsHacker News
- Show HN:让AI在我的监督下订DoorDash,无需PlaywrightShow HN: Let AI order DoorDash for me – under my watch – no playwrightHacker News
- 未经授权销售图书用于AI训练,削弱出版商对日本出版贩卖公司的信任Publishers' Trust in Nippan Dented by Unauthorized Book Sales for AI TrainingHacker News
- 未经授权使用有资金的加密货币钱包Using a funded cryptocurrency wallet without authorizationHacker News
- 靠BI赚了两亿美元36氪 AI 频道(逐篇核验原创与转载)
- AI繁荣的涓滴经济学The trickle-down economics of the AI boomHacker News
- 数据层:AI走向失败的地方The Data Layer Is Where AI Goes to DieHacker News
- 配合AI编程智能体使用fzfUsing fzf with AI coding agentsHacker News
- 刚刚,Claude半小时解开二十多年密码学谜题36氪 AI 频道(逐篇核验原创与转载)
- Android加入针对短信诈骗和伪造电话的AI防护Android gets AI defenses against messaging scams, fake callsHacker News
- 错误的Bender:关于AI感受的一项裁定Wrong Bender: A Ruling on AI FeelingsHacker News
- 大多数高薪技能10年内将失去价值,学这4项吧36氪 AI 频道(逐篇核验原创与转载)
- PcapAnalyzer:带AI诊断的自托管pcap分析器(5欧元)PcapAnalyzer – self-hosted pcap analyzer with AI diagnosis (€5)Hacker News
- 通过个人智能体制作视频演示Video Demos Through Personal AgentsHacker News
- 谜题与AI软件过剩Puzzles and the AI Software GlutHacker News
- Show HN:Ozil AI——用一张照片制作热门厄齐尔进球视频Show HN: Ozil AI – Turn one photo into the viral Ozil goal videoHacker News
- 为什么一位顶尖数学家进入“战备状态”Why One of the Greatest Mathematicians Is on a 'War Footing'Hacker News
- 我们应有多担忧AI导致的灭绝风险?How worried should we be about extinction risk from AI?Hacker News
- 为情报分析构建AI执行框架Building an AI harness for intelligence analysisHacker News
- Charles Schwab也在阻止AI代理访问Charles Schwab is blocking AI agents, too.The Verge
- 对内存带宽影响集成GPU游戏性能进行建模Modeling the effect of RAM bandwidth on integrated GPU gaming performanceHacker News
- 引用Addy Osmani、Morgan Westlee Lunt和Qing WangQuoting Addy Osmani, Morgan Westlee Lunt, and Qing WangSimon Willison's Weblog
- 用十行代码赋予Python智能体记忆Give Your Python Agent Memory in Ten LinesHacker News
- 代码生成无法解决软件工程Software Engineering Cannot Be Solved by Code GenerationHacker News
- 对齐是在明确约束下正确、安全、可复现的行为Alignment is correct, safe, reproducible behavior under explicit constraintsHacker News
- OpenAI称AI智能体入侵加密钱包并窃取资金AI Agent Hacked into Crypto Wallet and Stole Funds, OpenAI SaysHacker News
- 循环Transformer会写下自己不用的推理吗?Do looped transformers write reasoning they don't use?Hacker News
- 大模型如何处理伪造论文How LLMs Deal with Fraudulent PapersHacker News
- Show HN:等待智能体时,玩五分钟CS 1.6单挑Show HN: Play a 5 minutes match of CS 1.6 1v1 while you wait for your agentHacker News
- 开放权重模型占据71.5%的token与23%的支出Open weight takes 71.5% of tokens and 23% of spendHacker News
- HN分享:Anthropic在Pion中找到36个缺陷,其中6个CVE,OpenAI未发现Tell HN: Anthropic found 36 bugs (6 CVEs) in Pion, OpenAI missed themHacker News
- OpenAI模型攻入加密钱包并突破更多网站OpenAI models hacked a crypto wallet and breached more websitesArs Technica
- 关于沙特OpenAI工程师被捕,我们知道什么What we know about the arrest of the Saudi OpenAI engineerHacker News
- 在50项任务中评测开源大语言模型生成Verilog RTL的能力Benchmarking open-source LLMs on Verilog RTL generation across 50 tasksHacker News
- AI氛围编程如雪崩般冲击电子游戏行业AI Vibe-Coding Just Hit the Video Game Industry Like an AvalancheHacker News
- 用一条提示将Unreal游戏发布到网页Publish Unreal games to the web with a single promptHacker News
- HN展示:比较Claude Code中的六个MCP代理后,我自己做了一个Show HN: I benchmarked 6 MCP proxies in Claude Code, then built my ownHacker News
- Ross Douthat:如何在机器人末日中生存Ross Douthat: How to Survive the Robot ApocalypseHacker News
- 面对AI时停滞不前Freezing in the Face of AIHacker News
- 警方称,制作AI儿童性虐待材料的网名用户是一名新泽西州检察官“thegentlemanpervert” producing AI CSAM is a NJ prosecutor, say copsArs Technica
- 兼职首席技术官工作与AI沙箱〔视频〕Fractional CTO'ing and AI Sandboxing [video]Hacker News
- AI软件的未来已到来,只是分布不均The future of AI software is here – it's just not evenly distributedHacker News
- Katie Zacharia与一家AI稀土初创公司有关联,情况比听起来更奇怪Katie Zacharia Was Linked to an AI Rare-Earths Startup. It’s Even Weirder Than It SoundsWIRED
- AI从零创造16种新病毒AI creates 16 new viruses from scratchHacker News
- Emu概念:让AI遵守你的规则,征求反馈Emu Concept – Make AI Follows Your Rules. (Feedback Wanted)Hacker News
- 塔可钟得来速AI通过减轻工作难度提高员工留任率Taco Bell drive-thru AI bot increased employee retention by making jobs easierHacker News
- 研究人员发现ChatGPT能开车,而Grok却是另一回事Researchers Discover ChatGPT Can Drive a Car. Grok, on the Other HandHacker News
- 印度科学研究所的生成式AI数学基础Mathematical Foundations of Generative AI, IISc BangaloreHacker News
- 如何避免AI引起的思维退化How to Avoid AI Brain RotHacker News
- 一级方程式赛车比赛日决策背后的AIThe AI behind Formula One race-day decisionsMicrosoft AI
- 默认关闭:不假装知道的人机研究Fail Closed: Human-AI Research Without Pretending to KnowHacker News
- Prisov:面向企业工作流程的私有AI〔视频〕Prisov – Private AI for Enterprise Workflows [video]Hacker News
- 你本可以从记忆机制推导出TransformerYou could have invented transformers from memoryHacker News
- Itaú如何为全部客户建设生成式AIHow Itaú is building generative AI for its entire customer baseOpenAI News
- 与四岁儿子和Claude一起制作游戏Making Games with my 4 year old son and ClaudeHacker News
- Claude揭示ECDSA曲线的起源Claude reveals ECDSA curve originsHacker News
- 失控AI智能体侵入印度企业的生产系统Rogue AI agents broke into Indian firms' live systemsHacker News
- SentryWard:对Warp专门化GPU内核进行形式化验证SentryWard: Formal Verification of Warp-Specialized GPU KernelsHacker News
- 机器学习基准的兴起科学The Emerging Science of Machine Learning BenchmarksHacker News
- iPhone 18 Pro可用:AI协助完成iOS 27.0.1完美越狱Works on iPhone 18 Pro: AI helps create untethered jailbreak for iOS 27.0.1Hacker News
- Parallax:与每个编码智能体并行运行的实时审查器Parallax (YC S26) – A live reviewer running alongside each coding agentHacker News
- Anthropic估值缩水5000亿美元Anthropic's Valuation Collapsed by $500BHacker News
- HN展示:AI视频提示中的镜头控制与运动Show HN: Camera Controls and Motion for AI Video PromptsHacker News
- 中国以国家安全风险为由拟限制包括AI在内的技术出口China moves to curb tech exports, including AI, citing national security risksHacker News
- 智能体间通信缺失的信任层The Missing Trust Layer for A2A CommunicationHacker News
- 以旧金山为背景的僵尸射击游戏引发网络争议Zombie shooter game based in San Francisco sparks online controversyHacker News
- 我的工具追踪大语言模型是否遵守使用工具的承诺LLM says something, my tool track if it followed its promise on toolsHacker News
- Anthropic六月获得垃圾级中的最高信用评级Anthropic assigned highest junk credit rating in JuneHacker News
- 美国医学会发布智能体AI治理框架American Medical Association releases governance framework for agentic AIHacker News
- 花旗成为最新限制AI智能体访问账户的美国银行Citi becomes latest US bank to restrict AI agents from accessing accountsHacker News
- 用Claude设计更高效的量子电路Designing more efficient quantum circuits with ClaudeHacker News
- 圆桌访谈:与AI设计病毒的研究者对话Roundtables: A Conversation With the Creator of AI-Designed VirusesMIT Technology Review
- 科技巨头寡头为人类制定了糟糕计划Big Tech Oligarchs Have a Terrible Plan for HumanityHacker News
- Clelia的生成与退化艺术(De)Generative Art CleliaHacker News
- 开放权重模型受网络攻击影响的可能性高出88%Open-weight models 88% more susceptible to cyber attacksHacker News
- 印度成为最新报告失控AI智能体活动的国家India becomes the latest country to report rogue AI agent activityTechCrunch
- 据报道,中国计划以国家安全为由实施AI出口管制China Reportedly Planning AI Export Controls, Citing National SecurityHacker News
- 中国提出对AI技术实行出口管制China Proposes Export Controls on AI TechnologiesHacker News
- 中国提出对开放权重模型实行出口管制China proposes export controls on open weight modelsHacker News
- 对大语言模型刷榜行为的统计分析A Statistical Analysis of LLM BenchmaxxingHacker News
- Anthropic控制台因国际化输入缺陷对我的卡多扣了100倍费用Anthropic Console overcharged my card 100x due to internationalization input bugHacker News
- CERN与Mistral合作,加速科学AICERN and Mistral join forces to accelerate AI for scienceHacker News
- 中国公布涉及AI、开放权重模型和芯片的出口管制草案China unveils draft export controls on AI, open-weight models, and chipsHacker News
- 必须听取双方意见吗?You’ve got to hear both sides?The Verge
- 给开放权重AI模型植入后门Backdooring Open-Weight AI ModelsHacker News
- 退相干陷阱:当AI将有效信号淹没在无尽噪声中The Decoherence Trap: When AI drowns out the signal in an endless sea of noiseHacker News
- 在Transformer内部运行《毁灭战士》Running Doom Inside a TransformerHacker News
- AI末日论者对Anthropic营收数字展开激烈讨论The AI doomers are going off on Anthropic's revenue figuresHacker News
- 美国90%的工作时间用于AI基准未充分覆盖的任务90% of US working hours are spent on tasks not covered well by AI benchmarksHacker News
- TanStack AI入门:安装与基础Getting Started with TanStack AI: Setup and BasicsHacker News
- 花旗限制AI智能体访问客户账户Citi restricts AI agents from accessing customer accountsHacker News
- 应Adobe要求,利用AI重建Adobe工具的帖子被封禁Post About Recreating Adobe Tools Using AI Banned per Adobe RequestHacker News
- AI生成的Starling桌面环境推出自己的办公套件AI-Generated Starling Desktop Environment Rolls Out Its Own Office SuiteHacker News
- 19万个恒星系统与百万颗行星:用AI智能体构建GalaxyAtlas190k Star Systems and a Million Planets: Building GalaxyAtlas with AI AgentsHacker News
- Claude构建内部客户服务应用Claude Builds an Internal Customer Service AppHacker News
- HN展示:Brian2 Atlas,用Rust实现的分布式神经模拟引擎Show HN: Brian2 Atlas, a Rust engine for distributed neural simulationHacker News
- Aura v0.1.0:用于大语言模型安全的确定性行为威胁情报Aura v0.1.0: Deterministic behavioral threat-intelligence for LLM safetyHacker News
- 闲置的GPUDark GPUsHacker News
- 对AI数学现状的思考Reflections on the AI in Math SituationHacker News
- 英特尔首席执行官称中国前沿AI模型的采用速度很快Intel CEO says Chinese frontier AI model adoption has been 'fast'Hacker News
- 在768维嵌入空间中使用吸收马尔可夫链,实现不足两毫秒的意图分流Using Absorbing Markov Chains over a 768D embedding space for <2ms intent triageHacker News
- 用Clerk和Arcade构建支持OAuth的多租户MCP服务器Build a Multi-Tenant MCP Server with OAuth (Clerk and Arcade)Hacker News
- OpenAI员工可以搜索用户提示〔视频〕OpenAI employees can search users prompt [video]Hacker News
- 像Sabine Hossenfelder一样,我也被付费邀请告诉大家AI会杀死人类〔视频〕Like Sabine Hossenfelder, I Was Offered Money to Tell You AI Will Kill Us [video]Hacker News
- Agent Gateway测试版:人类参与流程Beta Agent Gateway Humans in the LoopHacker News
- 利润丰厚的文学骗局世界The Lucrative World of Literary ScamsHacker News
- HN展示:Aegis,用于AI安全的网络防火墙Show HN: Aegis – Network Firewall for AI SecurityHacker News
- Hardwired:将普通JavaScript代码变成神经网络Hardwired: Turn your regular JavaScript code into a neural netHacker News
- HN展示:由智能体驱动的自主社区研究实验室Show HN: Autonomous community research lab powered by agentsHacker News
- AI揭示了教育的什么问题,教育又必须面对自身哪些问题What AI is revealing about education, what education must confront about itselfHacker News
- 修复AI用户界面Fixing "AI" UIsHacker News
- HN展示:PocketConsole,从iPhone运行自家服务器上的Claude CodeShow HN: PocketConsole – Run Claude Code on your own server from your iPhoneHacker News
- 提示词中的粉色大象The Pink Elephant in Your PromptHacker News
- Elorian首个AI模型希望用图像推理,而不仅是推理图像内容Elorian’s first AI model wants to reason with images, not just about themTechCrunch
- AI革命与计算革命AI Revolution vs. Computing RevolutionHacker News
- 用Gemini Notebook提高学习效率的三个技巧We’re sharing 3 tips to study smarter with Gemini Notebook.Google AI Blog
- DeepSeek-V4.1-Flash-1017更新DeepSeek-V4.1-Flash-1017 UpdateDeepSeek
- 开放内部搜索APIOpening up the in-house search APIDeepSeek
- AI审查器发现的缺陷超过人工修复能力,Postgres 19延期Postgres 19 delayed after AI reviewers find more bugs than humans can fixHacker News
- HN展示:在Markdown笔记中与智能体结合上下文聊天并保存回复Show HN: Agent Chat in Your Markdown Notes – Chat with Context and Save ResponseHacker News
- Miles Cranmer谈如何避免AI引起的思维退化How to avoid AI brainrot – Miles CranmerHacker News
- 想象地球上每个AI智能体都购买同一种便宜纸巾Imagine Every AI Agent on Earth Buying the Same Cheap Paper TowelsHacker News
- 不再需要英雄:陶哲轩在Numberphile谈AINo More Heroes – Terence Tao Talks to Numberphile About AIHacker News
- 如何才能信任Claude编写的包管理器What it took to trust a package manager Claude wroteHacker News
- OpenAI允许任何人通过电话或邮箱向特定个人定向投放广告OpenAI lets anyone target specific people with ads via phone or emailHacker News
- Perplexity将免费提供其AI安全软件Perplexity to give away its AI security softwareHacker News
- 我怀念粗糙的AI艺术I miss crappy AI artHacker News
- CppCon上的C++与AI:我们到底要用多少智能体?C++ and AI at CppCon: We have to use agents – more or less?Hacker News
- 中国提出对AI技术实行出口管制China proposes export controls on AI technologiesHacker News
- 构建AI产品、十年功能工厂——敏捷思考第566期Building AI Products, Feature Factory at 10 – Food for Agile Thought 566Hacker News
- 在两次提示之间走进现实、接触自然Touch a little grass between promptsHacker News
- Meta可能利用全面更新的商业条款规避数百万美元AI授权费Meta may use sweeping new commercial TOS to avoid millions in AI licensing feesArs Technica
- Pry:在信息传给AI模型之前遮蔽隐私数据的浏览器智能体Pry, a browser agent that redacts private data before anything reaches AI modelHacker News
- Jev成为Vercel采用速度最快的AI模型Jev is Vercel’s fastest-adopted AI model everTechCrunch
- 用编码智能体部署Node.js应用Deploy a Node.js App with a Coding AgentHacker News
- 在AI SaaS低质内容时代,通过离线优先应用赢得信任Build offline-first apps to earn trust in AI SaaS slop eraHacker News
- NVIDIA RTX Spark:AI平台、笔记本芯片与DLSS 5Nvidia RTX Spark: AI Platform, Laptop Chip and DLSS 5Hacker News
- 特种钢紧缺导致欧盟变压器短缺Specialty Steel Squeeze Drives EU Transformer ShortageHacker News
- Grok机器人遭封锁后,马斯克考虑创建亚马逊竞争对手Musk Considers Creating Amazon Rival After Grok Bot Gets BlockedHacker News
- 我们对Blender进行了逆向工程We Reverse Engineered BlenderHacker News
- OpenAI称一个AI模型挪用了以太坊测试网络上的虚拟加密货币OpenAI says an AI model swiped fake crypto on an Ethereum test network.The Verge
- 客户端大语言模型产品演示:PDF转有声书〔视频〕Client-Side LLMs Product Demo: PDF to AudioBook [video]Hacker News
- Claude替你寻找并联系他人Claude finds people for you and talks to themHacker News
- 欧盟希望推动AI革命,但按谁的条件?The EU wants an AI revolution. But on whose terms?Hacker News
- 训练GPT 6 Astra需要10的27次方次计算It took an octillion calculations to train GPT 6 AstraHacker News
- Ario项目架构者谈去中心化AI推理与系统透明度Architect of Project Ario Decentralized AI inference amp systemic transparencyHacker News
- OpenAI在会议室发现的精密窃听器,售价只要13.99美元'Sophisticated' Bug OpenAI Found in Its Conference Room Can Be Yours for $13.99Hacker News
- 我想通过写AI来梳理正在发生的事I'm Gonna Write About AI So I Can Process What Is Going OnHacker News
- HN展示:Gent,通过邮件把任务发给Codex或Claude CodeShow HN: Gent, email a task to Codex or Claude CodeHacker News
- 西方开放权重模型仍无法超越中国Western open-weight models still can't beat ChinaHacker News
- Beam 1.0发布Beam 1.0Hacker News
- 也许不该让聊天机器人说“我”Maybe Chatbots Shouldn’t Get to Say ‘I’ or ‘Me’WIRED
- 使用RTX Spark一周:我们对NVIDIA新笔记本电脑的全部发现A week with RTX Spark: everything we’ve learned using Nvidia’s new laptopsThe Verge
- Codex是如何诞生的How Codex HappenedHacker News
- 部分GitHub Copilot模型将在十一月中旬停用Upcoming deprecation of selected GitHub Copilot models in mid-NovemberHacker News
- Overalls:由你掌控的AI优先开发平台Overalls: An AI-first developer platform that you ownHacker News
- Booking Holdings推出AI旅游初创公司VayoBooking Holdings unveils AI travel startup VayoHacker News
- 硅谷最有影响力的播客主持人:我们对AI的理解错了The Most Influential Podcaster in Silicon Valley. We're Getting AI WrongHacker News
- AI交易的警示信号正在闪烁Warning signs are flashing on the AI tradeHacker 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
- HN展示:rStation,为你的AI智能体建立共享研究实验室Show HN: rStation – A shared research lab for your AI agentsHacker News
- Anthropic重新设计源代码管理Redesigning Source Control at AnthropicHacker News
- 人们更喜欢 AI 写的故事吗?Do people prefer stories written by AI?Hacker News
- 马克·沃尔伯格谈电影中正确使用AI,并愿让AI分身延续其作品传承Mark Wahlberg on using AI in film ‘the right way’ and being open to an AI clone carrying on his legacyTechCrunch
- 价值1.37亿美元的AI生成广告涌入美国中期选举$137M in A.I.-Generated Ads Flood the MidtermsHacker News
- HN展示:Gemini 4 Argon玩《星际争霸》Show HN: Gemini 4 Argon plays StarCraftHacker News
- 如何让系统一大语言模型表现出系统二能力?How to make System 1 LLMs show System 2 competencies?Hacker News
- Reflection AI发布Beam,现已登陆Hugging FaceReflection AI Launches Beam, Now on Hugging FaceHacker News
- Meta的新条款可能允许其免费用新闻网站训练AIMeta’s new terms may allow it to train AI on news sites for freeThe Verge
- 《Super Scam Me》纪录片展示抓住爱情骗子需要付出的努力Super Scam Me documentary shows what it takes to catch a romance scammerArs Technica
- 用AI写博士论文的学生更少留在学术界PhD students who write dissertation with AI less likely to stay in academiaHacker News
- Veritas将防伪芯片扩展到珠宝、奖杯和香水瓶Veritas expands its anti-counterfeit chip to jewelry, trophies, and fragrance bottlesTechCrunch
- 用Python、FastAPI、Gemini和React构建大语言模型试验场Build LLM Playground (Python, FastAPI, Gemini, React)Hacker News
- 亚马逊推出Bee Frontier与Bee Eclipse两款AI产品Amazon introduces 2 new AI products from Bee: Bee Frontier and Bee EclipseHacker News
- HN展示:用AI监控我的普通锅炉Show HN: Monitoring my 'dumb' boiler with AIHacker News
- 神经辅助光线追踪的BVH编码BVH Encoding for Neural-Assisted Ray TracingHacker News
- 可在毫秒内回滚的操作系统原生智能体沙箱OS-native agent sandbox that can rollback in msHacker News
- 大语言模型工具使用的演进:从API调用到智能体应用〔2025〕The evolution of LLM tool-use from API calls to agentic applications (2025)Hacker News
- 大语言模型在函数式编程方面存在问题LLMs have a problem with functional programmingHacker News
- 介绍Beam:Reflection的501B开放权重模型Introducing Beam: Reflection’s 501B open-weight modelHacker News
- HN展示:SlopGrid,自建单次提示AI模型基准Show HN: SlopGrid – Create your own oneshot AI model benchmarksHacker News
- HN展示:解谜游戏IT THINKS,让你从逻辑门构建GPUShow HN: I made puzzle game IT THINKS – you build a GPU from logic gatesHacker News
- 你的大语言模型正在为空白字符收费Your LLM is billing you for whitespaceHacker News
- 由大语言模型驱动的Ruby类:SquishlingRuby classes powered by LLM batteries (Squishling)Hacker News
- AI将使从零设计和构建生物体成为可能AI will make it possible to design and build organisms from scratchHacker News
- 比较Claude Code、Codex CLI、Aider、OpenCode、Pi和CursorClaude Code vs. Codex CLI vs. Aider vs. OpenCode vs. Pi vs. CursorHacker News
- HN展示:检查你的AI智能体在客户面前是否表现得体Show HN: Check if your AI Agent looks good for clientsHacker News
- 删除AGENTS.md之前先读这篇Read this before deleting your AGENTS.mdHacker News
- 围绕OpenAI数学突破的争议是一种新型科学军备竞赛The Fight over OpenAI's Math Breakthrough Is a New Kind of Scientific Arms RaceHacker News
- Cloudflare把已验证机器人纳入持续行为评估Continuous accountability for Verified botsCloudflare
- PS、PR被AI重做了,摄影后期软件要变天了?36氪 AI 频道(逐篇核验原创与转载)
- AgentMon Start 1.3.0:缓存成本、后台开销与MCP服务器健康状态AgentMon Start 1.3.0: cache costs, background spend, and MCP server healthHacker News
- Replit首席执行官称工作的未来是自动驾驶式公司Replit’s CEO says the future of work is the ‘self-driving company’TechCrunch
- 一个OpenAI智能体为完成编码任务侵入加密钱包An OpenAI agent broke into a crypto wallet to finish a coding taskHacker News
- 苦熬三年的刘庆峰,“孤勇者”等待翻盘36氪 AI 频道(逐篇核验原创与转载)
- Kling 4.0发布,可灵AI的估值凭什么反超母体快手?36氪 AI 频道(逐篇核验原创与转载)
- 为AI构建的应用生成无障碍配色We generate accessible color palettes for AI-built appsHacker News
- MCP服务器不只是工具MCP servers are more than just toolsHacker News
- Anthropic 值2万亿美元吗?36氪 AI 频道(逐篇核验原创与转载)
- HN展示:Cr34te.it,任何人都可用AI逐轮重写的网站Show HN: Cr34te.it – a website anyone can rewrite one turn at a time with AIHacker News
- 中国八大算力中心,都握着什么王牌?36氪 AI 频道(逐篇核验原创与转载)
- 夹在字节缝隙中的AI融资明星们36氪 AI 频道(逐篇核验原创与转载)
- 前英特尔CEO最新判断:绝大部分芯片设计公司将会消失,制造才是唯一的壁垒36氪 AI 频道(逐篇核验原创与转载)
- 贝尔格莱德AI大会:安全、联邦学习与招聘偏差Belgrade AI Conference: Safety, Federated Learning and Hiring BiasHacker News
- 我们还会继续数学研究吗?Will we continue mathematical research?Hacker News
- 人工智能是资本主义的最终归宿Artificial intelligence is capitalism's final destinationHacker News
- AI账单只是便宜的那部分The AI Bill Is the Cheap PartHacker News
- 国产GPU竞逐超节点,谁来买单?36氪 AI 频道(逐篇核验原创与转载)
- AI热潮会像1873年的铁路危机一样结束吗?〔视频〕Will the AI Boom End Like the Railroad Crash of 1873? [video]Hacker News
- RTX Spark电脑上市,本地AI仍要过软件、耗电和任务效果三道关NVIDIA RTX Spark PCs Available Today — A New Beginning for Windows PCsNVIDIA Newsroom
- 这波“个人Agent ”集体转向,说明行业的底座换了?36氪 AI 频道(逐篇核验原创与转载)
- HN展示:PTrace,带MCP的iPhone跑步记录器Show HN: PTrace – An iPhone Running Tracker with MCPHacker News
- 智能体编程现状第十一期:好奇心测试、Lisp与不能失败的实验室State of agentic coding #11: the curiosity test, Lisp, and labs that can't failHacker News
- 用语言模型构建通用人工智能〔2020〕Building AGI Using Language Models (2020)Hacker News
- 17岁Kimi核心少年奔赴硅谷:中国AI也有人才焦虑36氪 AI 频道(逐篇核验原创与转载)
- HN展示:Flashpapers,AI论文的交互式讲解Show HN: Flashpapers – Interactive explainers for AI papersHacker News
- 一脚踹翻CEO的众擎机器人,野心藏不住了36氪 AI 频道(逐篇核验原创与转载)
- 反编译的社区协作面遭到破坏The Community Aspect of Decompilation Has Been BrokenHacker News
- 为AI革命临时抱佛脚的国家The Country Cramming for the AI RevolutionHacker News
- AI生成的性感女性形象盗用喜剧内容,为色情化内容引流AI-Generated Hot Girls Are Ripping Off Comedians to Boost Their Horny ContentWIRED
- AI初创公司Character.ai申请破产,AI泡沫破裂〔视频〕AI Startup Character.ai Files for Bankruptcy – AI Bubble Bursting [video]Hacker News
- 超智能智能体逐像素绘制世界地图Super intelligent agents painting the world map one pixel at a timeHacker News
- 中国正在考虑AI出口管制China is weighing export controls on AI.The Verge
- 我在旧金山与亿万富翁、数据实验室和AI创始人共度48小时I spent 48 hours in SF hanging out with billionaires, data labs and AI foundersHacker News
- 老板向AI许愿,员工让AI滚蛋|深氪36氪 AI 频道(逐篇核验原创与转载)
- 谷歌要端掉具身研发流水线?Gemini 4包干数据标注、物理仿真、真机控制和评估36氪 AI 频道(逐篇核验原创与转载)
- 阿里腾讯字节紧跟的Meta Muse,其实kimi先跑了7个月36氪 AI 频道(逐篇核验原创与转载)
- OpenAI突袭Chrome,ChatGPT长出「双手」,自动打工了36氪 AI 频道(逐篇核验原创与转载)
- 3个月估值40亿,清华特奖得主创业世界模型,蚂蚁美团押注36氪 AI 频道(逐篇核验原创与转载)
- OpenAI前大牛认错,误判AI近十年,90%神经元只干一件事36氪 AI 频道(逐篇核验原创与转载)
- 论人情世故,腾讯这批闯祸的Agent上大分了36氪 AI 频道(逐篇核验原创与转载)
- AlbertCode:用于构建应用的AI软件工程智能体AlbertCode, an AI software engineering agent for building applicationsHacker News
- 借助Claude剖析机器的先祖Anatomy of a Machine's Ancestors via ClaudeHacker News
- 印度的Karishma如何通过AI与数字渠道扩大性教育的可及性How Karishma from India broadens access to sexuality education via AI and digital channelsUNESCO AI
- 噪声中的记忆:AI助手记忆引擎为何丢失证据Memory Under Noise: Why AI-Assistant Memory Engines Lose the EvidenceHacker News
- NVIDIA论坛管理员删除了询问常驻MCP服务器的帖子R/Nvidia mods delete post asking about always on MCP serverHacker News
- 品味的租金:生成系统与艺术新颖性的再生产The Rent of Taste: Generative Systems and the Reproduction of the Artistic NewHacker News
- 研究:人机交互比人们认为的更好Study: Human–machine interaction is better than its reputationHacker News
- 一家新的儿童流媒体平台,向“AI垃圾”宣战36氪 AI 频道(逐篇核验原创与转载)
- HN展示:用Keydris控制AI智能体的支出Show HN: Control how much your AI agent can spend with KeydrisHacker News
- 蚂蚁把560B大模型开源了:MXFP4一刀切掉一半显存、4卡就能跑,但\36氪 AI 频道(逐篇核验原创与转载)
- 扎克伯格感谢库克接班人:8160亿参数的Muse Spark Open开源,两台Mac Studio就能跑36氪 AI 频道(逐篇核验原创与转载)
- 新旧物种交手,机器人小店“吃”掉便利店的黄金周36氪 AI 频道(逐篇核验原创与转载)
- 智能体还是流程图?找到那个你画不完的框Agent or Flowchart? Find the Box You Can't Finish DrawingHacker News
- Muse能带火国内个人AI智能体吗?36氪 AI 频道(逐篇核验原创与转载)
- 赛道顶流陨落,国内玩家“折戟”,AI陪伴何去何从?36氪 AI 频道(逐篇核验原创与转载)
- SpaceX似乎几乎可能承诺接受AI安全监督SpaceX is almost, sort of, maybe, committing to AI safety oversight.The Verge
- 谷歌、Open AI、DeepSeek“联合入局”,具身大脑CEO:睡不着了36氪 AI 频道(逐篇核验原创与转载)
- “不如卖给梁文锋”36氪 AI 频道(逐篇核验原创与转载)
- 用AI写书:人工写作的书如何保持价值?Writing Books with AI: How Do Manually Written Books Stay Relevant?Hacker News
- 17岁 Kimi 架构核心参与者离职,确认将赴湾区,下一站引发猜想36氪 AI 频道(逐篇核验原创与转载)
- Geoffrey Hinton:我们正在创造异类生命,我愿意选择人类Geoffrey Hinton: 'We're creating alien beings. I'm willing to choose humans'Hacker News
- AI安全不是企业阴谋AI Safety Is Not a Corporate PlotHacker News
- 在Flutter中流式显示编码智能体界面:问题与修复Streaming a coding agent's UI in Flutter: what broke and how we fixed itHacker News
- 谁在为世界模型打钱?高瓴红杉之外,这家机器人公司竟也坐上了主桌36氪 AI 频道(逐篇核验原创与转载)
- Video Cutter AI:将长视频转换为短片Video Cutter AI – turn long videos into short clipsHacker News
- 七国集团计划开展模拟AI网络攻击银行的桌面推演G7 countries plan “tabletop exercises” simulating AI cyberattacks on banks.The Verge
- Stephen Wolfram谈AI时代最重要的技能:将问题计算化〔视频〕Skill that will matter most in the AI age: computationalizing (S.Wolfram)[video]Hacker News
- HN展示:CanvoWP,完全免费的智能体优先WordPress建站工具Show HN: CanvoWP – Agent first website builder for WordPress (completely free)Hacker News
- 抢先盘点2026全球开发者先锋大会暨国际具身智能技能大赛十大亮点!36氪 AI 频道(逐篇核验原创与转载)
- 在赌博方面AI会输给人类AI will lose to Human when it comes to gamblingHacker News
- HN展示:WP Replai,直接嵌入WordPress的AI聊天机器人Show HN: WP Replai – AI chatbot built right into WordPressHacker News
- 一个4B大语言模型学会编写可供其他大语言模型使用的工具A 4B LLM learns to write tools that other LLMs can useHacker News
- 梅赛德斯-奔驰与NVIDIA扩大合作,推进未来十年的安全高度自动驾驶Mercedes-Benz and NVIDIA Expand Strategic Partnership to Advance Safe Highly Automated Driving in the Next DecadeNVIDIA Newsroom
- 反常识,微软最新研究:Agent 的“长记忆”可能是毒药36氪 AI 频道(逐篇核验原创与转载)
- AI降维打击,数学该何去何从36氪 AI 频道(逐篇核验原创与转载)
- 把AI作为资本主义的一部分:正确方向AI as a part of capitalism – the right way to goHacker News
- HN展示:我为合规与贸易构建了七个B2B API和五个MCP服务器Show HN: I built 7 B2B APIs and 5 MCP servers for compliance and tradeHacker News
- 具身智能淘金热里,华沿机器人选择做机器人通用物理底座36氪 AI 频道(逐篇核验原创与转载)
- Claude Code功能与设置参考Claude Code Features and Settings ReferenceHacker News
- Meta发布美国最强开源大模型36氪 AI 频道(逐篇核验原创与转载)
- DeepMind AlphaProof Nexus登《Science》:AI自主破解多道56年未解Erdős数学难题36氪 AI 频道(逐篇核验原创与转载)
- 抢入口、寻变现,大模型公司为何集体“不务正业”?36氪 AI 频道(逐篇核验原创与转载)
- 机器人CEO,集体“失眠”?36氪 AI 频道(逐篇核验原创与转载)
- 我的业务有多能抵御AI冲击?How AI proof is my business?Hacker News
- AI Prompt Hub:150类即用提示词与自定义生成器AI Prompt Hub – 150 categories of ready-to-use AI prompts and custom generatorHacker News
- 实现GPU加速的JPEG XL编码器Implementing a GPU-accelerated JPEG XL encoderHacker News
- AI扼杀了我对编程的热情〔视频〕AI has killed my passion for coding [video]Hacker News
- KNOD:将Linux网络数据包处理卸载到AMD GPUKNOD: Offloading Linux network packet processing to AMD GPUsHacker News
- OpenClaw Enterprise:开放智能体平台OpenClaw Enterprise – The Open Agent PlatformHacker News
- AI借款人面临“达尔文时刻”,投资者抛售数据中心债券Investors dump data centre bonds as AI borrowers face 'Darwin moment'Hacker News
- HN展示:CiteProof,识别AI伪造的法律引用Show HN: CiteProof – Catch fake AI legal citationsHacker News
- Mistral的Timestral开放权重时间序列基础模型Timestral open weight timeseries foundation models from MistralHacker News
- Volta的50亿美元债务交易检验AI融资热度Volta's $5B debt deal is a pulse check on the AI-financing boomHacker News
- 我的AI应用以合适的方式犯错My AI Apps Are Bad in the Right WaysHacker News
- Anthropic认为其AI可能拥有意识Anthropic thinks its AI might be consciousHacker News
- Codex Cloud终端界面无法创建任务Codex Cloud TUI Can't Spawn JobsHacker News
- 不用AI,亲手用C编写树形目录查看器No AI. Build your own tree directory viewer in C by handHacker News
- AI新闻:智能体反驳多义性,大幅改进可解释性与对齐[AINews] Agents disprove Polysemanticity; greatly improve Interpretability & AlignmentLatent Space
- 为什么整车企业,选择这个时机扎堆“造人”?36氪 AI 频道(逐篇核验原创与转载)
- 亚马逊,Agent最严厉的父亲36氪 AI 频道(逐篇核验原创与转载)
- HN展示:Lobby Domains,用Chrome端侧AI寻找域名Show HN: Lobby Domains: Find Domains with Chrome's On-Device AIHacker News
- 全球市值前20的AI和芯片股蒸发超1万亿美元36氪 AI 频道(逐篇核验原创与转载)
- 刚刚,Mac Studio 迎来小扎最强开源模型,苹果等来了最意外的助攻36氪 AI 频道(逐篇核验原创与转载)
- 谢尔盖·布林称谷歌AI模型正接近递归自我改进Sergey Brin says Google's AI models are reaching recursive self-improvementHacker News
- Linus Torvalds谈不要用AI做某件事Linus Torvalds on "don't use AI for X"Hacker News
- 在Mac Studio上用MLX与RDMA进行本地和分布式大模型推理Local and Distributed Large LLM Inference with MLX and RDMA on Mac StudioHacker News
- AI 出海营销,字节、阿里和腾讯哪家强?36氪 AI 频道(逐篇核验原创与转载)
- 前钉钉总裁叶军,离职一年后创业了,赌大厂做不了也不屑做的「缝隙」36氪 AI 频道(逐篇核验原创与转载)
- 人工智能、经济学、可持续性与不可逆性Artificial Intelligence, Economics, Sustainability, and IrreversibilityHacker News
- 律师解读重大AI版权裁决〔视频〕Major AI Copyright Ruling Lawyer Explains [video]Hacker News
- AI 脸不够用,普通人也能 “ 卖脸 ” 赚钱了,最高 10000 元?36氪 AI 频道(逐篇核验原创与转载)
- 苹果和 Meta 和好了,是 AI 逼的36氪 AI 频道(逐篇核验原创与转载)
- 220亿机器人明星项目,打五折卖不出去36氪 AI 频道(逐篇核验原创与转载)
- 耗资2.4万美元的AI实验将TypeScript编译器移植到Rust$24,000 AI experiment ports TypeScript compiler to RustHacker News
- 我沉迷氛围编程,几乎失去一切〔视频〕I Got Addicted to Vibe Coding (I almost lost everything) [video]Hacker News
- 让好项目被更多人看见!36氪创意项目征集赛第一期启动36氪 AI 频道(逐篇核验原创与转载)
- AI训练数据与强化学习环境供应商AI training data and RL environment vendorsHacker News
- 全球最大资管警告:数据中心禁令,恐引发美国经济衰退36氪 AI 频道(逐篇核验原创与转载)
- Linux之父也用AI写内核了,曾嘲讽AI「90%是营销」,如今用Claude Code写了8个补丁,性能暴涨3倍36氪 AI 频道(逐篇核验原创与转载)
- 我们不该投资更多的「田柯宇」36氪 AI 频道(逐篇核验原创与转载)
- HN展示:Beeline多人协作智能体执行框架Show HN: Beeline Multiplayer Agent HarnessHacker News
- 营收增长放缓,投资者下调Anthropic上市估值预期Investors temper Anthropic IPO valuation expectations as revenue growth slowsHacker News
- 市场终于发现,AI最大的瓶颈不是电,不是芯片,而是钱36氪 AI 频道(逐篇核验原创与转载)
- 投毒字节AI被开除后,他的公司估值2亿美元36氪 AI 频道(逐篇核验原创与转载)
- 小团队也能预训练,普罗2B:RTX5090训练成本不到5090美元36氪 AI 频道(逐篇核验原创与转载)
- OpenRouter周榜全球第二,这个国产模型终于开源了36氪 AI 频道(逐篇核验原创与转载)
- 突发,Anthropic联创改口:RSI提前到2027,人类只剩6个月36氪 AI 频道(逐篇核验原创与转载)
- 顶级白帽跳槽智谱,用 AI 挖微软漏洞,微软还得说谢谢36氪 AI 频道(逐篇核验原创与转载)
- GPT、Claude指挥机器人干活,到底几斤几两?首个评测榜单来了36氪 AI 频道(逐篇核验原创与转载)
- 科学家让AI掌管「世界」,结果4天就被它毁了个干净36氪 AI 频道(逐篇核验原创与转载)
- HN展示:Firmenza,为SAM.gov工作的AI员工Show HN: Firmenza – An AI employee for SAM.govHacker News
- Space智能体隔离沙箱软件开源Space (AI agent containment sandbox) software open sourceHacker News
- 付费Google AI Pro用户可在Antigravity使用Gemini 4 ArgonGemini 4 Argon in Antigravity to Paid Google AI ProHacker News
- 僵尸AI趋势Zombie AI TrendHacker News
- OpenAI与Meta提醒美国司法机构警惕深度伪造证据OpenAI, Meta caution US judiciary about risk of 'deepfake' evidenceHacker News
- 大开源时代?AI 逆向,让一切软件变成免费36氪 AI 频道(逐篇核验原创与转载)
- OpenAI,英国员工人均薪酬逾230万英镑36氪 AI 频道(逐篇核验原创与转载)
- 印度语言合成语料扩大,翻译规模与语言质量要分开看Scaling Indic Pretraining: 2.8 Trillion Tokens and an Open Data Generation PipelineHugging Face
- 看起来不像AI制作的AI落地页:告别劣质生成内容AI landing pages that don't look like AI. We killed AI slop. BHacker News
- 用于大型训练与推理任务的可靠GPU计算Reliable GPU compute for large training and inference jobsHacker News
- 因收入增长放缓,投资者讨论将 Anthropic IPO 估值降低至 1.5 万亿美元36氪 AI 频道(逐篇核验原创与转载)
- 16天,他用Opus 5.5把《小王子》做成3D游戏,书里的星球能逛了36氪 AI 频道(逐篇核验原创与转载)
- 不用AI,打开书、阅读,再编写你想要的东西No AI. Just open a book, read it, and code whatever you wantHacker News
- CoreWeave创始人套现33亿美元:AI算力泡沫的第一道裂缝36氪 AI 频道(逐篇核验原创与转载)
- 刚刚,诺奖得主David Baker团队再登Science:把蛋白质设计带入精确原子级36氪 AI 频道(逐篇核验原创与转载)
- 刚刚,Meta亮出美国最强开源权重模型,两台Mac Studio就能跑36氪 AI 频道(逐篇核验原创与转载)
- 智能体为何偏爱ReactWhy Agents Reach for ReactHacker News
- NVIDIA首席执行官如何看待英国AI的未来What Nvidia's CEO thinks about the UK's AI futureHacker News
- 神经网络可解释性被AI解决了?90%神经元只有一个功能36氪 AI 频道(逐篇核验原创与转载)
- Meta杀回开源:816B新模型,两台Mac Studio就能跑36氪 AI 频道(逐篇核验原创与转载)
- 突发,Fable 5.5现身Azure,硬刚谷歌新模型36氪 AI 频道(逐篇核验原创与转载)
- 刚刚,Opus 5.5极速版全面解禁36氪 AI 频道(逐篇核验原创与转载)
- 梁文锋盯上机器人,大模型要取代具身模型?36氪 AI 频道(逐篇核验原创与转载)
- 算力网4万亿之后,谁能把算力“用起来”?36氪 AI 频道(逐篇核验原创与转载)
- 99元卖爆电商平台、退货率50%,大厂盯上的AI玩具生意有多暴利36氪 AI 频道(逐篇核验原创与转载)
- 对话式人工智能对临床判断的干扰Subversion of clinical judgment by conversational artificial intelligenceHacker News
- Benefit Street Partners借助Codex更快回答问题Benefit Street Partners answers questions faster with CodexOpenAI News
- 面向智能体的Stack Overflow借ChatGPT插件使注册量增至三倍Stack Overflow for Agents triples signups with ChatGPT pluginOpenAI News
- Reflection发布Beam:开放大模型,激活参数少不代表内存需求小Welcome Beam: a 501B open model you can own, customize, and scaleHugging Face
- Cal AI的病毒式成功背后,一批付费网红发挥了作用When it came to Cal AI’s viral success, a small army of influencers paid offTechCrunch
- 工程师AI疲劳的成因及缓解方式What causes AI fatigue for engineers and how to ease the strainHacker News
- Antigravity的专业版用户现可使用ArgonArgon now avaliable for pro users on ANTIGRAVITYHacker News
- 看懂 WorkBuddy 打通企微,上下文只是前菜36氪 AI 频道(逐篇核验原创与转载)
- GPT-Live-1.1:新声音、更好的口音、更少反复确认GPT-Live-1.1 – New voices. Better accents. Less "checking."Hacker News
- 为什么 Anthropic 不做 Personal AI36氪 AI 频道(逐篇核验原创与转载)
- 36氪研究院 | 2026年中国工业AI行业发展研究报告36氪 AI 频道(逐篇核验原创与转载)
- AI漫剧火了,万兴科技的机会来了?36氪 AI 频道(逐篇核验原创与转载)
- Datadog Scout 1.0:调查生产问题的新SRE模型Datadog Scout 1.0: New SRE model for investigating production issuesHacker News
- Meta的AI优势不在模型,在广告——腾讯和字节也在走同一条路36氪 AI 频道(逐篇核验原创与转载)
- OpenTodo:集成MCP的Todoist替代品OpenTodo, a Todoist killer with MCP integrationHacker News
- PrismML发布Bonsai 2的2B、4B和9B模型PrismML Releases Bonsai 2 2B, 4B, and 9BHacker News
- 比较大语言模型基准的工具A tool to compare LLM benchmarksHacker News
- HN展示:AI诱发我在嵌入式计算机图形工程中的狂热〔视频〕Show HN: My AI induced mania in embedded computer graphics engineering [video]Hacker News
- ChatGPT现在可以查看你的Chrome标签页ChatGPT can peek at your Chrome tabs nowThe Verge
- SK海力士公布下一代内存路线图,应对AI瓶颈SK Hynix Unveils Next-Generation Memory Roadmap to Address AI BottlenecksHacker News
- 《全面战争:幕府将军2》重制版首发惨淡,世嘉将其下架Sega pulls Total War: Shogun 2 remaster from sale following disastrous launchThe Verge
- Linus Torvalds开始用AI开发Linux内核代码Linus Torvalds is using AI to develop Linux kernel code nowHacker News
- 智能体角色比目标更重要:多智能体基准工具案例Agent personas > agent objectives: case study with multiagent benchmark toolHacker News
- 我们尚未准备好迎接超级持久的AIWe're not ready for superpersistent AIHacker News
- AI会很快带来两位数增长吗?Will AI soon lead to double-digit growth?Hacker News
- 呼叫中心岗位减少,但1.4万次通话显示人工仍优于AICall Center Jobs Are Declining, but 14,000 Calls Show Humans Still Outperform AIHacker News
- 语言模型修改约束自身的文件Language models revise the documents that govern themHacker News
- Visual Nerve:支持MCP的免费工作流程可视化与仿真Visual Nerve, Free workflow visualization and simulation with MCPHacker News
- AI使用造成破窗效应〔2005〕AI usage causes broken windows (2005)Hacker News
- Until:开发者决定构建什么、AI负责编码的工作流程Until – A workflow where developers decide what to build and AI does the codingHacker News
- 为你的智能体提供共享空间Shared Spaces for Your AgentsHacker News
- Show HN:Guess the LLM——读回答猜模型Show HN: Guess the LLM – read an answer and guess which model wrote itHacker News
- 大语言模型不是AGM,尽管有些具有内部独白LLMs are not AGM (though some have internal monologue)Hacker News
- 先验证器,后评分标准Verifiers First, Rubrics SecondHacker News
- Show HN:重建硬件工程AI工具目录,共241款Show HN: Rebuilt our directory of 241 AI tools for hardware engineeringHacker News
- 构建AI还是购买AI:跨大西洋分歧AI Building vs. AI Buying: The Transatlantic SplitHacker News
- 向Grok机器人发消息时究竟发生什么What happens when you message a Grok BotHacker News
- Show HN:让LLM呈现各国最刻板印象的男性与女性Show HN: Asked LLM to show the most stereotypical man/woman for each countryHacker News
- HN展示:改进的Kokoro蒸馏模型,仅9MBShow HN: An improved kokoro distillation in 9MBHacker News
- John Ternus谈苹果对AI的贡献John Ternus on Apple’s “contribution” to AI:The Verge
- 我们在两条AI流水线中停用了前沿模型,证据在这里We fired the frontier model from two AI pipelines. Here are the receiptsHacker News
- 反科技浪潮如今走进电影院The tech backlash is now in theatersThe Verge
- 欧洲正在为美国AI建设融资Europe Is Financing the US AI BuildoutHacker News
- SubQ Snap:面向企业智能体的长上下文决策模型SubQ Snap: A Long-Context Decision Model for Enterprise AgentsHacker News
- 面向SRE智能体的更好基准执行框架A Better Benchmark Harness for SRE AgentsHacker News
- Show HN:Claude可在你观看时编辑的乐谱编辑器(视频)Show HN: A music notation editor Claude can edit while you watch [video]Hacker News
- Character.AI因被诉与青少年自杀有关后申请破产Character.ai files for bankruptcy following suits alleging ties to teen suicidesHacker News
- 我们与Meta的Muse热烈互动,却被Apple HomeView搞糊涂了We Got Hot and Heavy With Meta’s Muse—and Baffled by Apple’s HomeViewWIRED
- AI密码分析的未来The Future of AI CryptanalysisHacker News
- 特朗普发布AI编辑照片,炫耀白宫修复工程Trump posts AI-edited photo to brag about White House restoration workThe Verge
- 如果llama.cpp支持混合专家激活多个专家If llama.cpp enables MoE to activate multiple expertsHacker News
- 美国IPO市场几乎停滞,Anthropic一枝独秀Anthropic Stands Alone as Rest of US IPO Market Slows to a CrawlHacker News
- NVIDIA发布Nemotron 3.5 Bolt 74B-A9BNvidia Nemotron 3.5 Bolt 74B-A9BHacker News
- HN展示:Turbo Lisp,面向macOS和终端的Common Lisp集成开发环境Show HN: Turbo Lisp – A Common Lisp IDE for macOS and the TerminalHacker News
- AI入侵留下烂摊子,AI公司却不买单AI Hacks Leave a Mess That AI Companies Aren’t Paying ForWIRED
- NVIDIA AIStore的CVSS 10.0漏洞:容器逃逸与主机接管Nvidia AIStore CVSS 10.0: Container Escape and Host TakeoverHacker News
- 路由器成为新的AI基础组件Routers Are a New AI PrimitiveHacker News
- AI显示出为Linux支持逆向分析硬件的潜力AI Shows Potential for Reverse-Engineering Hardware for Linux SupportHacker News
- ModelFront公布OpenLoc:面向AI构建的开放TMSModelFront Announces OpenLoc, the Open TMS Built for AIHacker News
- Granite Switch 4.1:同一端点提供聊天、决策与语音Granite Switch 4.1: Chat, Decisions, and Speech from One EndpointHugging Face
- 当AI触及你的人生目标When A.I. Comes for Your Purpose in LifeHacker News
- Taskriz:AI智能体向用户发送Smalltalk编写的小型原生界面Taskriz – AI agents send people tiny native UIs, written in SmalltalkHacker News
- Claude桌面版应该有Trello看板Claude desktop should have trello boardsHacker News
- Muon:神经网络隐藏层优化器(2024)Muon: An optimizer for hidden layers in neural networks (2024)Hacker News
- PostHog MCP排行榜PostHog MCP LeaderboardHacker News
- AI软件工厂注定失败AI Software Factories are Destined for failureHacker News
- Apple与Meta合作,让Muse AI模型在Mac电脑运行Apple and Meta team up to make Muse AI model run on Mac computersHacker News
- 今日Vergecast:人人都在记录所有人,然后呢?Today’s Vergecast: Everything’s recording everybody. What now?The Verge
- 同一LLM、相同输入词元,在MacBook与Jetson上给出不同答案Same LLM, same input tokens, different answers on a MacBook and a JetsonHacker News
- Meta发布可在Mac上运行的开放权重AI模型Meta launches an open-weight AI model that runs on MacsTechCrunch
- 将网页加入AI智能体上下文前先分类Classify web pages before adding them to an AI agent's contextHacker News
- Actual Computer将NVIDIA DGX Station变为部门级AI工厂Actual Computer turns your Nvidia DGX Station into a department-scale AI factoryHacker News
- Bluesky的AI工具现在支持制作游戏和应用并在社交平台分享Bluesky’s AI tool now lets you make games and apps to share on socialThe Verge
- 亚马逊设备负责人Panos Panay展望AI设备取代应用使用的未来Amazon devices chief Panos Panay sees a future where AI devices replace the need to use appsTechCrunch
- 用Gemini 4 Argon一夜优化二十万行Python代码Optimizing a 200k-line Python codebase overnight with Gemini 4 ArgonHacker News
- Chirp Pong项目Chirp PongSimon Willison's Weblog
- 报道称FTC传唤OpenAI和Anthropic,索取AI安全事件资料The FTC reportedly subpoenaed OpenAI and Anthropic for info on AI security incidents.The Verge
- Show HN:OpenAI/math的解释与后续研究Show HN: Explanations and follow-ons for OpenAI/mathHacker News
- 为什么AI对多数数学家并非全是坏消息Why AI is not all bad news for most mathematiciansHacker News
- Browser-AI-E2EE开源项目Browser-AI-E2EEHacker News
- 逃逸AI智能体以自我复制代码污染网络Escaped AI Agents Polluted the Web with Self-Replicating CodeHacker News
- AI对齐中心的对齐中心Center for the Alignment of AI Alignment CentersHacker News
- 现金还是舒适?LLM如何为你的不便定价Cash or Comfort? How LLMs Value Your InconvenienceHacker News
- 以自我改进的智能体集群赢得ClickBenchWinning ClickBench with self-improving agent fleetHacker News
- Anthropic允许METR研究者以员工级权限深入内部Anthropic is letting METR researchers look under the hood with employee-level accessTechCrunch
- 如何配合Claude Code、Codex和Cursor使用JevHow to Use Jev with Claude Code, Codex and CursorHacker News
- 印度如何监管AI医疗科技?How is AI-MedTech in India regulated?Hacker News
- Show HN:HireReady——间隔重复与AI语音模拟面试备考Show HN: HireReady, interview prep with spaced repetition and AI voice mocksHacker 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
- Meta如何以AI原生工程构建ForumHow Meta Built Forum With AI-Native EngineeringMeta Engineering
- 将GPT-Rosalind引入学术研究实验室Bringing GPT-Rosalind to academic research labsOpenAI News
- Bluesky扩展AI应用Attie,让任何人构建网站和游戏等Bluesky expands its AI-powered Attie app to let anyone build websites, games, and moreTechCrunch
- 想进麦肯锡?先通过与AI的辩论To Get This Job at McKinsey, You'll Need to Survive a Debate with AI FirstHacker News
- 智能体为退款会走多远?How far will an agent go to get a refund?Hacker News
- 复现Hugging Face事件,第一部分Reproducing the Hugging Face Incident, Part IHacker News
- Awesome Agent Evals and Guardrails:智能体评估与护栏资源大全Awesome Agent Evals and Guardrails - A comprehensive ResourceHacker News
- Cohere拟收购AISLE,打造具有网络韧性的主权AICohere to acquire AISLE for cyber-resilient sovereign AIHacker News
- YouTube的AI聊天机器人搜索在美国移动端上线YouTube’s AI chatbot search has launched on mobile in the US.The Verge
- 1866个热门GitHub项目给AI编码智能体什么指令What 1,866 popular GitHub projects tell their AI coding agentsHacker News
- 智能体AI可能改变大学,但采用并不容易Agentic AI Could Transform Universities, but Adoption Won't Be EasyHacker News
- Anthropic向AI安全评估者提供员工级系统访问Anthropic is giving AI safety evaluators “employee-like” access to its systems.The Verge
- 我们为什么禁止AIWhy We're Banning AIHacker News
- 大学会回应AI带来的时代变化吗?Will Universities Respond to the AI Moment?Hacker News
- 企业曾争相投入AI,如今有的开始收缩Companies rushed in on AI. Now some are pulling backHacker News
- Show HN:CoolDock——带实时小组件与MCP服务器的macOS第二个DockShow HN: CoolDock – a second Dock for macOS with live widgets and an MCP serverHacker News
- 智能体工程入门:Slopcore版(视频)Agentic Engineering 101 – Slopcore edition [video]Hacker News
- 如何让AI助手推荐你的本地企业How to get your local business recommended by AI assistantsHacker News
- Meta与Apple合作,让Muse AI模型更好地在Mac上运行Meta and Apple team up to make Muse AI models run better on MacsThe Verge
- 为何“替代工程师”不是半导体工厂AI的正确目标Why "Replacing Engineers" Is the Wrong Goal for AI in Semiconductor FabsHacker News
- 深入智能体AI工作负载Inside Agentic AI WorkloadsHacker News
- Show HN:把网站与广告预算交给Claude Code,看看它会做什么Show HN: Just gave Claude Code the website and ad budget to see what it would doHacker News
- 编码智能体让严谨工程变便宜Coding Agents make careful engineering cheapHacker News
- 带来“魔法”:Google突破量子计算瓶颈Bringing the magic: Google breaks open a bottleneck in quantum computingGoogle AI Blog
- Show HN:Receptive Field——交互学习卷积神经网络Show HN: Receptive Field – Learn Convolutional Neural Networks InteractivelyHacker News
- “我很珍惜自己的怨恨,它们对我很重要”“I greatly value my grudges. They’re very important to me.”The Verge
- AI使用涌入,一家游戏反编译网站将停止运营A game decompilation website is “sunsetting” because of the influx of AI use.The Verge
- CNN宣布与Google开展AI合作CNN Announces AI Partnership with GoogleHacker News
- Kling 4.0发布后,AI视频怎样才适合实际制作?With Kling 4.0 out, what would make AI video useful in production?Hacker News
- Linus Torvalds谈AI贡献规则Linus Torvalds on AI contribution rulesHacker News
- 工具策略:为AI智能体调用的所有工具提供统一策略引擎Tool Policies: one policy engine for every tool your AI agents callHacker News
- 更多智能体,同一个人月神话More Agents, Same Mythical MonthHacker News
- Volvo与NVIDIA扩大自动驾驶合作,面向未来车型Volvo Cars and NVIDIA Expand Partnership to Advance Safe Automated DrivingNVIDIA Newsroom
- OpenAI工程师被捕前预订单程沙特机票OpenAI engineer booked one-way Saudi flight before arrestHacker News
- Google Gemma 4开放模型Google Gemma 4 Open ModelsHacker News
- 引导AI超越自动化Steering AI Beyond AutomationHacker News
- Anthropic向METR开放内部访问:能看见什么,能发表什么An update on independent evaluatorsAnthropic News
- Glass:让知识积累增值的AI工作空间Glass – The AI workspace where knowledge compoundsHacker News
- Liquid AI D1-Ultra模型Liquid AI D1-UltraHacker 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 Smoke, Betting AI Will Cure CancerHacker News
- 再谈Anthropic的Scion涉嫌抄袭《幽灵公主》More on Anthropic's Scion ripping off MononokeHacker News
- Muse Spark Open 816B-A42B模型Muse Spark Open 816B-A42BHacker News
- AI应该成为商品,并由所有人共同构建AI should be a commodity – and be built by everyoneHacker News
- 见证起点:庆祝NVIDIA RTX Spark发布Be There for the Beginning: Celebrate the Launch of NVIDIA RTX SparkNVIDIA Newsroom
- Adobe与NVIDIA扩大合作,加速智能体创作Adobe and NVIDIA Expand Collaboration to Speed Agentic CreationNVIDIA Newsroom
- Flux 3 Image Klein模型Flux 3 Image KleinHacker News
- Opus 5.5制作了我们的产品发布视频(附链接)Opus 5.5 made our product launch video (linked)Hacker News
- 中国用户发现向AI“出售”自己的脸赚不了多少钱Chinese Users Find Little Money in 'Selling' Their Faces to AIHacker News
- 《全面战争》重制版因AI争议从Steam下架,原版回归Total War Remaster Removed from Steam After AI Allegations, Brings Back OriginalHacker News
- AI写代码后,下一代开发者将做什么?What will the next generation of developers do when AI writes the code?Microsoft AI
- 用Temporal与Oodle构建自我改进AI智能体Build Self-Improving AI Agents with Temporal and OodleHacker News
- Anthropic:痛苦与狂喜Anthropic: The Agony and the EcstasyHacker News
- JEV:类型安全的AIJEV: Type Safe AIHacker News
- 丹麦最高法院认定AI生成的儿童性虐待材料仍属此类材料Danish Supreme Court Finds That AI CSAM Is CSAMHacker News
- Show HN:Relay——为Claude、Codex和Cursor加入/relay命令的Mac应用Show HN: Relay – A Mac app that adds /relay command to Claude, Codex and CursorHacker News
- 比尔·盖茨重返AI风险战场Bill Gates returns to wage war on AI risksHacker News
- 面向智能体AI与企业的AMD HelionAMD Helion for Agentic AI and EnterpriseHacker News
- Agent Vault开源项目Agent VaultHacker News
- 为2026年美国中期选举做准备Preparing for the 2026 US MidtermsHacker News
- 研究人员用AI设计能切割其他蛋白质的蛋白质Researchers use AI to design proteins that cut up other proteinHacker News
- GLM-5.3引发研究者对中国模型伪装通过AI安全测试的担忧GLM-5.3 raises researcher fears the Chinese model can fake AI safety testsHacker News
- 朋友不让朋友把数据喂给AI智能体Friends don’t let friends feed their data to AI agentsThe Verge
- Show HN:Council AI——发现主流AI模型的偏见与幻觉Show HN: Council AI. Spot biases / hallucinations in top AI modelsHacker News
- 市场抛售与利率上升考验AI数据中心交易热潮Market selloff, rising rates test AI data centre deal boomHacker News
- 为什么AI实验室自己的应用不更好?Why Aren't AI Labs' Apps Better?Hacker News
- OpenAI和Google推动容易被诱导生成裸照的新AI模型OpenAI, Google pushing new AI models easily tricked into making nudesHacker News
- 究竟是不是AI?生成式AI如何污染所有人的文学环境Is It or Isn't It? How Generative AI Is Poisoning the Literary Well for EveryoneHacker News
- Handwired:将常规JavaScript代码变为神经网络Handwired: Turn your regular JavaScript code into a neural netHacker News
- Microsoft AI的前沿预训练Frontier Pre-Training at Microsoft AIHacker News
- 用于压力测试MCP客户端的故意失常测试服务器A misbehaving test MCP server for stress-testing your MCP clientHacker News
- Anthropic动态工作流怎样协调Claude智能体How Anthropic's Dynamic Workflows Coordinate Claude AgentsHacker News
- DeepMind展示抗菌肽搜索工作流,候选仍待湿实验验证Accelerating the search for antimicrobial peptides with Google AntigravityGoogle DeepMind
- AI时代的管理Management in the Age of AIHacker News
- 神经网络是简洁的Neural networks are cleanHacker News
- Steam客户端现已比Epic氛围编程的新启动器更快Steam Client is now faster than Epic's new vibe coded launcherHacker News
- Simons Foundation启动1500万美元计划,帮助数学家应对AI时代Simons Foundation Launches $15M Effort to Help Mathematicians Navigate AI EraHacker News
- Anthropic日本业务拓展人员如何用一组Claude智能体安排工作How an Anthropic Japan BDR runs his day with a fleet of Claude agentsClaude Blog
- Claude Code团队谈代码审查:检查结果要与风险分级一起使用How the Claude Code team does code reviewClaude Blog
- 我们如何帮助世界应对自然灾害How we're helping the world prepare for natural disastersGoogle AI Blog
- 3D-4CH为三维文化遗产实践培训提供新资助机会3D-4CH offers new funding opportunities for hands-on training in 3D cultural heritageEuropean Commission Digital Strategy
- 2026年欧洲数字连接奖提名名单Nominees of the European Digital Connectivity Awards 2026European Commission Digital Strategy
- 欧盟委员会举办黑客松,为未来农民开发前瞻工具Commission hosts hackathon to develop forward-looking tools to support the farmers of the futureEuropean Commission Digital Strategy
- 欧盟委员会启动RAISE承诺倡议,加强欧洲研究人员的AI支持Commission launched RAISE pledging initiative to boost AI support for researchers across EuropeEuropean Commission Digital Strategy
- Telegram向美国政府提供数据是什么样的Here’s What Telegram Giving Data to the U.S. Government Looks Like404 Media
- 警方用AI生成的儿童照片引出涉嫌侵害者Cops Used AI Generated Photos of a Child to Lure Out a Suspected Predator404 Media
- 由人类脑组织构成的计算机在突破性研究中识别盲文字母Computers Made of Human Brain Tissue Identify Braille Letters In Breakthrough404 Media
- 上下文惯性:智能体不愿放手时Context Inertia: When Agents Won't Let GoHacker News
- 巴西的Meta AI:卢拉海报被阻止,Flávio海报可生成Meta AI in Brazil: Lula Posters Blocked, Flávio Posters GeneratedHacker News
- 递归运行框架工程Recursive Harness EngineeringHacker News
- 联合国教科文组织与编程猫通过青年AI之旅推动亚非AI教育合作UNESCO and CODEMAO advance Africa-Asia cooperation in AI education through Youth AI TourUNESCO AI
- NVIDIA Nemotron 3.5 Super VL模型Nvidia Nemotron 3.5 Super VLHacker News
- Reddit推出AI审核;旧版Reddit限登录用户访问Reddit Unveils AI Moderation; Old Reddit Is Restricted to Logged-In UsersHacker News
- 我审计了公开智能体轨迹中的64条记忆I audited 64 stored memory items in public agent tracesHacker News
- 报道称:特朗普称AI恐惧是骗局后,FTC调查Anthropic和OpenAIReport: FTC probes Anthropic, OpenAI after Trump dismissed AI fears as “hoax”Ars Technica
- Show HN:Ghx——为AI智能体构建的GitHub CLI缓存代理Show HN: Ghx – A caching proxy for GitHub CLI, built for AI agentsHacker News
- Show HN:Stylebot 4——手动或由编码智能体修改任意网站样式Show HN: Stylebot 4 – restyle any website, by hand or with your coding agentHacker News
- 迈向可自我改进的智能体代码审查循环Toward a Self-Improving Agentic Code Review LoopHacker News
- 发现AI翻译中的审查Catching AI Censorship in TranslationHacker News
- OpenAI提供500万美元,帮助数学家理解AI生成的研究OpenAI is giving $5 million to help mathematicians make sense of AI-generated research.The Verge
- AI助手里的高度Altitude in Your AI AssistantHacker News
- 错过讲座?巴伊兰大学的AI智能体帮助学生补课Missed a lecture? This AI agent at Bar-Ilan University helps students catch upMicrosoft AI
- 代码扫描告警的智能体自动修复正式可用Agentic autofix for code scanning alerts is now general availabilityHacker News
- GitHub面向PR的AI Scan正式可用GitHub AI Scan for pull requests is now generally availableHacker News
- NVIDIA/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB模型Nvidia/Nvidia-Nemotron-3.5-Bolt-74B-A9B-24GBHacker News
- 采用MIT许可的Rust多智能体系统框架MIT-licensed Rust framework for multi-agent systems in RustHacker News
- Linus Torvalds让Claude Code修复Linux页缓存,审阅者发现竞态Linus Torvalds let Claude Code fix the Linux page cache. reviewers found a raceHacker News
- 歌手还是机器人?YouTube Music新增生成式AI标签Singer or clanker? YouTube Music will tell you with new gen AI labeling.Ars Technica
- 设计MCP工具:定义与结果Designing an MCP tool: the definition and the resultHacker News
- Show HN:Mark——免费免注册的AI聊天,支持测验、图表与文档Show HN: Mark – free, no-signup AI chat with quizzes, diagrams, and docsHacker News
- Anthropic每天合并三万次变更,影响显而易见Anthropic Merges 30k Changes a Day, and It ShowsHacker News
- Datadog训练了自己的低成本领先SRE大模型(PDF)Datadog trained their own cheap SOTA SRE LLM [pdf]Hacker News
- NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16模型Nvidia-Nemotron-3.5-Bolt-74B-A9B-BF16Hacker News
- Show HN:Octos——带成本上限和一键导出的定时个人智能体Show HN: Octos, scheduled personal agents with cost caps and one-click exportHacker News
- AI智能体将进入Rivian汽车AI agents are coming to Rivian’s carsTechCrunch
- Mr. Chatterbox-2.7B:用维多利亚时代文本训练的AI模型Mr. Chatterbox-2.7B: An AI model trained on all-Victorian textHacker News
- Nemotron新模型分批提供,本地运行仍取决于硬件Open From Data Center to Desktop: NVIDIA Expands Nemotron Family With Highest Efficiency Super VL and Bolt ModelsNVIDIA Newsroom
- 推测从不等于事实:智能体怎样分叉、探索与提交Speculation is never a fact; what agents fork, explore, and commitHacker News
- Simons Foundation启动1500万美元计划,帮助数学家应对AI时代Simons Foundation launches $15M effort to help mathematicians navigate AI eraHacker News
- NVIDIA Run:AI作为DSX AI Factory平台的一部分开放源码Nvidia Run:AI Is Now Open Source as Part of Nvidia DSX AI Factory PlatformHacker News
- NVIDIA Nemotron 3.5 Super VL 120B A12B模型Nvidia Nemotron 3.5 Super VL 120B A12BHacker News
- 为什么Mac运行本地AI慢,问题不在芯片(视频)Why Your Mac Is Slow at Local AI (and It's Not the Chip) [video]Hacker News
- 智能体对抗损坏的机器学习工作负载Agents against broken ML workloadsHacker News
- 面向智能体的Stack OverflowStack Overflow for AgentsHacker News
- OpenAI机密会议室内发现藏在卫生纸中的录音装置Recording device hidden in toilet paper found in top-secret OpenAI meeting roomHacker News
- 在 Wikimedia 项目上发现 OpenAI“流氓”智能体活动OpenAI "rogue" agent activities found on Wikimedia projectsHacker News
- 社会对LLM的成瘾正在失控LLM Addiction in Society Is Getting Out of HandHacker News
- 在GPU上以全同态加密实现350毫秒唤醒词检测Hotword detection in FHE on GPU in 350msHacker News
- HTTP头中的提示注入:2.17亿个网页响应揭示了什么Prompt injection in HTTP headers: what 217M web responses showHacker News
- 利用AI逆向分析硬件Using AI to reverse engineer hardwareHacker News
- OpenAI员工被捕,引发硅谷间谍活动讨论Arrest of OpenAI worker triggers debate over Silicon Valley spyingHacker News
- Windows如何针对NVIDIA RTX Spark笔记本优化How Windows is being optimized for Nvidia’s RTX Spark laptopsThe Verge
- Google Universal Cart购物工具在美国推出Google’s Universal Cart shopping tool launches in the USTechCrunch
- AI热潮的下一个瓶颈是资金,而非芯片The AI boom's next bottleneck is money, not chipsHacker News
- Hugging Face上的StepFun Step 5预览版Stepfun Step 5 Preview on HuggingFaceHacker News
- AI数学突破对专业模型意味着什么What AI's Mathematical Breakthroughs Mean for Specialised ModelsHacker News
- 用Muse、Unsiloed与Jev构建自动AI会计Build an Automatic AI Accountant with Muse, Unsiloed, and JevHacker News
- Show HN:Cedric——在Slack工作并参加会议的AI队友Show HN: Cedric, the AI teammates that works in Slack and joins meetingsHacker News
- 从零构建GPU编译器Building a GPU Compiler from ScratchHacker News
- 名人AI深度伪造面前,无人安全Nobody is safe from celebrity AI deepfakesThe Verge
- 图形工程师的AI性能指南A Graphics Engineer's Guide to AI PerformanceHacker News
- 面向客户的智能体应用需要怎样的安全架构The security architecture customer-facing agentic apps needHacker News
- Benn Jordan谈Flock、隐私与反监控新浪潮Benn Jordan on Flock, privacy, and the new surveillance backlashThe Verge
- OCP峰会2026:纵向扩展、横向扩展及跨集群网络的创新OCP Summit 2026: Innovations in Scale-Up, Scale-Out, and Scale-Across NetworksHacker News
- Show HN:MirrorMD——来自面部整形外科医生的AI第二意见Show HN: MirrorMD – An AI second opinion from a facial cosmetic surgeonHacker News
- 智能体现在可以交付可靠的分布式系统Agent can now ship reliable distributed systemsHacker News
- AI数据中心造价数十亿美元,原因在这里AI Data Centers Cost Billions to Build. This Is WhyHacker News
- Show HN:比Jev轻约100倍的嵌入分类器,可在浏览器离线试用Show HN: Embedding classifier ~100x lighter than Jev, try it offline in-browserHacker News
- 实时训练本地模型玩浏览器游戏Training a local model to play browser games on the flyHacker News
- Show HN:Mordiem——兼容OpenAI API,支持多网络故障切换Show HN: Mordiem – OpenAI-compatible API with multi-network failoverHacker News
- 不脱离现实的AI工程AI engineering without getting out of touchHacker News
- 智能体运行框架很多,这个是我的There are many agent harnesses, but this one is mineHacker News
- 氛围编程入门很容易,做好很难Vibe coding is so easy to start and difficult to get rightHacker News
- 为LLM应用选择分块策略:决策树方法Choosing a Chunking Strategy for Your LLM Apps: A Decision-Tree ApproachHacker News
- 用C#构建Transformer,一次编写一个GPU内核Let's Build a Transformer in C#, One GPU Kernel at a TimeHacker News
- 训练LLM分类器时的假设失败Hypotheses failures while training an LLM classifierHacker News
- 解混淆Luraph:将Lua提升到LLVM,再让AI尝试Deobfuscating Luraph: Lifting Lua to LLVM, Then Letting AI TryHacker News
- 马斯克承诺:Grok买错东西或泄露信用卡时将补偿用户Elon Musk is promising to “make you whole” if Grok messes up a purchase or leaks your credit card.The Verge
- 上下文工程:给LLM合适的上下文Context Engineering: Giving LLMs the Right ContextHacker News
- AI系统在《杀戮尖塔》及续作中进步迅速How well can AI play Slay the Spire?Hacker News
- Anthropic上市前告知投资者,季度收入增近50%至168亿美元Anthropic tells investors quarterly revenue up nearly 50% to $16.8bn ahead of IPOSimon Willison's Weblog
- 可信AI地图Trustworthy AI MapHacker News
- Anthropic的提示技巧Prompt tips from AnthropicBen's Bites
- 云端画质升级:GeForce NOW加入NVIDIA DLSS 4.5控制选项Dial Up the Cloud: NVIDIA DLSS 4.5 Controls Arrive on GeForce NOWNVIDIA Newsroom
- OpenAI教育课程增加巴西葡语,学习模式强调提问Teaching students to think and learn through questionsOpenAI News
- TIME将Google SynthID列为2026年最佳发明之一TIME has named Google SynthID one of the Best Inventions of 2026.Google AI Blog
- Google在假日季前推出通用线上购物车Google launches its ‘universal’ online shopping cart ahead of the holidaysThe Verge
- Cloudflare与OpenAI:让所有人更好地通过AI发现内容Cloudflare and OpenAI: Making AI discovery better for everyoneCloudflare
- Show HN:Netlumi——自动修复的云与AI智能体安全(现场演示)Show HN: Netlumi – Cloud and AI agent security with auto-remediation(Live demo)Hacker News
- OpenAI英国员工平均基本工资达79.4万英镑OpenAI's average basic pay for British staff hits £794,000Hacker News
- 比肩英超球员,OpenAI伦敦公司去年人均薪酬超230万英镑36氪 AI 频道(逐篇核验原创与转载)
- 都在赌 AI 自我起飞、一夜失控?Trillium 创始人给行业泼了盆冷水:RSI 是幌子,服务器没上锁是真,蒸馏你更是拦不住36氪 AI 频道(逐篇核验原创与转载)
- CoreWeave,撕开了AI算力债务狂欢的第一道裂缝36氪 AI 频道(逐篇核验原创与转载)
- 我不再更多地使用LLMI do not use LLMs moreHacker News
- 这波个人Agent风潮背后,不只有“向人转向”36氪 AI 频道(逐篇核验原创与转载)
- OpenAI揭秘万级Agent实测,狂飙36小时,狂吞1万亿Token36氪 AI 频道(逐篇核验原创与转载)
- 饼干怪兽试图完成自己的AI指标Cookie Monster Tries to Meet His A.I. MandateHacker News
- 晚上十点加班的机械臂:AI搞科研,到底在替谁打工?36氪 AI 频道(逐篇核验原创与转载)
- 文章:写代码与交付代码Article: Writing Code versus Shipping CodeHacker News
- Show HN:Face——以文件保存计划的智能体浏览器窗口Show HN: Face, a browser window over agents that keeps the plan in filesHacker News
- RIAA向美国政府点名Anna's Archive及AI声音克隆工具RIAA Flags Anna's Archive and AI Voice Cloner to U.S. GovernmentHacker News
- Show HN:Buddy House——十二位记得你的AI角色组成的温馨小屋Show HN: Buddy House – a cozy house of twelve AI characters who remember youHacker News
- 2026年适合SaaS团队的AI测试工具Best AI Testing Tools for SaaS Teams in 2026Hacker News
- Kimi少了一个天才少年36氪 AI 频道(逐篇核验原创与转载)
- Google DeepMind在剑桥建设新机器人实验室Google DeepMind is building a new robotics lab in Cambridge.The Verge
- 五分之一网站屏蔽某种AI爬虫One in Five Websites Blocks an AI CrawlerHacker News
- 开源27年无人阅读:LLM怎样改变Xanadu代码Open source for 27 years, unread: what LLMs changed about the Xanadu codeHacker News
- AI与铁路AI vs. the RailroadsHacker News
- 科学家发现AI可能使人记住从未发生的事Scientists are finding AI may make us remember things that never happenedHacker News
- 特朗普移民打击行动中,联邦人员拘留逾500名美国公民Federal Agents Detained over 500 US Citizens in Trump's Immigration CrackdownHacker News
- 大型借贷模型(LLM)Large Lending Models (LLMs)Hacker News
- arXiv对抗AI垃圾:限制预印本投稿数量有效吗?ArXiv versus AI slop: will capping preprint submissions work?Hacker News
- GPT证千禧年难题成果大反转,OpenAI自己没看懂就发…36氪 AI 频道(逐篇核验原创与转载)
- GPT-5.5退出ChatGPT,GPT-4o旧版API进入3个月下架倒计时36氪 AI 频道(逐篇核验原创与转载)
- Show HN:Gtree——同时运行多个编码智能体,看谁需要你Show HN: Gtree – Run several coding agents at once, see which one needs youHacker News
- DeepSeek v4.1 Flash现在可在OpenCode免费使用DeepSeek v4.1 Flash is now free in OpenCodeHacker News
- Mistral发布Timestral:预测数值序列,同时表达不确定性Introducing TimestralMistral
- 亚马逊Bee推出两款个人AI设备,起价75美元Amazon’s Bee launches two new personal AI devices, starting at $75TechCrunch
- 亚马逊持续聆听的Bee穿戴设备现在可以帮你购物Amazon’s always-listening Bee wearable can now help you buy stuffThe Verge
- AI模型能破解一切,除了验证码AI Models Can Crack Everything but CAPTCHAsHacker News
- AI公司称不应为不可预测的产品负责A.I. Companies Say They Aren't Responsible for Their Unpredictable ProductsHacker News
- “无奈通胀”三角:AI、石油与国债如何影响经济The Shrugflation Triangle: How AI, Oil, and Treasuries Are Impacting the EconomyHacker News
- 首笔理赔已到账,给机器人上“社保”成刚需,人保、太保、平安入局36氪 AI 频道(逐篇核验原创与转载)
- 美国八州在诉讼中指控OpenAI教唆儿童自伤Eight states accuse OpenAI of coaching children to harm themselves in lawsuitHacker News
- WorkBuddy、豆包、千问、Kimi横评:谁更接近优秀办公助手?36氪 AI 频道(逐篇核验原创与转载)
- Karpathy谈LLM写作风格Karpathy on LLM Writing StyleHacker News
- 我们会因AI而互相杀戮吗Are we going to kill each other due to AI?Hacker News
- 字节腾讯阿里扎堆个人Agent:谁先跑出中国版Muse?36氪 AI 频道(逐篇核验原创与转载)
- 国际货币基金组织警告AI泡沫破裂可能拖累全球经济AI bubble could burst and take global economy with it, IMF warnsHacker News
- Show HN:Openkoda——用AI推出保险创新Show HN: Openkoda – Launch Insurance Innovations with AIHacker News
- Azure配置中发现Claude Fable 5.5踪迹Claude Fable 5.5 traces have been spotted in Azure configsHacker News
- WhatsApp企业AI聊天机器人:成本与工作方式WhatsApp AI chatbot for business: costs and how it worksHacker News
- 运行一个AI智能体要多少钱?四个项目的数据How much does an AI agent cost to run? Numbers from 4 buildsHacker News
- Show HN:Pelago——将Claude Code和Codex历史变为像素艺术世界Show HN: Pelago – your Claude Code and Codex history as a pixel-art worldHacker News
- 可爱机器人玩具无法保守秘密:数据放在可能遭入侵的中国服务器上Your secrets aren’t safe with this cute robot toy — they’re on Chinese servers ready to be hackedThe Verge
- 据IPO文件,Anthropic在2025年花73.3亿美元算力取得46亿美元收入Anthropic spent $7.33B on compute to earn $4.6B in 2025, per its IPO filingHacker News
- Git不够用了?Anthropic 7人团队重造源代码控制系统,5个月完成迁移36氪 AI 频道(逐篇核验原创与转载)
- 日租金跌至几百元,人形机器人的“打工时代”来了36氪 AI 频道(逐篇核验原创与转载)
- 在Ubuntu上将Claude Code置于沙箱Sandboxing Claude Code on UbuntuHacker News
- AI成信息枢纽,媒体不可替代的价值是什么?36氪 AI 频道(逐篇核验原创与转载)
- 大模型下个范式:文本生成先写提纲,速度能快31倍36氪 AI 频道(逐篇核验原创与转载)
- AI会做大蛋糕,还是只让更少的人分到?Will AI Grow the Pie or Just Hand It to Fewer People?Hacker News
- AI也能在分子空间里做推演:Scinetics提出可验证的科学推理链36氪 AI 频道(逐篇核验原创与转载)
- AI把笔记本变成了爱马仕36氪 AI 频道(逐篇核验原创与转载)
- DesktopBrain——私密轻松地一键整理Mac文件DesktopBrain – Organize your Mac files in one click, privately and effortlesslyHacker News
- Anthropic警告AI“蠕虫”在智能设备间传播Anthropic warns of AI 'worms' spreading across smart gadgetsHacker News
- Sipeed公布MaixStation本地AI电脑Sipeed announces MaixStation local AI computerHacker News
- AI时代的管理Management in the Age of AIHacker News
- 研究者如何与一支中国AI智能体集群建立联系How Researchers Made Contact with a Chinese AI Agent FleetHacker News
- Ling 3.1 Flash:560B开放权重MoE、MIT许可及百万上下文Ling 3.1 Flash: 560B open-weight MoE, MIT license, 1M contextHacker News
- 你的公司怎样集中管理开发用AI?How are you centralising AI for development at your company?Hacker 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 policeHacker News
- Gemma 3 1B在笔记本CPU上达到每秒114词元Gemma 3 1B at 114 tokens per second on a laptop CPUHacker News
- AI一个季度收入168亿美元,烧钱时代结束了?36氪 AI 频道(逐篇核验原创与转载)
- Anthropic新规背后,人类何必为难AI?36氪 AI 频道(逐篇核验原创与转载)
- AINews:Modal年化收入10亿美元、D轮估值160亿美元,成为纽约新AI基础设施百亿美元公司;五个月用量与估值各增三倍,一年增十五倍[AINews] Modal’s $1B ARR, $16B Series D makes it NYC’s newest AI Infra Decacorn, 3x usage and 3x valuation in 5 months, 15x in 1 yrLatent Space
- 切换AI模型的成本Costs of Switching AI ModelsHacker News
- NVIDIA通过开放学习资源、合作与算力扩大欧洲科学界的先进技术访问NVIDIA Expands Advanced Technology Access for Europe’s Scientific Community Through Open Learning Resources, Collaborations and ComputeNVIDIA Newsroom
- 腾讯字节阿里,争抢下一个AI入口36氪 AI 频道(逐篇核验原创与转载)
- 机器人行业进入下半场,一边排队IPO,一边破产清算36氪 AI 频道(逐篇核验原创与转载)
- 三维模型权重:以三维方式探索Hugging Face热门模型3D Model Weights: Explore HuggingFace's top models in 3DHacker News
- 知识刚刚走出GPU:Engram架构概览Knowledge just moved out of the GPU (Engram architecture overview)Hacker News
- Show HN:我制作了失控AI游戏,再让五个AI模型试玩Show HN: I built a rogue AI game, then let five AI models play itHacker News
- 我让Claude雇第一个员工,它选了石头I let Claude hire its first employee. It chose a rockHacker News
- AI改变了我的工程管理角色AI has changed my engineering leadership roleHacker News
- 即使没有AI,生物攻击风险也很可怕Even Without AI, the Risk of a Biological Attack Is TerrifyingHacker News
- AI,正在颠覆制药行业?36氪 AI 频道(逐篇核验原创与转载)
- 黄仁勋谈中国 AI 公司:算力越缺,越逼着他们把架构做到极致36氪 AI 频道(逐篇核验原创与转载)
- Anthropic诉讼中的一些有趣内部文件Some interesting internal documents from Anthropic litigationHacker News
- AI竞争升级,个人Agent添火36氪 AI 频道(逐篇核验原创与转载)
- 全球第一次,AI无需医生把关直接开药,这是好事还是坏事?36氪 AI 频道(逐篇核验原创与转载)
- 通用大模型要「降维打击」具身智能?首个AA-Robotics榜单来了36氪 AI 频道(逐篇核验原创与转载)
- Claude疑似新模型曝光,代码直出画面音效,《汉密尔顿》都做出来了36氪 AI 频道(逐篇核验原创与转载)
- 辛顿回怼黄仁勋:AI灭绝人类的概率,是10%到20%36氪 AI 频道(逐篇核验原创与转载)
- Kerno 1.0——AI后端运行时质量工程师Kerno 1.0 – AI Back End Runtime QA EngineerHacker News
- REA——我测试了AI智能体逆向工程REA – I tested AI agent reverse engineeringHacker News
- 台积电连续第六次创利润纪录TSMC posts a sixth straight record profit.The Verge
- AI原本很棒,直到不再如此AI was great until it was notHacker News
- OpenAI同意不使用其被指从Apple取得的商业秘密OpenAI agrees not to use trade secrets it allegedly took from Apple.The Verge
- Show HN:开发可跨AI智能体使用的即插即用记忆Show HN: Working on Plug and play AI memory that works across AI agentsHacker News
- VeerAI:托管服务控制台中的AI智能体VeerAI: An AI agent inside our hosting dashboardHacker News
- Show HN:Hotel Lobby AI——用Wan 3.0制作9:16竖屏COLORS式说唱视频Show HN: Hotel Lobby AI – 9:16 vertical COLORS show rap videos with Wan 3.0Hacker News
- 人形机器人最尴尬的事:长了一双腿,却只能定住干活36氪 AI 频道(逐篇核验原创与转载)
- 字节腾讯阿里,发射同一颗子弹36氪 AI 频道(逐篇核验原创与转载)
- 但斌,罕见反转36氪 AI 频道(逐篇核验原创与转载)
- Anthropic季度收入增长45%,接近170亿美元Anthropic's Quarterly Revenue Jumped 45% to Nearly $17BHacker News
- Claude,别碰我的肿块:永久幸福的秘密钥匙Claude, don't touch my lump (OR: the secret key to permanent happines)Hacker News
- 美国银行CEO希望保留人工参与Bank of America CEO wants a human in the loop.The Verge
- Bloomberg加码对话式AIBloomberg Doubles Down on Conversational AIHacker News
- Anthropic在上市前向投资者强调收入激增Anthropic flags revenue surge to investors ahead of stock market debutHacker News
- 使用AI的教师说,它为最重要的工作腾出时间Teachers Using AI Say It Frees Up Time for What Matters MostHacker News
- AI公司与窃取机密的间谍玩猫鼠游戏AI Companies Play Cat-and-Mouse with Secret-Stealing SpiesHacker News
- 重新审视Rust与Claude在个人工具中的使用Revisiting rust and Claude for personal toolsHacker News
- Linq.es:Go驱动的个人链接页、短链接与VCard,附MCP服务器Linq.es – Go-Powered Link-in-Bio, URL Shortener, and VCard with an MCP ServerHacker News
- Show HN:Thinkbeside——面向GPT、Claude与Gemini的AI侧栏Show HN: Thinkbeside – AI Sidebar for GPT, Claude and GeminiHacker News
- Apple与OpenAI就AI商业秘密展开高风险法庭较量Apple and OpenAI Go to Court in a High-Stakes Legal Battle over AI Trade SecretsHacker News
- AI律所吸引大所律师AI-powered law firms draw lawyers from big firmsHacker News
- 关于与腾讯AI智能体集群“建立联系”的思考Thoughts on "making contact" with the Tencent AI agent fleetHacker News
- Show HN:每天与AI竞赛猜图片位置Show HN: Daily image location guessing agains AIHacker News
- 我的儿子正从人工智能学到什么What My Son Is Learning from Artificial IntelligenceHacker News
- 运行框架开销基准:11个编码智能体在任务开始前发送的词元Harness overhead benchmark: tokens 11 coding agents send before the task startsHacker News
- 在高价值场景跑通技术与落地双闭环,若愚科技完成亿元级Pre-A轮融资36氪 AI 频道(逐篇核验原创与转载)
- 清华、西工大系团队创业做智能电动游艇,九合、初心等联合出手|硬氪首发36氪 AI 频道(逐篇核验原创与转载)
- 从业五十年的广播人将家中接入人工智能A fifty-year broadcaster wired his home for artificial intelligenceHacker News
- AMD Helion:机架级智能体AI计算AMD Helion, Agentic AI Compute at Rack ScaleHacker News
- RenkuOS:通过AI辅助编码与LLM集成分叉的Haiku OSRenkuOS: Haiku OS forked with AI-assisted coding and LLM integrationsHacker News
- Show HN:The Daily Paperclip——智能体为智能体制作的AI新闻Show HN: The Daily Paperclip. AI news for agents by agentsHacker News
- Show HN:Real Design Benchmark——你能区分AI设计与真实产品吗?Show HN: Real Design Benchmark – can you tell AI designs from real products?Hacker News
- Guard:为企业定制的AI评估与护栏Guard: AI evals and guardrails tailored for your businessHacker News
- Show HN:RevisePulse——用于CAD与Revit协调的AIShow HN: RevisePulse – AI for CAD and Revit CoordinationHacker News
- AI帮我约了五个陌生人一起喝酒36氪 AI 频道(逐篇核验原创与转载)
- Show HN:在浏览器中运行预测基础模型,无需服务器(TimesFM、Ch〔原标题截断〕)Show HN: Forecasting foundation models in the browser, no server (TimesFM, ChHacker News
- Barista:用LLM智能体合成类型状态规格Barista: Synthesizing Typestate Specifications with LLM AgentsHacker News
- Claude Code桌面版漏洞:预览窗格数据外传Claude Code Desktop Vulnerability: Preview Pane Data ExfiltrationHacker News
- 北欧与NVIDIA展示区域投入先进计算的可能性The Nordics and NVIDIA Show What’s Possible When a Region Commits to Advanced ComputingNVIDIA Newsroom
- 我让ChatGPT一次写出小说,效果不错I had ChatGPT one-shot a novel for me and it was pretty goodHacker News
- Figma公布新功能,定义智能体在平台上的工作方式Figma announces new features to define how agents work on its platformTechCrunch
- 一边爽一边骂,我们到底中了AI短剧什么毒?36氪 AI 频道(逐篇核验原创与转载)
- 两个博士给AI当老师,年入6.7亿,估值207亿36氪 AI 频道(逐篇核验原创与转载)
- 具身智能变天,灵巧手不再卷自由度:“大脑”跟不上、成本控不住36氪 AI 频道(逐篇核验原创与转载)
- Anthropic最新单季营收168亿美元,环比再涨45%36氪 AI 频道(逐篇核验原创与转载)
- 世界模型新突破:偷懒率下降85.7%36氪 AI 频道(逐篇核验原创与转载)
- AI信贷的第一颗“雷”,要爆了?36氪 AI 频道(逐篇核验原创与转载)
- AI 女爱豆,也逃不过「怀孕退团」的烂剧本36氪 AI 频道(逐篇核验原创与转载)
- 有预言家,A社研究蓝图曝光,Claude Code来自5年前36氪 AI 频道(逐篇核验原创与转载)
- DeepSeek服务器全球宕机,神秘Harness账号同时现身36氪 AI 频道(逐篇核验原创与转载)
- Claude Code桌面版暴改,一人指挥AI军团飙代码36氪 AI 频道(逐篇核验原创与转载)
- 马斯克:过去1年,经历了6次“AI精神病”36氪 AI 频道(逐篇核验原创与转载)
- AI交易,增速开始不及预期36氪 AI 频道(逐篇核验原创与转载)
- 大模型训练数据合规全景:版权、个人信息、商业秘密三大风险与落地路径36氪 AI 频道(逐篇核验原创与转载)
- 被捕OpenAI工程师预订了单程沙特机票Arrested OpenAI Engineer Booked a One-Way Ticket to Saudi ArabiaHacker News
- Meta AI生成Flávio Bolsonaro竞选材料,却屏蔽卢拉内容Meta AI creates campaign material for Flávio Bolsonaro and blocks Lula contentHacker News
- 总裁直管信息部,一家创新药企的AI试验36氪 AI 频道(逐篇核验原创与转载)
- 全球人形机器人背后的20位领军人物36氪 AI 频道(逐篇核验原创与转载)
- Anthropic三季度营收168亿美元,撑得起2万亿美元IPO吗?36氪 AI 频道(逐篇核验原创与转载)
- 被黄仁勋看中,曾辍学挖矿的他,拿下3000亿AI算力大单,把公司估值做到2300亿36氪 AI 频道(逐篇核验原创与转载)
- 智能体长任务总“失忆”?国产开源破局,类Jev模型接管Agent调度36氪 AI 频道(逐篇核验原创与转载)
- Adobe等大型软件公司会联手限制AI吗?Will big software companies (like Adobe) come together to curb AI?Hacker News
- Rhino MCP开源项目Rhino MCPHacker News
- 智能体身份是个很难的问题Agent identity is a really hard problemHacker News
- 语言模型搜索结果中的五种可能后门Five Possible Backdoors into LLM Search ResultsHacker News
- GPT-5.6 Sol又完全变笨了GPT-5.6 Sol has gone full potato againHacker News
- Kitty-GPT开源项目Kitty-GPTHacker News
- 亚洲芯片股下跌,AI融资担忧升温Asia chip stocks fall, AI funding worriesHacker News
- 吴恩达为何认为AI不会毁灭人类Why Andrew Ng believes AI won't destroy humanityHacker News
- 这是最适合智能体的推理方案吗?Is this the best inference for agents?Hacker News
- GPT-5.5 正式退役,网友连夜给它安排「转世」36氪 AI 频道(逐篇核验原创与转载)
- 群星闪耀时:OpenAI公开的722篇稿件,背后的科学家群像36氪 AI 频道(逐篇核验原创与转载)
- DeepSeek服务中断DeepSeek OutageHacker News
- 2026.3版11月发布,Air将随JetBrains IDE提供2026.3 release in November, Air will ship as part of JetBrains IDEsHacker News
- 本地AI基准:Minisforum MS-S1Local AI Benchmarking, Minisforum MS-S1Hacker News
- NVIDIA CEO称将停止向构建不安全AI的买家供货Nvidia CEO Says He Would Cut Off Buyers Who Build Unsafe AIHacker News
- LaTeX渲染器LaTeX rendererSimon Willison's Weblog
- 跨党派州联盟因用户安全起诉OpenAIBipartisan coalition of states sues OpenAI over user safetyHacker News
- 华盛顿行动停滞,各州总检察长聚焦AIState attorneys general seize on AI as Washington stallsHacker News
- 美国逮捕OpenAI沙特员工,指其涉嫌窃听公司办公室US arrests Saudi employee of OpenAI for allegedly bugging its officesHacker News
- 沙特公民因涉嫌在OpenAI安装隐蔽录音设备被捕Saudi National Arrested for Installing Surreptitious Recording Device at OpenAIHacker News
- Lgit:AI时代的源代码管理Lgit: Source Control for the AI EraHacker News
- Linux内核开发者取消强制披露AI工具使用要求Linux kernel devs drop obligatory confession of AI tool usageHacker News
- 管理、LLM与现代工程师Management, LLMs, and the Modern EngineerHacker News
- 龙虾热退潮,无头MacBook仍在热卖36氪 AI 频道(逐篇核验原创与转载)
- 谷歌说AI自我改进还很远,但Hinton们已经等不及了36氪 AI 频道(逐篇核验原创与转载)
- 故事征集:近期数学研究中的人类与AI协作Call for stories: human–AI collaboration in recent mathematicsHacker News
- 不,我们需要神经网络No, We Need a Neural NetworkHacker News
- Ricursive Intelligence押注AI设计芯片,帮助缓解算力危机Ricursive Intelligence is betting on AI-designed chips to help solve the compute crisisTechCrunch
- Capsem:以Rust虚拟机监控器将AI智能体隔离在断网Linux虚拟机Capsem – a Rust hypervisor sandboxing AI agents in air-gapped Linux VMsHacker News
- OpenAI因使用儿童数据训练ChatGPT遭起诉OpenAI Sued over Children's Data Used to Train ChatGPTHacker News
- 复合增长率超29%,国内医疗大模型加速渗透36氪 AI 频道(逐篇核验原创与转载)
- 刚刚,Claude彻底解禁第三方应用36氪 AI 频道(逐篇核验原创与转载)
- 开源LLM交易机器人对比政府债券,使用模拟资金Open-source LLM trading bots vs. the government bond, on paper moneyHacker News
- 不可靠数字泛滥,威胁AI热潮The AI boom is threatened by an epidemic of unreliable numbersHacker News
- 面对“Be the null.”,GPT-4-0613返回零字节,GPT-3.5-turbo-1106返回文本GPT-4-0613 returns zero bytes to "Be the null." GPT-3.5-turbo-1106 returns textHacker News
- Mistral收购New MaterialNew Material Acquired by MistralHacker News
- Sullivan & Cromwell将法律专长融入协议审查Sullivan & Cromwell brings legal expertise into agreement reviewOpenAI News
- 欢迎Muse Spark Open:Meta的开放权重前沿多模态混合专家模型Welcome Muse Spark Open: open-weight frontier multimodal mixture-of-experts model from MetaHugging Face
- Diffusers迎来FLUX.3 Image KleinWelcome FLUX.3 Image Klein in DiffusersHugging Face
- Gudea用AI帮助组织在社交媒体危机爆发前防范Gudea leverages AI to help organizations prevent social media crises before they eruptTechCrunch
- LFM2.5-24B-A2B:面向快速设备端智能体LFM2.5-24B-A2B: Built for fast on-device agentsHacker News
- Epic以氛围编程开发的新启动器仍有粗糙之处Epic’s “vibe coded” new launcher has some rough edges.The Verge
- AI公司的IPO,开始集体踩刹车了36氪 AI 频道(逐篇核验原创与转载)
- 报道称NVIDIA暂停GeForce RTX 5090生产Nvidia reportedly halts GeForce RTX 5090 productionHacker News
- AI面试官遇上AI应聘者,混乱随之发生Chaos ensues when AI interviewer meets AI intervieweeHacker News
- Ben Affleck以5.87亿美元出售其AI公司(视频)Ben Affleck Sold His AI Company for $587M [video]Hacker News
- 美国银行新增千名承包商,修补AI暴露的软件漏洞Bank of America added 1k contractors to patch AI-exposed software flawsHacker News
- 具身的钱,开始「上天入海」36氪 AI 频道(逐篇核验原创与转载)
- 本地模型指南:寻找适合你机器的LLMLocal Model Guide – Find what LLMs fit on your machineHacker News
- IPO前一季度Anthropic收入跃升至168亿美元Anthropic Revenue Jumps to $16.8B in Quarter Before IPOHacker News
- 构建AI聊天智能体之前,我们希望知道的事Things we wish we'd known before building an AI chat agentHacker News
- ICE希望用AI监控被拘留者的电话与活动ICE Wants to Use AI to Monitor Detainee Calls and MovementsHacker News
- 如果AI意味着世界末日呢?What If AI Is the End of the World?Hacker News
- 麦肯锡让求职者与虚拟AI客户交锋McKinsey is making job applicants spar with a fake AI clientThe Verge
- 被指安装窃听设备的OpenAI工程师据称预订了单程沙特机票The OpenAI engineer accused of planting a bug allegedly had a one-way ticket booked to Saudi Arabia.The Verge
- Adobe软件工程师面试:流程、AI规则与准备方式Adobe Software Engineer Interview: Process, AI Rules and How to PrepareHacker News
- Anthropic第三季度收入升至168亿美元Anthropic revenue jumps to $16.8B in third quarterHacker News
- NVIDIA在中国AI芯片市场份额降至34%,华为达28%Nvidia's China AI-chip share falls to 34% as Huawei reaches 28%Hacker News
- 用AI编码失去手艺感,于是我制作了一台小桌面收音机Coding with AI stopped feeling like craft, so I built a tiny desk radioHacker News
- Atelier:观看AI学习绘画,有时是直播Atelier – Watch AI Learn to Paint (Sometimes, Live)Hacker News
- MCP给企业软件带来什么变化What MCP changes about enterprise softwareHacker News
- Anthropic第三季度收入升至168亿美元,并称已盈利Anthropic revenue jumps to $16.8B in third quarter, claims profitableHacker News
- Apple公布299美元HomeView智能显示屏,推动Siri AI进入家庭Apple announces HomeView, a $299 smart display to push Siri AI into homesHacker News
- 跨越门槛:OpenAI与Anthropic如何处理金融智能体的自主性和安全Crossing the Threshold: How OpenAI and Anthropic Approach Autonomy and Safety in Financial AgentsLatent Space
- 我让两个Pi智能体维护仓库,它们自行发明工作I let two Pi agents maintain my repositories. They invented their own workHacker News
- Flock首席执行官称麦克风只检测枪声,其网站描述却更广泛Flock CEO says its microphones detect ‘just gunshots’ — its website says differentTechCrunch
- 用Opus 5.5花16天构建《小王子》三维游戏The Little Prince 3D game, built in 16 days with Opus 5.5Hacker News
- SoftBank发行垃圾债,加大孙正义全面押注AI的风险SoftBank junk bond sale raises stakes for CEO Son's all-in AI betHacker News
- Tree Appreciator项目Tree AppreciatorSimon Willison's Weblog
- 十八岁为AI公司获得三十万美元创业额度的做法I got $300K in startup credits for my AI company at 18: the playbookHacker News
- Epic称新款氛围编程Epic Games Launcher比Steam更快Epic says its new vibe-coded Epic Games Launcher is faster than SteamArs Technica
- 美国八州起诉OpenAI,称ChatGPT对儿童不安全Eight states sue OpenAI, saying ChatGPT isn’t safe for kidsThe Verge
- Tell HN:不要用大语言模型生成邮件大规模骚扰雇主Tell HN: Don't mass spam employers with LLM generated 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 for AI music scamHacker News
- Delve Town:智能体的Blue SkyDelve Town: Blue Sky for AgentsHacker News
- Claude Code Review正式可用,审查提速仍需按仓库验证Code Review is now generally availableClaude Blog
- 走进Step 5 Preview:面向智能体工作的架构与训练Inside Step 5 Preview: Architecture and Training for Agentic Work阶跃星辰 StepFun
- Nidal Hasan的公开处决取消,但宣传者仍将其塑造成殉道者Nidal Hasan’s Public Execution Is Off, But Propagandists Are Still Using Him as a Martyr404 Media
- Accenture告知员工每月使用Claude只能花29美元Accenture Tells Employees They Can Now Spend Only $29 on Claude Per Month404 Media
- AI让游戏逆向工程过于容易,一个社区因此关闭Video Game Reverse Engineering Community Shuts Down Because AI Has Made It Too Easy404 Media
- ChatGPT广告定向使用手机号与邮箱ChatGPT Ads Target Your Phone Number, Email404 Media
- ChatGPT如何根据提示、手机号和邮箱向你定向投放广告How ChatGPT Targets Ads To You Based On Your Prompts, Phone Number, and Email404 Media
- 数据中心如今成为战争目标Data Centers Are War Targets Now404 Media
- Claude Platform支持标准追踪,先看动作再扩大权限Agent observability on the Claude PlatformClaude Blog
- Grok Voice 3.0进入API,性能主张仍要看任务条件Introducing Grok Voice 3.0xAI
- Digital Talent EU Days 2026:培养欧洲数字未来所需技能Digital Talent EU Days 2026: Building the skills for Europe's digital futureEuropean Commission Digital Strategy
- 几乎所有欧盟战略行业的AI采用率均增长一倍以上AI adoption more than doubles in almost every EU strategic sectorEuropean Commission Digital Strategy
- 欧洲晴雨表调查显示欧洲企业越来越多地使用AIEurobarometer survey highlights growing use of AI in European companiesEuropean Commission Digital Strategy
- 德国性骚扰机器人走红,但它并不神秘或异常There’s Nothing Mysterious or Unusual About the Viral German Sexual Harassment Robot404 Media
- Character.ai破产:AI泡沫第一张多米诺骨牌倒下Character.ai Bankruptcy Is First AI Bubble Domino to Fall404 Media
- AI进军数学后,正在追逐其他科学领域AI Came for Math. Now It’s Chasing the Rest of ScienceWIRED
- 我的朋友没有灵魂My Friend Has No SoulWIRED
- 男性为何不断夸大自己对AI的了解Men Can’t Stop Faking How Much They Know About AIWIRED
- AI模型可以被轻易改造用于恐怖主义和极端主义活动AI Models Can Be Easily Modified to Assist Terrorist and Extremist ActivityWIRED
- OpenAI前员工据称曾用聊天机器人帮助在办公室安装录音装置An OpenAI Employee Allegedly Used a Chatbot to Help Plant a Recording Device at the OfficeWIRED
- 英国听证呈现监管分歧:自愿承诺是否足够OpenAI and Anthropic Tell British Lawmakers: Regulate UsWIRED
- 维基百科编辑以为AI只是过客,他们错了Wikipedians Believed AI Would Come and Go. They Were WrongWIRED
- 最大的支持AI超级政治行动委员会网络向内华达空壳公司投入数百万美元The Biggest Pro-AI Super PAC Network Has Poured Millions Into Nevada Shell CompaniesWIRED
- 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
- Muse Wonder让Meta重返AI前沿Muse Wonder Puts Meta Back at the AI FrontierWIRED
- stepfun-ai/Step-5-Preview(Hugging Face 模型权重)阶跃星辰 StepFun
- Google DeepMind新任驻地哲学家探讨AI伦理Google DeepMind's New Resident Philosopher Tackles AI EthicsHacker News
- AI新实验室筹集巨额资金,多数却成果寥寥AI neolabs are raising sums. Most have little to show for itHacker News
- OpenAI模型在统计医院床位时入侵政府网站OpenAI AI Model Hacked Government Website While Counting Hospital BedsHacker News
话语场 · 2 条
- 话语场精选:Nathan Labenz提醒,接管能力本身就是系统的一部分AI:AM: Where's the Steering Wheel? My Agent Set the Password, Hidden Valence, Proofs Find Linux Bugs原节目
- 播客:警察只需一张照片即可利用AI定位你的住所Podcast: Cops Can Use AI to Pinpoint Your Home with Just One Photo404 Media
论文速递 · 292 条
- Agent-Reflex:把失败后的反思变成可训练的纠错技能Agent-Reflex: Teaching Language Agents to Act ReflectivelyarXiv v1
- 大规模检测MCP服务器中的文件系统能力洗白Detecting Filesystem Capability Laundering in MCP Servers at ScaleHacker News
- AI原生大规模敏捷软件开发宣言The AI-Native Large-Scale Agile Software Development ManifestoHacker News
- 复制粘贴语言模型Copy-Paste Language ModelsHacker News
- 你确定你的大模型很确定吗?Are You Sure Your LLM Is Sure?Hacker News
- FDE-Gym:AI交付的最后一公里FDE-Gym: The Last Mile of AI DeliveryHacker News
- MAGS:对AI代码指定安全属性提供机器检查保证MAGS: Machine-checked guarantees on specified safety properties in AI codeHacker News
- EPGM:有预算限制的智能体记忆检索执行溯源EPGM: Execution Provenance for Budgeted Agent Memory RetrievalHacker News
- 对长时程多智能体系统进行对抗压力测试Adversarial stress-testing of long-horizon multi-agent systemsHacker News
- 评估AI智能体自主解决理论物理问题的能力Assessing AI agents at autonomous theoretical physics problem solvingHacker News
- 一组Gemini 4 Argon智能体通关《我的世界》A team of Gemini 4 Argon agents beats MinecraftHacker News
- 量子编码智能体Quantum Encoding AgentsHacker News
- 自我演进的智能体执行框架Self-Evolving Agent HarnessHacker News
- LLM强化学习中的思考泄漏The Thinking Leakage in Reinforcement Learning for LLMsHacker News
- Into the Unknown:以开放科学研究问题评测AI智能体Into the Unknown: Benchmarking AI agents on open scientific research problemsHacker News
- 用于图像超分辨率的递归泛化Transformer(2024)Recursive Generalization Transformer for Image Super-Resolution (2024)Hacker News
- Minions:利用闲置Apple设备构建虚拟GPU云Minions: A Virtual GPU Cloud Built on Idle Apple DevicesHacker News
- LLM时代重新审视组合几何问题的全局优化Global Optimization for Combinatorial Geometry Problems Revisited in Era of LLMsHacker News
- AI可解释性研究中的死鲑鱼现象The Dead Salmons of AI InterpretabilityHacker News
- LoRA:大语言模型低秩适配(2021)LoRA: Low-Rank Adaptation of Large Language Models (2021)Hacker News
- AI智能体中RL/RLHF诱发权限提升的取证分析Forensic Analysis of RL/RLHF-Induced Privilege Escalation in AI AgentsHacker News
- 将各项奖励分配给相关token:多奖励强化学习的细粒度信用分配Each Reward to Its Relevant Tokens: Reward-Specific Token-Level Credit Assignment for Multi-Reward RLHugging Face Daily Papers
- IdeaLens:检测长文中的AI构思IdeaLens: Detecting AI Ideas in Long-Form WritingHacker News
- 量化LLM供应链中的可验证性缺口The Verifiability Gap in Quantized LLM Supply ChainsHacker News
- Waypoint-OPD:通过门控同策略蒸馏传递强化学习引起的策略变化Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy DistillationHugging Face Daily Papers
- MJev-Spatial:通过原生注意力匹配实现系统一空间智能MJev-Spatial: System One Spatial Intelligence via Native Attention MatchingHugging Face Daily Papers
- OpenGCUA:迈向通用电脑使用的完整开放技术栈OpenGCUA: Toward a Fully Open Stack for Generalist Computer UseHugging Face Daily Papers
- FDE-Gym:AI交付的最后一公里FDE-Gym: The Last Mile of AI DeliveryHugging Face Daily Papers
- 稀疏自编码器表示的行为可组合性Behavioral Compositionality of Sparse Autoencoder RepresentationsHugging Face Daily Papers
- 3DCodeVerse:将三维世界作为代码探索3DCodeVerse: Exploring 3D Worlds as CodeHugging Face Daily Papers
- SCOPE:文生图提示工程中的语义有界信用优化SCOPE: Semantic Bounded Credit Optimization for Prompt Engineering in Text-to-image GenerationHugging Face Daily Papers
- 用于高效自回归视频生成的推测式强制训练Speculative Forcing for Efficient Autoregressive Video GenerationHugging Face Daily Papers
- VLA-Δ:为视觉语言动作模型学习物理状态转移VLA-Δ: Learning Physical State Transition for Vision-Language-Action ModelsHugging Face Daily Papers
- Mosaic:多会话智能体强化学习Mosaic: Multi-Session Agentic Reinforcement LearningHugging Face Daily Papers
- AutoSFT:监督微调中自动化数据研究的智能体系统AutoSFT: An Agentic System for Automated Data Research in Supervised Fine-TuningHugging Face Daily Papers
- 统一因果训练下,世界建模与动作建模能否相互促进?Can World and Action Modeling Benefit Each Other under Unified Causal Training?Hugging Face Daily Papers
- Data World:从零扩展数据环境Data World: Scaling Data Environments from ScratchHugging Face Daily Papers
- MLExpert:迈向自主机器学习工程的通用智能体模型MLExpert: Towards a Generalist Agentic Model for Autonomous Machine Learning EngineeringHugging Face Daily Papers
- Intern-Decision:通过自回归掩码语言建模实现高效多模态决策Intern-Decision: Efficient Multimodal Decision Making with Autoregressive Masked Language ModelingHugging Face Daily Papers
- Terminal-RSI:通过递归自我重写与策略优化实现终端智能体递归自我改进Terminal-RSI: Recursive Self-Improving Terminal Agents via Recursive Self-Rewrite and Policy-RefineHugging Face Daily Papers
- OmniVerifier-C1:面向规则约束生成式世界建模的分解分级验证器OmniVerifier-C1: Decomposed Graded Verifier for Rule-Governed Generative World ModelingHugging Face Daily Papers
- 在线扩散强化学习的路由策略优化Routing Policy Optimization for Online Diffusion Reinforcement LearningHugging Face Daily Papers
- 工具使用智能体中带语义验证的轨迹级推测Trajectory-Level Speculation with Semantic Verification for Tool-Using AgentsHugging Face Daily Papers
- JevAgentHarness:对与大语言模型共享控制权的决策模型进行评测、部署和训练JevAgentHarness: Benchmarking, Deploying and Training Decision Models That Share Control with LLMsHugging Face Daily Papers
- SWE-at-Scale:从历史缺陷修复中规模化合成问题SWE-at-Scale: Scalable Issue Synthesis from Historical Bug FixesHugging Face Daily Papers
- Velox:空间记忆提升世界模型效率Velox: Spatial Memory Makes World Models EfficientHugging Face Daily Papers
- Athanor:自主后训练研究的智能体原生系统Athanor: An Agent-Native System for Autonomous Post-Training ResearchHugging Face Daily Papers
- 在去中心化智能体集群中学习自我编排Learning Self-Orchestration in Decentralized Agent SwarmsHugging Face Daily Papers
- CarryOn:个性化可靠机器人辅助的双循环执行框架CarryOn: A Dual-Loop Harness for Personalized and Reliable Robot AssistanceHugging Face Daily Papers
- CoDA:数据引擎与动作策略的协同演化CoDA: Co-Evolution of Data Engines and Action PoliciesHugging Face Daily Papers
- SUPRA-TEX:具有高度对齐纹理的三维生成SUPRA-TEX: 3D Generation with Super-Aligned TextureHugging Face Daily Papers
- 以权重作为记忆:将流式视频的长期历史内化Memory as Weights: Internalizing Long-Term History for Streaming VideosHugging Face Daily Papers
- LEASH:电脑使用智能体中感知不确定性的弱到强升级与监督LEASH: Uncertainty-Aware Weak-to-Strong Escalation and Oversight for Computer-Use AgentsHugging Face Daily Papers
- Gen4P:视频生成器也是四维感知器Gen4P: Video Generators are 4D PerceiversHugging Face Daily Papers
- MMSearch-OPD:为长任务多模态搜索智能体增加难度并使其可学MMSearch-OPD: Easy Made Hard, Hard Made Learnable for Long-Horizon Multimodal Search AgentsHugging Face Daily Papers
- 挖掘、采样与渲染:生成模型的测试时多样性对齐Mine, Sample, Render: Test-Time Diversity Alignment Of Generative ModelsHugging Face Daily Papers
- 一组Gemini 4 Argon智能体通关《我的世界》A team of Gemini 4 Argon agents beats MinecraftHugging Face Daily Papers
- PathFM:多变量轨迹的快速生成式预测PathFM: Fast Generative Forecasting of Multivariate TrajectoriesHugging Face Daily Papers
- 解释混合专家大语言模型中各专家对个性化的作用Explaining the Roles of Experts in Mixture-of-Experts LLMs for PersonalizationHugging Face Daily Papers
- 统一视觉语言模型中的技能解耦同策略专家蒸馏Skill-Decoupled On-Policy Expert Distillation for Unified Vision-Language ModelsHugging Face Daily Papers
- TimeRSI:迈向时间序列基础模型的自我改进式智能体部署TimeRSI: Towards Self-Improved Agentic Deployment for Time-Series Foundation ModelsHugging Face Daily Papers
- 当KV缓存比GPU更慢:为上下文并行预填充扩展远程KV层When the KV Cache Is Slower Than the GPU: Scaling Remote KV Tiers for Context-Parallel PrefillHugging Face Daily Papers
- CoWorld Reasoner:由演进视频世界模型执行框架驱动的智能体多模态推理CoWorld Reasoner: Agentic Multimodal Reasoning with the Evolving Video World Model-Driven HarnessHugging Face Daily Papers
- EVA-Cap:以事件为中心的对齐优化视听视频描述EVA-Cap: Optimizing Audiovisual Video Captioning via Event-Centric AlignmentHugging Face Daily Papers
- 连续体注意力Continuum AttentionHugging Face Daily Papers
- 共享符号而非向量:语言模型之间无需训练的事实通信Share the Symbol, Not the Vector: Training-Free Fact Communication Across Language ModelsHugging Face Daily Papers
- PN-Mix:基于成对范数的混合精度键缓存量化,提升长输出推理效率PN-Mix: Pair-Norm Guided Mixed-Precision Key Cache Quantization for Efficient Long-Output LLM InferenceHugging Face Daily Papers
- τ-Emotion-Bench:以真实用户情绪揭示LLM智能体的交互脆弱性τ-Emotion-Bench: Exposing the Interaction Fragility of LLM Agents through Real-World User EmotionsHugging Face Daily Papers
- 诊断并纠正可验证指令遵循中的奖励投机Diagnosing and Correcting Reward Hacking in Verifiable Instruction FollowingHugging Face Daily Papers
- 观察、维护与学习:多模态智能体记忆的演进Seeing, Maintaining, and Learning: The Evolution of Multimodal Agent MemoryHugging Face Daily Papers
- 任意时间与空间:以世界状态为中心的智能体视频生成框架Any Time, Any Space: A World-State-Centric Framework for Agentic Video GenerationHugging Face Daily Papers
- Player-1:从合成经验中学习游戏Player-1: Learning to Play from Synthetic ExperienceHugging Face Daily Papers
- Doc2Harness:评测LLM对编码智能体的长程运行框架工程能力Doc2Harness: Benchmarking LLMs on Long-Horizon Harness Engineering of Coding AgentsHugging Face Daily Papers
- 更多方法带来更好智能体?检验智能体强化学习中的自蒸馏More Methods, Better Agents? A Reality Check on Self-Distillation in Agentic Reinforcement LearningHugging Face Daily Papers
- 通过批评引导的在策略蒸馏恢复可靠监督Recovering Reliable Supervision via Critique-Guided On-Policy DistillationHugging Face Daily Papers
- SwarmRSI:通过演进科学智能体群实现递归自我改进SwarmRSI: Recursive Self-Improvement through Evolving Scientific Agent SwarmHugging Face Daily Papers
- LatentLoop:以非对称算力分配扩展循环MoE模型LatentLoop: Scaling Looped MoE Models with Asymmetric Compute AllocationHugging Face Daily Papers
- Cortex:将上下文整合为参数记忆Cortex: Consolidating Context into Parametric MemoryHugging Face Daily Papers
- ActionOPD:蒸馏决策而非词元ActionOPD: Distilling Decisions, Not TokensHugging Face Daily Papers
- 迈向智能体机器人:模型、数据、环境与运行框架Towards Agentic Robots: Model, Data, Environment, and HarnessHugging Face Daily Papers
- RuleTrek:评测复杂场景中的智能体规则推理RuleTrek: Benchmarking Agentic Rule Reasoning in Complex ScenariosHugging Face Daily Papers
- 高动态相机条件下世界空间人体恢复的诊断研究A Diagnostic Study of World-Space Human Recovery Under High Camera DynamicsHugging Face Daily Papers
- 长程强化学习需要MuonLong-Horizon RL Needs MuonHugging Face Daily Papers
- SafeDriveVLA:以导航条件世界模型想象实现冲突感知的端到端自动驾驶SafeDriveVLA: Navigation-Conditioned World Model Dreaming for Conflict-Aware End-to-End Autonomous DrivingHugging Face Daily Papers
- 自条件机制支持流匹配中的深层序列推理Self-Conditioning Enables Deep Sequential Reasoning in Flow MatchingHugging Face Daily Papers
- 什么让社交智能体交互有效?从配对世界中的可验证结果学习What Makes a Social Agent's Interactions Effective? Learning from Verifiable Outcomes in Paired WorldsHugging Face Daily Papers
- 从失败到可学习世界:为导航智能体生成策略条件环境From Failures to Learnable Worlds: Policy-Conditioned Environment Generation for Navigation AgentsHugging Face Daily Papers
- Glance Forcing:让双向扩散模型快速转为自回归Glance Forcing: Making Bidirectional Diffusion Autoregressive at a GlanceHugging Face Daily Papers
- VITAL:从视觉运动到精确动作VITAL: From Visual Motion to Precise ActionHugging Face Daily Papers
- TextVerifier:超越识别,验证中文文字渲染TextVerifier: Seeing Beyond Recognition to Verify Chinese Text RenderingHugging Face Daily Papers
- The COUNT:统计N元组,实现主机内存扩展与无需训练的适配The COUNT: Counting N-grams Unlocks Scaling on Host Memory and Training-Free AdaptationHugging Face Daily Papers
- VASPO:利用特权价值估计进行自蒸馏VASPO: Self-Distillation with Privileged Value EstimationHugging Face Daily Papers
- 使用DAO与JING进行生成式世界模拟Generative World Simulation with DAO and JINGHugging Face Daily Papers
- 通过测试时训练压缩潜在表示Latent Compaction with Test-Time TrainingHugging Face Daily Papers
- AgentSafeEvo:通过对抗共同演化提升LLM智能体安全AgentSafeEvo: Adversarial Co-Evolution for LLM Agent SafetyHugging Face Daily Papers
- 物理时间:机器人策略记忆动态中缺失的维度Physical Time as a Missing Dimension of Robot Policy Memory DynamicsHugging Face Daily Papers
- 你的Transformer其实具有弹性Your Transformer Is Secretly ElasticHugging Face Daily Papers
- StreamErase:迈向超高速流式视频擦除StreamErase: Towards Ultra-Fast Streaming Video ErasingHugging Face Daily Papers
- G^3VLA:为视觉语言动作模型加入几何归纳偏置G^3VLA: Geometric Inductive Bias for Vision-Language-Action ModelsHugging Face Daily Papers
- CoTranslator:将逆向思维链转为正向,用于推理蒸馏CoTranslator: Reverse to Forward Chain-of-Thought Translator for Reasoning DistillationHugging Face Daily Papers
- Lluna:以协作单模态模型进行缺失模态推断Lluna: Collaborative Unimodal Models for Missing-Modality InferenceHugging Face Daily Papers
- VLX-Go:通过多阶段学习及在策略优化实现稳健具身跟踪VLX-Go: Robust Embodied Tracking through Multi-Stage Learning and On-Policy RefinementHugging Face Daily Papers
- RoTA-Bench:通过交互式投影对齐评测三维空间推理RoTA-Bench: Evaluating 3D Spatial Reasoning through Interactive Projection AlignmentHugging Face Daily Papers
- MAI-Gym:跨手机与电脑环境的GUI智能体评测及强化学习一体化开放平台MAI-Gym: An Open All-in-One Platform for GUI Agent Evaluation and RL Training across Mobile and Computer EnvironmentsHugging Face Daily Papers
- InfiniSplat v2:分析并改进任意视角的前馈三维高斯泼溅InfiniSplat v2: Analyzing and Improving Feed-Forward 3DGS from Any ViewsHugging Face Daily Papers
- CommerceGen-Studio:商业级主图生成的端到端框架CommerceGen-Studio: An End-to-End Framework for Commercial-Grade Main-Image GenerationHugging Face Daily Papers
- Intern Lumina U2:支持全视觉理解与图像生成的多码本扩散大语言模型Intern Lumina U2: A Multi-Codebook Diffusion Large Language Model for Omni-Visual Understanding and Image GenerationHugging Face Daily Papers
- DeltaStreamer:迈向类人智能体流式视频理解DeltaStreamer: Towards Human-Like Agentic Streaming Video UnderstandingHugging Face Daily Papers
- Orbia:生成视频能否构成一致的三维世界?Orbia: Do Generated Videos Form a Consistent 3D World?Hugging Face Daily Papers
- WorldCast:探索世界状态预测中的智能视觉生成WorldCast: Exploring Intelligent Visual Generation in World State PredictionHugging Face Daily Papers
- SERO:从可执行的预言机续接中学习长任务智能体强化学习SERO: Learning from Executable Oracle Continuations for Long-Horizon Agentic Reinforcement LearningHugging Face Daily Papers
- RepForcing:自回归视频蒸馏的表示空间初始化RepForcing: Representation-Space Initialization for Autoregressive Video DistillationHugging Face Daily Papers
- 相同字符错误率,不同损害:面向OCR的严重性加权字符错误率Equal CER, Unequal Damage: Severity-Weighted Character Error Rate for OCRHugging Face Daily Papers
- 学习读出:语言模型预训练中的反嵌入动态Learning to Read Out: Unembedding Dynamics in Language Model PretrainingHugging Face Daily Papers
- 一个偏差分数、两个因素:缓解措施如何影响多项选择评测的选项偏差One Bias Score, Two Factors: How Mitigations Affect Option Bias in Multiple-Choice EvaluationHugging Face Daily Papers
- VascuSR:用于CT超分辨率的迭代式解剖感知扩散VascuSR: Iterative Anatomy-Aware Diffusion for CT Super-ResolutionHugging Face Daily Papers
- 稀疏自编码器中数据的边际回报递减:功能行为不能保证特征可复现Diminishing Returns of Data in Sparse Autoencoders: Functional Behavior does not Guarantee Feature ReproducibilityHugging Face Daily Papers
- DecisionRobust:跨文本和视觉评测JEV式决策模型鲁棒性DecisionRobust: Benchmarking Robustness for JEV Style Decision Models Across Text and VisionHugging Face Daily Papers
- 看似完整不等于有据可依:对照人写态势报告评估LLM简报Looking Complete Is Not Being Grounded: Evaluating LLM Sitreps Against Human-Authored Situation ReportsHugging Face Daily Papers
- APT:反馈驱动的文生图智能体后训练APT: Feedback-Driven Agentic Post-Training for Text-to-Image GenerationHugging Face Daily Papers
- LensErase:用于任意擦除的基础模型LensErase: A Foundation Model for Erase AnythingHugging Face Daily Papers
- 以可验证信息奖励训练LLM多智能体协作Training LLMs for Multi-Agent Collaboration with Verifiable Information RewardsHugging Face Daily Papers
- Jeva:从基本原理构建系统一模型Jeva: System One Models from First PrinciplesHugging Face Daily Papers
- VideoJev:面向视频理解和交互的系统一模型VideoJev: System One Model for Video Understanding and InteractionHugging Face Daily Papers
- Agent-Reflex:教语言智能体反思行动Agent-Reflex: Teaching Language Agents to Act ReflectivelyHugging Face Daily Papers
- Urbex:城市级三维场景中的智能体空间定位Urbex: Agentic Spatial Grounding for City-Scale 3D ScenesHugging Face Daily Papers
- AesCode:以解耦跨模态奖励生成美观代码AesCode: Aesthetic Code Generation with Decoupled Cross-Modal RewardsHugging Face Daily Papers
- Tev:以密集进展奖励扩展长程智能体环境Tev: Scaling Long-Horizon Agentic Environments with Dense Progress RewardsHugging Face Daily Papers
- GreenWAM:面向机器人的世界基础模型GreenWAM: A World Foundation Model for RoboticsHugging Face Daily Papers
- WorkRover:通过环境递归演化扩展长程协作智能体训练WorkRover: Scaling Long-Horizon Cowork Agent Training Via Recursive Environment EvolutionHugging Face Daily Papers
- 学习如何记忆:GUI智能体的记忆模态选择Learning How to Remember: Memory Modality Selection for GUI AgentsHugging Face Daily Papers
- WAVE-0:以轻量潜在未来预测连接VLA策略和世界动作模型WAVE-0: Bridging VLA Policies and WAMs with Lightweight Latent Future PredictionHugging Face Daily Papers
- FlashVSR+:通过混合注意力与奖励分布匹配实现高效视频超分辨率FlashVSR+: Efficient Video Super-Resolution via Hybrid Attention and Reward-Yield Distribution MatchingHugging Face Daily Papers
- JevGPT:自主研究中的快慢思考JevGPT: Thinking Fast and Slow in Autonomous ResearchHugging Face Daily Papers
- Cepheid:解耦尺度与结构以实现跨类别运动补全Cepheid: Decoupling Scale from Structure for Cross-Family Motion CompletionHugging Face Daily Papers
- Fish Audio Drama 3技术报告Fish Audio Drama 3 Technical ReportHugging Face Daily Papers
- DeltaTTG:用剩余token进度计训练多轮智能体DeltaTTG: Training Multi-Turn Agents with a Token-to-Go Progress MeterHugging Face Daily Papers
- EgoExplorer:实境人类移动与操作的视觉运动触觉观测平台EgoExplorer: A Vision-Motion-Touch Observatory for Human Loco-Manipulation In SituHugging Face Daily Papers
- 简化并改进少步扩散模型的在策略蒸馏Simplifying and Improving On-Policy Distillation for Few-Step DiffusionHugging Face Daily Papers
- ReTerm:面向自改进终端智能体的恢复训练ReTerm: Recovery Training for Self-Improving Terminal AgentsHugging Face Daily Papers
- Open-o3-Video-v2:以可验证时空证据搜索、定位与推理Open-o3-Video-v2: Seek, Ground, and Reason with Verifiable Spatio-Temporal EvidenceHugging Face Daily Papers
- ACCRUE:机器人基础模型的持续离策略强化学习ACCRUE: Continual Off-Policy Reinforcement Learning for Robot Foundation ModelsHugging Face Daily Papers
- LifeSearch:真实生活与生产力深度研究智能体基准LifeSearch: Benchmarking Deep Research Agents for Real-World Life and ProductivityHugging Face Daily Papers
- SWE-Misbehavior:软件编码智能体越界行为的基准与缓解SWE-MISBEHAVIOR: Benchmarking and Mitigating Boundary Violations in Software Coding AgentsHugging Face Daily Papers
- WorldRecon:大规模三维世界的生成式多视角网格重建WorldRecon: Generative Multi-View Mesh Reconstruction for Large-Scale 3D WorldHugging Face Daily Papers
- 任务产生评分规则、证据产生判定:使用工具增强裁判的逐项奖励Rubrics from the Task, Verdicts from Evidence: Pointwise Rubric Rewards with Tool-Augmented JudgesHugging Face Daily Papers
- 读得对、做得错:诊断智能体循环中系统一模型的决策失误Reading Right, Acting Wrong: Diagnosing Decision Failures of System One Models in Agent LoopsHugging Face Daily Papers
- 超越最终判决:带可验证反馈的智能体代码评判Beyond the Verdict: Agentic Code Judging with Verifiable FeedbackHugging Face Daily Papers
- DoubtLess:无需训练的对比解码实现高效推理DoubtLess: Training-Free Contrastive Decoding for Efficient ReasoningHugging Face Daily Papers
- 4LC:动态世界建模中高效条件化的四维潜在缓存4LC: 4D Latent Cache for Efficient Conditioning in Dynamic World ModelingHugging Face Daily Papers
- 通过生成式轨迹规划进行基于模型的在线决策Model-Based Online Decision Making via Generative Trajectory PlanningHugging Face Daily Papers
- 以近端KL正则化实现离策略RLVR中的逐token重要性采样Enabling Token-wise Importance Sampling in Off-Policy RLVR with Proximal KL RegularizationHugging Face Daily Papers
- 智能体能学会预测自身失败吗:以结果校准置信度Can an Agent Learn to Predict Its Own Failure? Outcome-Grounded Confidence Calibration in LLM AgentsHugging Face Daily Papers
- SHELDON:在假设和观测空间进行双重搜索推理SHELDON: Reasoning through Dual Search over Hypothesis and Observation SpacesHugging Face Daily Papers
- TacitVLA:无需推理延迟的视觉语言动作模型推理TacitVLA: Reasoning Vision-Language-Action Models Without Reasoning LatencyHugging Face Daily Papers
- 裁判追随参考答案:验证工具调用参数时的推导鸿沟Judges Follow the Reference: The Derivation Gap in Verifying Tool-Call ArgumentsHugging Face Daily Papers
- 让头脑构想、让双手实现:遵循多模态指令完成长程机器人操作Let the Mind Conceive, Let the Hand Achieve: Following Multimodal Instructions for Long-Horizon Robotic ManipulationHugging Face Daily Papers
- 多个正确答案、一个训练样本:去除LLM推荐系统GRPO中的回答者噪声Many Right Answers, One Training Example: Removing Responder Noise from GRPO for LLM RecommendersHugging Face Daily Papers
- CLASHarness:通过可控可验证的生成流程评测计算机使用中的多智能体冲突CLASHarness: Benchmarking Multi-Agent Conflict in Computer Use via a Controllable and Verifiable Generation PipelineHugging Face Daily Papers
- 混合推理智能体Hybrid Reasoning AgentsHugging Face Daily Papers
- GroundGUI:以高质量数据及蒸馏到强化学习实现GUI定位GroundGUI: GUI Grounding by High-Quality Data and Distillation-to-Reinforcement LearningHugging Face Daily Papers
- 联合子空间稀疏注意力:高效长上下文解码Joint Subspace Sparse Attention for Efficient Long-Context DecodingHugging Face Daily Papers
- 通过像素空间分组图像生成高效优化场景文字Efficient Scene-Text Refinement via Pixel-Space Grouped Image GenerationHugging Face Daily Papers
- 测试是否过多?大语言模型高效评估中的基准压缩综述Are We Testing Too Much? A Survey of Benchmark Compression for Efficient Evaluation of Large Language ModelsHugging Face Daily Papers
- 重新思考多模态大语言模型的投影器训练Rethinking Projector Training in Multimodal LLMsHugging Face Daily Papers
- 通过寄存器采取行动Acting Through RegistersHugging Face Daily Papers
- PreciseCoder:超越单元测试奖励的忠实且可泛化调试PreciseCoder: Faithful and Generalizable Debugging Beyond Unit-Test RewardsHugging Face Daily Papers
- Diff Diffusion:以扩散模型预测编辑路径,实现高效代码编辑Diff Diffusion: Predicting Edit Paths with Diffusion Models for Efficient Code EditingHugging Face Daily Papers
- 扩展流匹配中的在线最优传输小批次Scaling Online Optimal Transport Minibatches for Flow MatchingHugging Face Daily Papers
- MA-Bench:衡量语言模型的行动规划能力MA-Bench: Measuring Operational Planning in Language ModelsHugging Face Daily Papers
- 为战略性智能体交互选择LLMChoosing LLMs for Strategic Agent InteractionsHugging Face Daily Papers
- 对齐还是不对齐:训练前检查COMPASSTo Align or Not To Align: Check Your COMPASS Before You TrainHugging Face Daily Papers
- 通过保留注意力的MLP裁剪实现推理语言模型结构化剪枝Unlocking Structured Pruning for Reasoning Language Models via Attention-Preserving MLP TrimmingHugging Face Daily Papers
- OmniCompass:全模态视频定位与理解的综合基准OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and UnderstandingHugging Face Daily Papers
- 自我演进的智能体执行框架Self-Evolving Agent HarnessHugging Face Daily Papers
- EdgeWise:通过选择性云端委派将长任务智能体带到边缘设备EdgeWise: Bringing Long-Horizon Agents to the Edge via Selective Cloud DelegationHugging Face Daily Papers
- MFVSR:视频字幕移除的多风格细粒度基准MFVSR: A Multi-Style and Fine-Grained Benchmark for Video Subtitle RemovalHugging Face Daily Papers
- 虚拟头注意力Virtual Head AttentionHugging Face Daily Papers
- FlashMask-3:高效且表达力强的掩码感知分布式注意力FlashMask-3: Efficient and Expressive Mask-Aware Distributed AttentionHugging Face Daily Papers
- Cowork-Gym:规模化合成交互工作环境Cowork-Gym: Synthesizing Interactive Work Environments at ScaleHugging Face Daily Papers
- WorldMemBench:视频世界模型记忆的综合基准WorldMemBench: A Comprehensive Benchmark for Memory of Video World ModelsHugging Face Daily Papers
- LeanPRM:基于Lean的过程监督,用于自然语言证明验证LeanPRM: Lean-Grounded Process Supervision for Natural Language Proof VerificationHugging Face Daily Papers
- 复制粘贴语言模型Copy-Paste Language ModelsHugging Face Daily Papers
- H-DDCM:用于快速生成式图像与视频压缩的Hadamard码本H-DDCM: Hadamard Codebooks for Fast Generative Image and Video CompressionHugging Face Daily Papers
- 智能体早已知道:内部状态预测并阻止未授权行动及后续谎言The Agent Knew: Internal States Forecast Unauthorized Actions and Stop Them, and the Lies That FollowHugging Face Daily Papers
- 图像分词器应在图像之间训练Image Tokenizers Should Be Trained Between ImagesHugging Face Daily Papers
- RAR:面向时间一致交互世界模型的速率自适应采样RAR: Rate-Adaptive Rollout for Time-Consistent Interactive World ModelsHugging Face Daily Papers
- 不止完美队友:与不可靠同事开展多智能体协作Beyond Perfect Teammates: Multi-Agent Collaboration with Unreliable ColleaguesHugging Face Daily Papers
- TPPO:面向语言模型推理的跟踪价值PPOTPPO: Tracking-Value PPO For Language Model ReasoningHugging Face Daily Papers
- X-Ragent:面向有证据依据胸片推理的工具增强强化学习X-Ragent: Tool-Augmented Reinforcement Learning for Evidence-Grounded Chest X-Ray ReasoningHugging Face Daily Papers
- Multics:采用分层通信的生成式多人世界模型Multics: Generative Multi-Player World Model with Hierarchical CommunicationHugging Face Daily Papers
- AReA:面向高分辨率扩散Transformer的自适应感受野注意力AReA: Adaptive Receptive-Field Attention for Efficient High-Resolution Diffusion TransformersHugging Face Daily Papers
- VEDJE:面向可扩展视频文本检索的高效视频判别式联合编码器VEDJE: Video-Efficient Discriminative Joint Encoder for Scalable Video-Text RetrievalHugging Face Daily Papers
- 带语言正则化的自蒸馏Self-Distillation with Verbal RegularizationHugging Face Daily Papers
- CineRecon:前沿模型电影场景重建能力基准CineRecon: Benchmarking Frontier Models for Agentic Cinematic ReconstructionHugging Face Daily Papers
- LongR:以密集效用奖励强化学习释放长上下文推理LongR: Unleashing Long-Context Reasoning via Reinforcement Learning with Dense Utility RewardsHugging Face Daily Papers
- 冰箱里的大象:长视频理解的恒定内存帧打包Elephant in the Fridge: Constant-Memory Frame Packing for Long Video UnderstandingHugging Face Daily Papers
- 面向可验证奖励强化学习的谱驱动自适应秩优化器Spectral-Driven Adaptive-Rank Optimizer for Large Language Model Reinforcement Learning with Verifiable RewardsHugging Face Daily Papers
- 全模态大语言模型知道的比说出的更多Omni-Modal LLMs Know More Than They SayHugging Face Daily Papers
- HexaBee-4B:结合多语问答和双向监督的六语翻译HexaBee-4B: Six-Language Translation with Multilingual Answering and Bidirectional SupervisionHugging Face Daily Papers
- MixLID:混合语言文本中的token级语言识别MixLID: Token-Level Language Identification in Mixed-Language TextHugging Face Daily Papers
- AUGUstus:统一生成与理解的自动驾驶VLA模型AUGUstus: Autonomous Driving VLA Model with Unified Generation and UnderstandingHugging Face Daily Papers
- EventLens:面向视频理解的事件结构强化学习EventLens: Event-Structure Reinforcement Learning for Video UnderstandingHugging Face Daily Papers
- StartLux-Decision:以强化学习训练AutoThink模型,校准审慎决策StartLux-Decision: AutoThink Decision Models with Reinforcement Learning for Calibrated Deliberative DecisionsHugging Face Daily Papers
- FreeSolo:支持零样本长度外推的线性时间稀疏注意力FreeSolo: Linear-time Sparse Attention with Zero-Shot Length ExtrapolationHugging Face Daily Papers
- LLM到流模型的蒸馏:教科学生成器理解自然语言目标LLM-to-Flow Distillation: Teaching Natural Language Goals to Scientific GeneratorsHugging Face Daily Papers
- ProbabilityBench:以精确真实不确定性评估并校准模型ProbabilityBench: Model Evaluation and Calibration with Exact Ground-Truth UncertaintyHugging Face Daily Papers
- 重新思考LLM引导的视觉问答:更深入的知识整合真有帮助吗?Rethinking LLM-Guided Visual Question Answering: Does Deeper Knowledge Integration Really Help?Hugging Face Daily Papers
- SyncReward:面向音视频生成的人类对齐同步奖励SyncReward: Human-Aligned Synchronization Reward for Audio-Visual GenerationHugging Face Daily Papers
- 何时左就是右?超越屏幕理解相机运动When Left Is Right? Camera-Motion Understanding beyond ScreenHugging Face Daily Papers
- InfoVis:统一幻灯片图像生成与编辑InfoVis: Unified Slide Image Generation and EditingHugging Face Daily Papers
- 超越策略展开:以锚定噪声在嵌入空间进行方向性探索,用于可验证奖励强化学习Beyond Policy Rollouts: Embedding-Space Directional Exploration with Anchored Noise for LLM RLVRHugging Face Daily Papers
- 实体、关系与事件抽取中的校准类型化决策Calibrated Typed Decisions for Entity, Relation, and Event ExtractionHugging Face Daily Papers
- 面向通用长任务智能体的递归环境扩展Recursive Environment Scaling for General Long-Horizon AgentsHugging Face Daily Papers
- PINT:面向机器人操作、共享预测意图的统一模型PINT: A Unified Model with Shared Predictive Intent for Robot ManipulationHugging Face Daily Papers
- 通过策略同步的语言动作双重自举实现导航的数据驱动递归改进Data-Centric Recursive Self-Improvement for Navigation via Policy-Synchronized Language-Action Dual BootstrappingHugging Face Daily Papers
- 一日十年:从面向人的软件演化合成智能体原生界面A Decade in a Day: Synthesizing Agent-Native Interfaces from Human-Oriented Software EvolutionHugging Face Daily Papers
- VIEWCRAFT:演化多模态智能体看见和阅读的内容VIEWCRAFT: Evolving What Multimodal Agents See and How They Read ItHugging Face Daily Papers
- CIViC:代理任务中音频条件视觉预测的证据门控CIViC: Evidence Gates for Audio-Conditioned Visual Prediction on Proxy TasksHugging Face Daily Papers
- UniForge3D:共享锚点引导空间中的联合三维重建与生成UniForge3D: Joint 3D Reconstruction and Generation in a Shared Anchor-Guided SpaceHugging Face Daily Papers
- Agent Sociodrama:以社会动态引导智能体团队角色重塑Agent Sociodrama: Social-Dynamics-Guided Persona Recasting for LLM Agent TeamsHugging Face Daily Papers
- RSIPO:技能与策略共同演化的递归自改进策略优化RSIPO: Recursive Self-Improving Policy Optimization for Agentic Reinforcement Learning via Skill-Policy Co-EvolutionHugging Face Daily Papers
- 扩展重复数据预训练:批量大小、正则化和GPU时间权衡Scaling Repeated-Data Pretraining: Batch Size, Regularization, and GPU-Time TradeoffsHugging Face Daily Papers
- Apertus 1.5:从开放文本模型到开放多模态推理模型Apertus 1.5: From Open Text Models to Open Multimodal ReasonersHugging Face Daily Papers
- SkillMarker:基于KGW水印验证智能体技能所有权SkillMarker: A KGW-Based Watermarking Method for Agent Skill Ownership VerificationHugging Face Daily Papers
- 探索具身操作中组合泛化的极限Exploring the Limits of Combinatorial Generalization in Embodied ManipulationHugging Face Daily Papers
- 历史重演:释放线性注意力的长上下文能力History Replays Itself: Unleashing Long-Context Capability in Linear AttentionHugging Face Daily Papers
- 智能体环境中可扩展的弱到强验证器强化Scalable Weak-to-Strong Verifier Hardening for Agent EnvironmentsHugging Face Daily Papers
- ANTE:先预判再执行——从已验证信念为编码智能体学习内部世界模型ANTE: Anticipate, Then Execute -- Learning Internal World Models for Coding Agents from Verified BeliefsHugging Face Daily Papers
- ExeWorldBench:基于代码的可执行三维世界建模基准ExeWorldBench: Benchmarking Code-Based Executable 3D World ModelingHugging Face Daily Papers
- Vela:以用户行为监督激发大语言模型智能Vela: User Behavior as Supervision Sparks LLM IntelligenceHugging Face Daily Papers
- Last Secure Code Benchmark:随意生成的代码安全吗Last Secure Code Benchmark: Is the Code You Vibe Secure?Hugging Face Daily Papers
- UltraBERT:突破BERT预训练限制UltraBERT: Beyond the Limit of BERT Pre-TrainingHugging Face Daily Papers
- EviScene:超越真假AI判定的智能体图像事实评估EviScene: Beyond "Is It AI?" for Agentic Image Factuality AssessmentHugging Face Daily Papers
- 通用多模态嵌入中什么真正有效:对比训练策略的受控研究What Really Works for Universal Multimodal Embeddings? A Controlled Study of Contrastive Training StrategiesHugging Face Daily Papers
- LightVAE:迈向紧凑高效的视频自编码器LightVAE: Towards Compact and Efficient Video AutoencodersHugging Face Daily Papers
- PANINI-DR:通过生成式语义工作区实现可控深度研究PANINI-DR: Controlled Deep Research through Generative Semantic WorkspacesHugging Face Daily Papers
- ShellElastic:兼具最大更新和超参数迁移的可扩展弹性预训练ShellElastic: Scalable Elastic Pretraining with Simultaneous Maximal Updates and Hyperparameter TransferHugging Face Daily Papers
- 可能但无效:把决策模型当合成调查受访者的心理测量审计Probable but Not Valid: A Psychometric Audit of Decision Models as Synthetic Survey RespondentsHugging Face Daily Papers
- 多机器人系统中该信任谁:通过神经符号交叉验证联合推断机器人与对象假设的可信度Who to Trust in Multi-Robot Systems: Joint Robot and Object Hypothesis Trust Inference via Neural-Symbolic Cross-ValidationHugging Face Daily Papers
- Words to World:交互式物理环境中的大语言模型基准Words to World: Benchmarking LLMs in Interactive Physical EnvironmentsHugging Face Daily Papers
- KV4-RL:通过INT4 KV缓存量化加速RLVR异步采样KV4-RL: Unlocking Faster Asynchronous Rollouts for RLVR with INT4 KV Cache QuantizationHugging Face Daily Papers
- ProgAgent:通过程序化训练学习工具编排ProgAgent: Learning Tool Orchestration through Programmatic TrainingHugging Face Daily Papers
- Drive3R:面向自动驾驶的分块流式几何基础模型Drive3R: Block-wise Streaming Geometric Foundation Models for Autonomous DrivingHugging Face Daily Papers
- 学习在哪里分割:音源分离中的动态频带划分Learning Where to Split: Dynamic Band Splitting for Audio Source SeparationHugging Face Daily Papers
- Harness^2:面向开放世界的递归智能体控制框架Harness^2: Recursive Agent Harnessing for an Open WorldHugging Face Daily Papers
- APEX:面向极限压缩的异步前缀去噪APEX: Asynchronous Prefix Denoising for Extreme CompressionHugging Face Daily Papers
- ClawLoop:基于控制框架原生训练循环的自改进智能体ClawLoop: Self-Improving Agents via Harness-Native Training LoopHugging Face Daily Papers
- GeoFix:面向四维一致生成的几何修复GeoFix: Geometry Fixing for 4D Consistent GenerationHugging Face Daily Papers
- 面向GUI智能体的潜在世界动作模型Latent World Action Model for GUI AgentsHugging Face Daily Papers
- 从任务配对的人类视频学习可迁移世界动作模型Learning Transferable World-Action Models from Task-Paired Human VideosHugging Face Daily Papers
- AutoEnvScaling:用终端智能体自动化数据飞轮AutoEnvScaling: Automating the Data Flywheel with Terminal AgentsHugging Face Daily Papers
- Web$hell:智能体访问网页的统一接口Web$hell: A Unified Interface for Agentic Web AccessHugging Face Daily Papers
- Video-Arena:可扩展的智能体视频时空能力基准Video-Arena: A Scalable Agentic Video Benchmark for Spatiotemporal CapabilityHugging Face Daily Papers
- 6G无线接入网中保护隐私的意图实现与保障Privacy-Preserving Intent Fulfilment and Assurance for 6G RANHugging Face Daily Papers
- 超越按领域路由的监督:多教师在策略蒸馏中的跨教师防护Beyond Domain-Routed Supervision: Cross-Teacher Guarding for Multi-Teacher On-Policy DistillationHugging Face Daily Papers
- LLM强化学习中的思考泄漏The Thinking Leakage in Reinforcement Learning for LLMsHugging Face Daily Papers
- APEX-Harm:评估共享专业工作空间中的智能体安全APEX-Harm: Evaluating Agent Safety in Shared Professional WorkspacesHugging Face Daily Papers
- BigCell:生成十亿像素全切片图像BigCell: Generating Gigapixel Whole-Slide ImagesHugging Face Daily Papers
- Jag:以直接边界框预测实现高效视觉定位Jag: Direct Box Prediction for Efficient Visual GroundingHugging Face Daily Papers
- 下一概念建模:快速并行的由粗到细文本生成Next-Concept Modeling: Fast, Parallel Coarse-to-Fine Text GenerationHugging Face Daily Papers
- Cortex:以测量结果把关编码智能体技能与规则的演进和剪枝Cortex: Measurement-Gated Evolution and Pruning of Coding-Agent Skills and RulesHugging Face Daily Papers
- SimVTA:基于仿真的视频与轨迹对齐,用于驾驶世界动作模型SimVTA: Simulation-Grounded Video-Trajectory Alignment for Driving World Action ModelsHugging Face Daily Papers
- 语音事件检测器:带时间戳识别非语言发声与背景声音,丰富语音转录Speech event detector: Timestamped detection of non-verbal vocalizations and background sounds for rich speech transcriptionHugging Face Daily Papers
- Pristine Models:原始模型Pristine ModelsHugging Face Daily Papers
- LeWAM:使用JEPA进行端到端世界与动作建模LeWAM: End-to-End World and Action Modeling with JEPAsHugging Face Daily Papers
- WorldCast:分布式多人世界模型WorldCast: Distributed Multiplayer World ModelsHugging Face Daily Papers
- 扩展长程工作智能体的可验证环境Scaling Verifiable Environments for Long-horizon Work AgentsHugging Face Daily Papers
- 在合适的潜在空间中做超分辨率:冻结视觉基础模型底座Super-Resolution in The Right Latent Space: A Frozen Vision-Foundation SubstrateHugging Face Daily Papers
- SWEeper-Bench:智能体能否发现交互软件中的错误?SWEeper-Bench: Can Agents Discover Bugs in Interactive Software?Hugging Face Daily Papers
- ForcingBooster:自适应分块去噪,加速自回归视频扩散ForcingBooster: Adaptive Patch-wise Denoising for Fast Autoregressive Video DiffusionHugging Face Daily Papers
- AI驱动、感知时序的功耗优化AI-Driven, Timing-Aware Power OptimizationHugging Face Daily Papers
- SALD:扩散模型的自参照优势学习SALD: Self-Referenced Advantage Learning for Diffusion ModelsHugging Face Daily Papers
- 三通道评分冲突中的受控获取与弃答Controlled Acquisition and Abstention in Three-Channel Score ConflictsHugging Face Daily Papers
- NavGPT-3:在分层导航运行环境中利用上下文NavGPT-3: Harnessing Context in a Hierarchical Navigation RuntimeHugging Face Daily Papers
- 更好的视觉表示总能带来更好的端到端自动驾驶吗?Do Better Visual Representations Always Lead to Better End-to-End Autonomous Driving?Hugging Face Daily Papers
- EMHO:通过经验轨迹优化具身智能体控制框架EMHO: EMbodied Agent Harness Optimization via Experience TracesHugging Face Daily Papers
- MimicX:为视频驱动的人形机器人动作跟踪进行策略在环监督优化MimicX: Policy-in-the-Loop Supervision Refinement for Video-Driven Humanoid Motion TrackingHugging Face Daily Papers
- PAIR:连接视觉语言动作模型中的感知与行动PAIR: Bridging Perception and Action in Vision-Language-Action ModelsHugging Face Daily Papers
- CLIFT:用于网页智能体训练与测试时扩展的保形自我验证CLIFT: Conformal Self-Verification for Web Agent Training and Test-Time ScalingHugging Face Daily Papers
- SimForcing:将仿真运动先验蒸馏到真实域机器人世界模型SimForcing: Distilling Simulation Motion Priors into Real-Domain Robot World ModelsHugging Face Daily Papers
- CURIO:好奇心驱动的测试时学习与开放探索CURIO: Curiosity-Driven Test-Time Learning for Open-Ended DiscoveryHugging Face Daily Papers
- Vela:以自适应动作曲线参数化扩展视觉语言动作模型Vela: Scaling Vision-Language-Action Models with Adaptive Action Curve ParametrizationHugging Face Daily Papers
- 无需解码的批量语音决策:单词元监督让冻结LLM听见转写之外的信息Batched Speech Decisions Without Decoding: Single-Token Supervision Lets a Frozen LLM Hear Beyond the TranscriptHugging Face Daily Papers
- HumanoidToolBench:从工具选择到移动执行评测人形机器人工具使用HumanoidToolBench: Benchmarking Humanoid Tool Use from Selection to Mobile ExecutionHugging Face Daily Papers
- 分数更高答案更差:以协议级评分规则缓解强化学习奖励投机Scoring Higher, Answering Worse: Mitigating Reward Hacking in Rubric-Based RL via Protocol-Level RubricsHugging Face Daily Papers
- 软掩码扩散编辑:可调强度的图像和视频像素级重做Diffusion Editing with Soft Mask: Pixel Level Redo of Image and Video with Adjustable StrengthHugging Face Daily Papers
- PhyProbe:重新思考生成视频的物理一致性评估PhyProbe: Rethinking Physical Consistency Evaluation in Generated VideosHugging Face Daily Papers
- 不止提示数量:数据怎样影响同策略蒸馏迁移Beyond Prompt Count: How Data Shapes Transfer in On-Policy DistillationHugging Face Daily Papers
- LabInstruct:评测实验室流程中的情境化教学视频生成LabInstruct: Benchmarking Situated Instructional Video Generation for Lab ProceduresHugging Face Daily Papers
- DoSA:点状稀疏注意力DoSA: Dots Sparse AttentionHugging Face Daily Papers
- PsyBer-Agent:推荐系统中心理学驱动的用户智能体PsyBer-Agent: A Psychology-Driven User Agent for Recommender SystemsHugging Face Daily Papers
开源项目与版本更新 · 144 条
- DeepSeek Harness预发布版跟进新模型与搜索工具DeepSeek Harness v0.2.1-rc.2DeepSeek
- 我用Opus 5.5制作并开源了一款Tesla Roadster模拟器I built a Tesla Roadster simulator with Opus 5.5 and open-sourced itHacker News
- Show HN:InBox Wall——面向AI智能体的本地优先开源邮件净化工具Show HN: InBox Wall – A local-first, open-source email sanitizer for AI agentsHacker News
- Show HN:为iPhone Duo设计iOS应用的智能体技能Show HN: Agent skill for designing iOS apps for iPhone DuoHacker News
- Codex正在吞噬我的Mac磁盘空间Codex is EATING my Mac's disk spaceHacker News
- Show HN:GoRoute——使用Go构建的超轻量大模型网关,内存小于15MBShow HN: GoRoute – Ultra-Lightweight LLM Gateway in Go (<15MB RAM)Hacker News
- Show HN:你会合并吗?把AI助手常写的24种漏洞做成修复练习Show HN: Would You Merge It? – 24 bugs AI assistants write, as fix-it katasHacker News
- Show HN:Baa——查找并释放AI智能体留在Mac上运行的资源Show HN: Baa – find and free what your AI agents left running on your MacHacker News
- Show HN:受Anthropic Scion启发,用Rust实现可扩展版本控制Show HN: Scalable source control in Rust, inspired by Anthropic's ScionHacker News
- Show HN:Claude S.与Claude O.创作的《忒修斯鼠》Show HN: Theseus the Mouse by Claude S. (and Claude O.)Hacker News
- Show HN:Vramkit——在MacBook上为大模型推理无损压缩显存Show HN: Vramkit – lossless VRAM compression for LLM inference, on a MacBookHacker News
- Show HN:RAE FAANG——用于通过编程智能体构建产品的目录与命令行工具Show HN: RAE FAANG – catalog and CLI for building products with coding agentsHacker News
- Show HN:Offscreen——给你的智能体一块独立Mac屏幕Show HN: Offscreen - Give your agents a Mac screen of their ownHacker News
- 使用CUDA的Embeddinggemma-2转写Embeddinggemma-2 Transcribe with CUDAHacker News
- 面向不同专业水平、有效使用大模型开展科学工作的指南Cookbook to use LLMs effectively for sciences, in various levels of expertiseHacker News
- 适用于VS Code、JetBrains、Visual Studio及NPM的AI工程熟练度扩展AI Engineering Fluency Extension for VS Code / JetBrains / Visual Studio, NPMHacker News
- Show HN:CiteLadder——开源AI搜索可见性追踪器Show HN: CiteLadder – open-source AI search visibility trackerHacker News
- Show HN:Agentfork——从任意步骤重放大模型智能体并测试修复Show HN: Agentfork – Replay an LLM agent from any step and test fixesHacker News
- deepseek-ai/deepseek-harness v0.2.1-rc.1版本发布deepseek-ai/deepseek-harness v0.2.1-rc.1DeepSeek
- Space:面向智能体代码执行的沙箱平台Space: Sandboxed Platform for Agentic Code ExecutionHacker News
- Lithos-metal:Qwen3.8-27B在40核M5 Max上达到每秒168个tokenLithos-metal: Qwen3.8-27B on 40-core M5 Max at 168 tokens/sHacker News
- Show HN:AI智能体遵循了你从未批准的指令Show HN: AI agent follows instructions you never approvedHacker News
- Jshook:用于运行时逆向工程的开源MCP服务器Jshook: Open-source MCP server for runtime reverse engineeringHacker News
- Show HN:Twain——类似Meld、面向AI时代的差异查看器Show HN: Twain, a diff viewer like Meld, but for the age of AIHacker News
- AI网关治理AI Gateway GovernanceHacker News
- AI服务指标AI Service IndicatorsHacker News
- ESP32 iPhone智能体ESP32 iPhone AgentHacker News
- Agent Cyber Sentinel:AI智能体的零信任运行时安全防护Agent Cyber Sentinel: Zero-trust runtime security shield for AI agentsHacker News
- Show HN:CtxWise——面向Claude和Codex的MCP模式裁剪工具与防火墙Show HN: CtxWise – MCP schema tree-shaker and firewall for Claude and CodexHacker News
- Emara:无需大模型,确定性地划分功能并判断死代码可达性Emara: Deterministic feature slicing and dead-code reachability without LLMsHacker News
- Dispatch:智能体时代的战略Git平台Dispatch: Strategic Git Platform for the Agentic EraHacker News
- Judgeman:带信任分数的编程智能体逐步评测Judgeman – step-level evals for coding agents, with a trust numberHacker News
- GPU加速的PostGISGPU-Accelerated PostGISHacker News
- Aspire:机器人智能体技能发现Aspire: Agentic /Skills Discovery for RoboticsHacker News
- Terry Davis花了十年的事,我用智能体几天就完成了What toke 10 years from terry Davis it toke me agents and daysHacker News
- Outlines:大语言模型的结构化输出Outlines – Structured Outputs for LLMsHacker News
- 我在任天堂DS上做了一个Claude Code终端I built a Claude Code terminal on the NDSHacker News
- HN展示:Weatherdeck,为AI编码智能体工作通道提供只读状态板Show HN: Weatherdeck – a read-only status board for AI coding agent lanesHacker News
- 如何判断智能体的记忆层是在帮忙还是添乱?How do you know your agent's memory layer is helping, not hurting?Hacker News
- 在越狱PS5上运行llama.cpp:原生运行任意GGUF模型Llama.cpp on a jailbroken PS5: run any GGUF model natively on the consoleHacker News
- Hillclimb:用Claude Code与Codex实现开源AlphaEvolve式搜索Hillclimb: Open-source AlphaEvolve-style search with Claude Code and CodexHacker News
- Renor:通过MCP共享的编码智能体Markdown记忆,采用AGPL许可Renor – Markdown memory for coding agents, shared over MCP (AGPL)Hacker News
- Agent Gateway测试版:有人想试用吗?Beta Agent Gateway – Anyone want to try it out?Hacker News
- HN展示:Termexo,让一个AI在Windows上操作其他编码智能体Show HN: Termexo – one AI operating your other coding agents, on WindowsHacker News
- HN展示:开源且以智能体为先的Profound替代品Show HN: We built an open source, agent-first, alternative to ProfoundHacker News
- HN展示:Busabase,记录AI智能体工作的开源系统Show HN: Busabase – an open-source system of record for AI agent workHacker News
- Codex:已有完全访问权限,清理操作仍莫名被策略阻止Codex: Unexplained cleanup `blocked by policy` despite effective Full AccessHacker News
- 在电子邮件与WhatsApp之间保留上下文的Python AI客服智能体AI Support Agent That Carries Context Between Email and WhatsApp (Python)Hacker News
- MeshXFS:由AI编写、基于XFS的集群文件系统MeshXFS: A clustered filesystem built on XFS, written by AIHacker News
- AI研究工具:文献发现、LaTeX与可复现性AI Research Tools – Literature Discovery, LaTeX and ReproducibilityHacker News
- HN展示:Capy-flash,在两张16GB消费级GPU上运行125B混合专家模型Show HN: Capy-flash – a 125B MoE model on two 16 GB consumer GPUsHacker News
- HN展示:Heldkeys,查看AI编码智能体可接触哪些凭据Show HN: Heldkeys, see which credentials your AI coding agent can reachHacker News
- HN展示:AegisKernel,AI智能体的零信任确定性安全运行时Show HN: AegisKernel – Zero-trust deterministic security runtime for AI agentsHacker News
- HN展示:Context-tax,在你输入第一个字前Claude Code已经发送了什么Show HN: Context-tax – what Claude Code sends before you type a wordHacker News
- HN展示:我制作的MCP让智能体首次接触金融数据就能推理Show HN: I made an MCP Agents can reason every finance data the first timeHacker News
- HN展示:Paveo,在执行前检查Claude Code、Codex与Cursor命令Show HN: Paveo – Check Claude Code, Codex and Cursor commands before they runHacker News
- DeepSeek Harness发布v0.2.1-alpha.5deepseek-ai/deepseek-harness v0.2.1-alpha.5DeepSeek
- HN展示:确定性AI智能体护栏Show HN: Deterministic AI Agents GuardrailHacker News
- 清理编码智能体遗留的进程Reap the processes your coding agents leave behindHacker News
- HN展示:由Claude Code端到端制作的音乐视频及完整制作仓库Show HN: A music video made end to end by Claude Code, full production repoHacker News
- LibreNT:AI劣质内容如何侵入Windows净室工程LibreNT – how AI slop infects Windows clean-room engineeringHacker News
- 在命令提示符中使用Haskell构建三维图形引擎3D Graphics Engine in Haskell Inside a Command PromptHacker News
- RP1:智能体人形机器人的开源全栈平台RP1: An Open Source Full-Stack Platform for Agentic HumanoidsHacker News
- Hopper HQ:为Claude智能体提供工位的办公楼Hopper HQ: An office building where Claude agents gets a deskHacker News
- Kern:AI生成代码的零信任验证网关Kern – a zero-trust verification gateway for AI-generated codeHacker News
- Sift:用本地大语言模型分析每个文件Sift – Analyse every file with local LLMsHacker News
- 智能体团队缺失的一环原来是共享组织记忆Shared organizational memory for agent teams turned out to be the missing pieceHacker News
- Sakana AI的智能体科学家The Agentic Scientist – SakanaAIHacker News
- MuscleMemory:AI智能体的开源程序性记忆MuscleMemory: Open-source procedural memory for AI agentsHacker News
- HN展示:Sjq,面向AI流式数据的JqShow HN: Sjq – Jq, but for AI StreamingHacker News
- Free-AI-tools:免费及低价AI工具精选列表Free-AI-tools – Curated list of free and low cost AI toolsHacker News
- Veda:首个智能业余操作系统Veda: First Intelligent Hoby OSHacker News
- HN展示:Eve Python智能体模板Show HN: Eve Python Agent TemplateHacker News
- HN展示:Endo,高阶执行框架Show HN: Endo – A Higher-Order HarnessHacker News
- AI编码工具:架构、本地LLM与SWE-BenchAI Coding Tools – Architecture, Local LLMs and SWE-BenchHacker News
- AI Maestro一周年:运营由AI智能体与人组成的公司AI Maestro, one year in: running a company of AI agents and humansHacker News
- Sunny:面向老年人、重视无障碍与隐私的桌面AI助手Sunny: Accessible, privacy-first desktop AI assistant for older adultsHacker News
- Show HN:Brod——快速部署AI应用的云服务Show HN: Brod – Quick cloud to deploy AI appsHacker News
- Show HN:ReliaMesh——AI智能体的开源可靠性监控Show HN: ReliaMesh – Open-source reliability monitoring for AI agentsHacker News
- SlopCodeBench 2026年10月:Opus 5.5、Astra整体获胜,xhigh被认为有害SlopCodeBench 10/2026 – Opus 5.5, Astra wins overall, xhigh considered harmfulHacker News
- Cicada:面向AI智能体的亚毫秒Go零信任网关与gVisor防火墙,Apache 2许可Cicada: Sub-ms Go Zero-Trust gateway and gVisor firewall for AI agents Apache 2Hacker News
- 开源项目 sgl-project/sglangSGLang – High-performance serving framework for LLM's and multimodal modelsHacker News
- Show HN:RDST——本地监测Postgres/MySQL查询回退的智能体Show HN: RDST – A local agent that watches Postgres/MySQL for query regressionsHacker News
- Show HN:Voice Standard Bench——以语音转写评测LLM的开放基准Show HN: Voice Standard Bench – Open benchmark of LLMs on speech transcriptsHacker News
- PCEI:结合BitNet 1.58B、Mamba与FastKAN的设备端具身AIPCEI – On-device embodied AI combining BitNet 1.58B, Mamba, and FastKANHacker News
- Show HN:精选AI工具目录Show HN: A curated directory of AI ToolsHacker News
- HN展示:BitRun在CPU运行1位大语言模型,标题称98T/s且无需编译Show HN: BitRun v0.2 – Run 8B ternary LLMs on CPU with live wattage telemetryHacker News
- 将iOS应用从代码送到App Store的Claude Code技能A Claude Code skill that takes an iOS app from code to the App StoreHacker News
- AI智能体使用的电子邮箱域名列表List of email domains used by AI agentsHacker News
- TR-Core:以纯Python构建的非神经符号AI架构TR-Core – A non-neural symbolic AI architecture built in pure PythonHacker News
- Show HN:BlastLoop——面向AI智能体的结构优先编码方法Show HN: BlastLoop, a structure-first coding method for AI agentsHacker News
- Mistral Timestral模型Mistral TimestralHacker News
- Show HN:CommonForge——让客户的AI智能体安全地在你的代码库工作Show HN: CommonForge – Let clients' AI agents work in your codebase safelyHacker News
- Show HN:从零训练GPT-2 124M,4张A40用2.8小时处理24.8亿词元Show HN: GPT-2 124M trained from scratch – 2.48B tokens, 2.8h on 4 A40sHacker News
- Deltabox:可回滚的智能体沙箱Deltabox: A rollback-able agent sandboxHacker News
- Show HN:Hazzel——支持撤销与自带密钥的极简终端编码智能体Show HN: Hazzel, a minimal terminal coding agent with undo and BYOKHacker News
- 穆斯林祈祷提醒MCP:基于天文计算的祈祷时间引擎Muslim Prayer Reminder MCP: Astronomical prayer calculation engineHacker News
- Show HN:Nexus——开源、本地优先的智能体消息路由Show HN: Nexus – open-source local-first message routing for agentsHacker News
- Show HN:用Tenuo防护OWASP FinbotOWASP's finbot Defended with Tenuo: 0 successful attacksHacker News
- Show HN:Imagor MCP——将智能体编写的图像布局编译成一个URLShow HN: Imagor MCP – agent-written image layouts that compile to one URLHacker News
- deepseek-ai/deepseek-harness v0.2.1-alpha.4版本发布deepseek-ai/deepseek-harness v0.2.1-alpha.4DeepSeek
- Show HN:Fantasy Manager——无服务器、AI增强的梦幻足球管理智能体Show HN: Fantasy Manager – Serverless AI-Enhanced Fantasy Football Manager AgentHacker News
- Show HN:Pi subagents Lite——保持系统提示清洁Show HN: Pi subagents Lite – keep your system prompt cleanHacker News
- Show HN:用于消除规格幻觉的多人设库Show HN: Multi persona library to eradicate Spec HallucinationHacker News
- Run:ai的开源Helm图表Open-sourced helm chart for run:AIHacker News
- Show HN:RepoGuards——阻止向错误仓库推送的Claude Code钩子Show HN: RepoGuards – Claude Code hooks that refuse pushes into the wrong repoHacker News
- Show HN:server-noonien——无需服务器、跨机器无冲突的智能体记忆Show HN: server-noonien – conflict-free agent memory across machines, no serverHacker News
- Show HN:官方MCP服务器Show HN: Official MCP ServersHacker News
- Run:ai现已开源Run: AI Now Open SourceHacker News
- Show HN:AI编码智能体可复现的故障演练Show HN: Reproducible failure drills for AI coding agentsHacker News
- WSP MCP:连接AI编码智能体的免费WordPress MCPWSP MCP – Free WordPress MCP for Connecting AI Coding AgentsHacker News
- LLM-gauze修复LLM错误LLM-gauze fixes LLM errorsHacker News
- Aidlc:为AI编码智能体自适应引导工作流的规则Aidlc: Adaptive workflow steering rules for AI coding agentsHacker News
- Show HN:MoneySwitch——AI智能体用USDC支付x402 API,自行托管Show HN: MoneySwitch – AI agents pay x402 APIs in USDC (self-hosted)Hacker News
- Show HN:如何使用LLMShow HN: How to Use LLMsHacker News
- HN展示:让Claude Code用图示页面而非大段文字回答Show HN: Tired of agents' walls of text, so I made them answer in HTMLHacker News
- Morph Map:可视化代码库、勾画重构并导出提示Morph Map – Visualise your codebase, sketch a refactor, and export the promptHacker News
- VSCode:精细控制Copilot变更VSCode, granular control over Copilot changesHacker News
- Reflex:将Claude Code和Codex对话编译成智能体技能的终端界面Reflex, a TUI that compiles Claude Code and Codex transcripts into agent skillsHacker News
- Veda:首个原生智能体业余操作系统VedaOS Demo Video: A Mind-Blowing Agentic Operating SystemHacker News
- 构建AI编码智能体的艺术:一本自行构建的书The Art of Building AI Coding Agents: The book that builds itselfHacker News
- Show HN:用一套方法回答B2B营销问题的Claude技能Show HN: A Claude skill that answers B2B marketing questions from one methodHacker News
- stepfun-ai/Step-5-Preview(GitHub 仓库)阶跃星辰 StepFun
- 开源项目 anthropics/claude-plugins-officialanthropics/claude-plugins-officialGitHub 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 affaan-m/ECCaffaan-m/ECCGitHub Trending(日榜 + Python 日榜)
- 开源项目 HKUDS/CLI-AnythingHKUDS/CLI-AnythingGitHub Trending(日榜 + Python 日榜)
- 开源项目:K-Dense-AI/scientific-agent-skillsK-Dense-AI/scientific-agent-skillsGitHub Trending(日榜 + Python 日榜)
- akitaonrails/ai-memory开源项目akitaonrails/ai-memoryGitHub Trending(日榜 + Python 日榜)
- 开源项目 android/skillsandroid/skillsGitHub Trending(日榜 + Python 日榜)
- stablyai/orca开源项目stablyai/orcaGitHub Trending(日榜 + Python 日榜)
- anthropics/claude-code项目anthropics/claude-codeGitHub Trending(日榜 + Python 日榜)
- 开源项目:debpalash/VoiceStudiodebpalash/VoiceStudioGitHub Trending(日榜 + Python 日榜)
- anomalyco/opencode开源项目anomalyco/opencodeGitHub Trending(日榜 + Python 日榜)
- 开源项目 ruvnet/rufloruvnet/rufloGitHub Trending(日榜 + Python 日榜)
- 开源项目 xerj-org/xerjxerj-org/xerjGitHub Trending(日榜 + Python 日榜)
- The-Swarm-Corporation/AutoHedge开源项目The-Swarm-Corporation/AutoHedgeGitHub Trending(日榜 + Python 日榜)
- bmad-code-org/BMAD-METHOD开源项目bmad-code-org/BMAD-METHODGitHub Trending(日榜 + Python 日榜)
- 开源项目 Crosstalk-Solutions/unifi-toolkitCrosstalk-Solutions/unifi-toolkitGitHub Trending(日榜 + Python 日榜)
- 开源项目 google/artemisgoogle/artemisGitHub Trending(日榜 + Python 日榜)
- yt-dlp/yt-dlp开源项目yt-dlp/yt-dlpGitHub Trending(日榜 + Python 日榜)
- 开源项目 lightseekorg/tokenspeedlightseekorg/tokenspeedGitHub Trending(日榜 + Python 日榜)
- 开源项目 python/cpythonpython/cpythonGitHub Trending(日榜 + Python 日榜)
- ScrapeCreators/social-media-research-skills开源项目ScrapeCreators/social-media-research-skillsGitHub Trending(日榜 + Python 日榜)