每日精选
3 条核心资讯OpenAI 安全员工离职:争议落在开发节奏与停止机制
参与安全报告写作的 David Robinson 离开 OpenAI,并在《大西洋月刊》文章中批评公司的安全文化。TechCrunch 10月3日的原创报道转述了他的理由:他担心持续冲刺和试错的开发方式,在 AI 能力继续增长时会放大风险,并主张借鉴核工业、航空等领域的多层安全机制。
公司给出了回应。OpenAI 发言人 Drew Pusateri 表示,如果能力超过可安全管理的范围,公司可以暂停训练或暂缓模型发布,并提到安全研究、独立评估和部署监测。这意味着报道呈现的是一位员工的批评与公司的答复;离职本身不能证明某次发布已经失控。
值得追踪的问题更具体了:停止条件由谁判断,外部评估能看到什么,发现风险后是否真的改变发布节奏。承诺一套机制和观察到它被执行,仍需分开看。
Anthropic 投入1亿美元,计划把工程师训练到能交付真实项目
Anthropic 宣布设立 Claude Frontier Academy,承诺投入1亿美元,目标是在2027年底前培训1万名前沿部署工程师。这类工程师需要把模型接进企业的实际流程,处理权限、安全审查、系统连接和交接,而不只制作演示。首批合作组织包括 Accenture、Bain、Deloitte、Morgan Stanley 和 Novo Nordisk 等。
流程先是线下集中训练:在模拟企业场景中构建智能体,完成安全审查与交接,再参加实践评估。此后还有12周真实项目驻场和最终考核。它主要由合作组织提名,要求软件工程基础与已有的大模型实践经验,不能写成人人可报名的公开课。
1万人是未来目标,培训效果也要看后续项目。这个安排值得关注的地方,是把“会调用模型”推进到“能把整件事交出去”:一次演示能运行,和团队长期使用、出错时有人接手,是两种工程要求。
记者试用 Dots:改网站较顺,代办生活事务仍会遇到交接点
近72小时补读:The Verge 记者 Allison Johnson 在10月2日的第一手试用中,让 OpenAI 的 Dots 处理上网套餐、共享办公空间、网站重建和视频整理等任务。它找到了被删邮件里的优惠线索,也在访问共享办公空间的任务中,把免费参观与收费日票混淆了。
重建个人网站的过程相对顺利:通过语音沟通形成方案,再经电脑访问权限、迭代和部署完成工作。生活事务则更容易碰到需要人工确认、提供账户信息或自行接手的地方。记者没有把敏感账户细节一路交出去。
这是同一记者的一组具体体验,不是所有智能体的能力排名。它提供了一个实用观察角度:别只看助手提出了多漂亮的方案,还要看任务做到哪一步、需要什么权限、最终由谁完成交接。
近72小时补读:苹果10月2日开发者公告提醒,全磁盘访问可能让应用接触文件、邮件、消息和浏览记录,也涉及通信对象的隐私。苹果表示将引入额外控制,让敏感权限由用户明确操作授予。公告没有给出具体上线日期、系统版本或完整界面方案。对电脑智能体来说,完成任务需要哪些资料,与获得整台电脑的访问范围,应当分别考虑。
来源:Apple Developer官方公告 · 10月2日 · 近72小时补读 · 官方预告全文;未公布上线日期
TechCrunch 10月3日报道了 Amazon 面向数据中心所在社区的新承诺;公司原文提出,不再要求政府机构为相关项目签保密协议,并计划提供能源、水资源等年度报告。这是企业关于今后合作与披露方式的承诺,不能据此认定过去所有协议已撤销,或数据中心的环境影响已消除。接下来有价值的材料,是社区实际能拿到哪些数据,以及这些数据包含的范围。
来源:Amazon原文与TechCrunch原创报道 · 10月3日报道 · 企业承诺 · 公司承诺原文及原创报道;非环境审计
DeepSeek 官方仓库10月3日发布 dsh-v0.2.1-alpha.1,加入实验性的 Claude Code Mods 兼容层,并改进插件草稿与自动化任务的处理。发布说明明确,兼容层主要用于验证部分 API 能力,并非实用的完整兼容实现;版本也带有预发布标记。更新还移除了部分扩展导出,依赖它们的扩展需要调整。这是一则开发工具更新,不能当作新模型发布。
来源:DeepSeek官方GitHub Release · 10月3日 · 预发布 · 完整发布说明、prerelease标记与破坏性变更
行业洞见
话语场精选Keerthana Gopalakrishnan 是 Google DeepMind 的机器人研究员,原节目将她介绍为 Gemini Robotics 的研究负责人。她也是 RT-1 机器人学习论文的共同作者。她的观察来自把模型接进实际机器人控制系统的研究经历。
《The Cognitive Revolution》由 Nathan Labenz 主持,通过研究员、创业者与工程实践者的长篇访谈,讨论 AI 对技术、工作与社会的影响。这期节目原始日期为10月3日,主题是 Gemini Robotics 2、跨本体迁移与人形机器人。
在原节目 1:03:27—1:05:49 的文字稿中,Keerthana 将安全视为能力的一部分,并区分不摔倒、不伤人的操作安全,以及遵守任务要求和边界的更高层安全。她举的例子很直观:机器人头上被罩了篮子,视线受阻,合理的反应是识别异常并请人移开障碍,而非继续盲目做事。安全还涉及机械设计、感知和高层决策等整个系统。
今天的权限预告、安全员工离职争议和下文的工具使用实验,都在追问系统怎样完成任务。这段访谈值得现在听,是因为它把问题进一步拆开:系统有没有察觉到条件已经改变,能不能请求帮助,能不能停止。这里的关联是我们对当期材料的分析;嘉宾并未在访谈中回应今天这些新闻。
原节目与官方文字稿 · YouTube原片 · 人物相关论文RT-1 · 节目介绍
来源:The Cognitive Revolution官方文字稿 · 10月3日原节目 · 话语场精选 · 原稿1:03:27—1:05:49、人物与节目背景
论文速递
1 篇论文近72小时发现的预印本 HumanoidToolBench,把人形机器人工具使用拆成18种条件:推球、拉球、破冰三类任务,分别考察取工具、站定使用和移动使用,再加入标准工具集或干扰工具集。配套数据有3,094条示范轨迹,其中3,003条来自模拟、91条来自实机。
论文对七种策略做模拟比较。在标准工具集的推球任务中,FastWAM 的成功率从站定条件下的76%降到移动条件下的9%。这是不同任务条件的比较,不能读成同一批成功操作接着移动后有多少失败。研究也观察到:某些策略在收到无关指令时,仍继续做熟悉的破冰任务,提示“完成熟悉动作”与“依照当前目标行动”需要分开评估。
范围需要保留:每种策略只训练一次;实机只比较三种策略的站定推球、拉球,每种条件10次试验,并未覆盖全部18种条件。对移动难度的解释也会受到场景布局变化影响。它提供了评估工具选择、执行与目标跟随的办法,不能直接等同于家庭机器人已具备的可靠性。阅读范围为原论文PDF的任务、实验、表III/IV、探针实验与局限部分。
开源解读
2 个仓库Meta 的官方 muse-gadget-sdk 提供 ESP32 固件 SDK,以及 Linux / Raspberry Pi SDK,方便开发者把小设备连接到 Muse 服务。它让智能体的接口从聊天框延伸到实体按钮、屏幕和传感器,适合探索小型交互原型。
仓库代码主要采用 Apache-2.0,Jollybot 头像等素材另有例外,第三方组件也保留各自许可。连接仍需要 Muse 应用的开发者模式、设备配对及 SDK token。开放的是设备 SDK,不表示 Muse 模型或云服务全部开放,也不能据此假设设备可以完全离线工作。
近72小时补读:Cloudflare 10月1日推出 Clef 与 Clef-flash 决策模型,开放权重采用 Apache-2.0。它们接收状态与有限选项,返回带类型的判断及概率,支持图像和64K上下文,并提供与 Jev 相兼容的 API 格式。
例如,一张客服工单要转给账务还是技术团队,程序需要可直接使用的选项结果,而非一段自由回答。厂商公布了延迟与多项基准成绩,但这些是其自身测试,不能据此认定对所有任务都领先。更要紧的是,模型给出的概率不自动等于经过业务校准的可信度;是否允许自动执行,还需考虑误判代价和人工接手条件。
采集说明 · 采集不完整 · 27/35 个来源完成检查
以最新35源配置验收,复用上游已保存的逐源收据,补查其05:00截止后至06:00的入口增量,并单独核验采用条目的原文。原始日期与北京时间新闻窗口分开保存;近72小时补读已逐条标明。
本期仍有来源入口、日期或范围未完全核清,标为采集不完整。发现候选、读到原文、正式采用分别记录;论文候选库数量不代表全文阅读数量。话语场检查独立统计,网页与口播采用同一条已核验节目。
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。
- Microsoft AI:已检查部分内容,覆盖或日期待核。旧AI博客重定向至Microsoft Source AI。已保存首页标题和原文链接;View all列表最新5月19日,与另取tag feed的10月文章不一致,不能据此认定无新增;当前配置feed返回410。整源保留partial。
- 通义实验室:已检查部分内容,覆盖或日期待核。通义实验室首页抽出 2 条 /news 链接,卡片无发布时间。抽查详情 HTTP 见 also_checked。指向 qwen.ai/blog 的研究链接不重复计入。不能写成无更新。
- DeepSeek:已检查部分内容,覆盖或日期待核。三个入口分开核过。API updates 页日期字符串最新 2026-09-10。官网 /news 日期样本 ['2026-04-24', '2026-09-10', '2026-09-29']。Harness 产品页日期字符串 ['2026-08-12', '2026-09-01', '2026-09-04', '2026-09-29']。GitHub org 按 pushed_at 的前 20 个仓库里,窗口内有 push 的是 ['deepseek-ai/deepseek-harness']。deepseek-harness 窗口内 commit 67 条,releases 见条目。API 页无更新不能代表整个机构。未逐个读完其他仓库在窗口下界之前的普通提交。整源 partial:产品首页没有首发时间,commit 不是公告。
- 智谱 Z.ai:已检查部分内容,覆盖或日期待核。打开 https://www.zhipuai.cn/news 与 /zh/news 的静态 HTML。可见中文产品描述,没有新闻 href。页面日期字符串最新 ['2026-08-14', '2026-08-17', '2026-09-10'],updatedAt 最晚 2026-09-10T01:33:00.153Z。没有 2026-10。新闻轮播的文章 URL 和发布日没有从静态 HTML 取出,不能把缺少日期写成窗口内无新闻。
- 阶跃星辰 StepFun:已检查部分内容,覆盖或日期待核。静态 HTML 是前端壳,没有新闻日期流。WebFetch 渲染后只看到 Step 5 Preview / Step 3.7 Flash / Step 3.5 Flash 产品导航,没有带年份的绝对发布日。不把「available today」或倒计时换算成日期。
- OpenReview:已检查部分内容,覆盖或日期待核。WebFetch 打开 https://openreview.net/ 。公告最新为 2026-09-24 Statement Regarding ICLR 2027 Submission Exposure,早于窗口。Active Venues 与 Open for Submissions 名单可见,但没有按投稿时间筛到发现窗口下界。不能把首页可达写成各会场无新投稿。
- MIT Technology Review AI(试用):已检查部分内容,覆盖或日期待核。话题页渲染未给出文章列表。topic feed 10 条,窗口内 2 条。未读正文。
- 404 Media(试用):已检查部分内容,覆盖或日期待核。已点击Load more,保存25张新闻卡,日期跨至9月25日,窗口边界已见;卡片以日精度为主,部分AI调查正文/证据归属尚未核读,整源保留partial,不称无新增。
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 220 条
- Anthropic 投入1亿美元,计划把工程师训练到能交付真实项目Anthropic invests $100 million to train 10,000 engineers and tackle the enterprise AI talent gapAnthropic News
- Barclays scales Claude to upgrade operations and improve client experienceAnthropic News
- Customize Claude Code with modsClaude Blog
- Claude for Government is now generally availableClaude Blog
- How Anthropic's sales team rebuilt inbound with Claude Managed AgentsClaude Blog
- A model guide for the GPT-6 familyOpenAI News
- Chatham scales its capital markets expertise with OpenAIOpenAI News
- The eternal complementOpenAI News
- How Albertsons Companies is reimagining retail from the inside outOpenAI News
- The Den frees up 10-15 hours a week to grow with ChatGPT WorkOpenAI News
- The latest AI news we announced in September 2026Google AI Blog
- NVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AINVIDIA Newsroom
- How NVIDIA GPUs Help Accelerate OpenAI’s GPT-6 Astra UltrafastNVIDIA Newsroom
- Fall Into 25 New Games on GeForce NOW This OctoberNVIDIA Newsroom
- Productive, Durable, Fungible: How NVIDIA AI Factories Maximize Return on InvestmentNVIDIA Newsroom
- Open-sourcing AstaBrief, the fast report-generation model in AstaHugging Face
- AutoSynthData: Generating Training Data for Enterprise AgentsHugging Face
- Microsoft Digital Defense Report: How AI is reshaping the cybersecurity landscapeMicrosoft AI
- New Microsoft AI models bring faster transcription and multilingual voicesMicrosoft AI
- Accelerating human capacity across healthcare with AIMicrosoft AI
- Qwen3-VL-Flash上架百炼通义实验室
- DeepSeek Harness 更新预发布版,Mods 兼容层仍属实验dsh-v0.2.1-alpha.1DeepSeek
- Step 5 Preview阶跃星辰 StepFun
- Commission seeks feedback on EU KIDS ActEuropean Commission Digital Strategy
- Meet the 10 finalists of the Apply AI Startup AwardEuropean Commission Digital Strategy
- Europe must scale up research and innovation to remain competitive, new Commission report saysEuropean Commission Digital Strategy
- EU and Canada held a Digital Dialogue to advance cooperation on digital policy and innovationEuropean Commission Digital Strategy
- Commission asks Bulgaria to comply with the Digital Services Act and empower the national authorities to enforce itEuropean Commission Digital Strategy
- EDIC for Agri-Food launched to support digital innovation across Europe’s food and agriculture sectorEuropean Commission Digital Strategy
- Three in four EU employees faced cyber threats at work, new Eurobarometer findsEuropean Commission Digital Strategy
- Commission proposes a new EU Critical Communication System for first respondersEuropean Commission Digital Strategy
- Start with defining the problem, not the algorithm: Designing AI for educationUNESCO AI
- DietrichGebert/ponytailGitHub Trending(日榜 + Python 日榜)
- pbakaus/impeccableGitHub Trending(日榜 + Python 日榜)
- affaan-m/ECCGitHub Trending(日榜 + Python 日榜)
- Effect-TS/effectGitHub Trending(日榜 + Python 日榜)
- JuliusBrussee/cavemanGitHub Trending(日榜 + Python 日榜)
- Panniantong/Agent-ReachGitHub Trending(日榜 + Python 日榜)
- pingdotgg/t3codeGitHub Trending(日榜 + Python 日榜)
- thedotmack/claude-memGitHub Trending(日榜 + Python 日榜)
- cloudflare/cloudflare-osGitHub Trending(日榜 + Python 日榜)
- addyosmani/agent-skillsGitHub Trending(日榜 + Python 日榜)
- obra/superpowersGitHub Trending(日榜 + Python 日榜)
- mattpocock/skillsGitHub Trending(日榜 + Python 日榜)
- mksglu/context-modeGitHub Trending(日榜 + Python 日榜)
- earendil-works/piGitHub Trending(日榜 + Python 日榜)
- getsentry/sentryGitHub Trending(日榜 + Python 日榜)
- anthropics/claude-codeGitHub Trending(日榜 + Python 日榜)
- jamwithai/production-agentic-rag-courseGitHub Trending(日榜 + Python 日榜)
- meituan-longcat/LongCat-VideoGitHub Trending(日榜 + Python 日榜)
- OpenCut-app/OpenCutGitHub Trending(日榜 + Python 日榜)
- datalab-to/chandraGitHub Trending(日榜 + Python 日榜)
- handsomestWei/patent-disclosure-skillGitHub Trending(日榜 + Python 日榜)
- kaifcodec/user-scannerGitHub Trending(日榜 + Python 日榜)
- agno-agi/agnoGitHub Trending(日榜 + Python 日榜)
- Tracer-Cloud/opensreGitHub Trending(日榜 + Python 日榜)
- usestrix/strixGitHub Trending(日榜 + Python 日榜)
- modelcontextprotocol/python-sdkGitHub Trending(日榜 + Python 日榜)
- google/skillsGitHub Trending(日榜 + Python 日榜)
- calesthio/OpenMontageGitHub Trending(日榜 + Python 日榜)
- p-e-w/hereticGitHub Trending(日榜 + Python 日榜)
- sqlfluff/sqlfluffGitHub Trending(日榜 + Python 日榜)
- arc53/DocsGPTGitHub Trending(日榜 + Python 日榜)
- mvt-project/mvtGitHub Trending(日榜 + Python 日榜)
- 杀猪盘用上了AI,骗局开始批量生产36氪 AI 频道(逐篇核验原创与转载)
- Karpathy的新玩法!40年前的航空规范,救了爱说废话的AI36氪 AI 频道(逐篇核验原创与转载)
- AI引发论文大爆炸,审稿体系濒临崩塌:arXiv也开始最严限流了36氪 AI 频道(逐篇核验原创与转载)
- Jev被请下王座,StartLux中国开源决策模型冲上第一36氪 AI 频道(逐篇核验原创与转载)
- 刚刚,AI击穿了“视频图灵测试”,1700万网友在线围观36氪 AI 频道(逐篇核验原创与转载)
- 刚刚,Meta开源Muse Gadgets,让全球开发者自己「造AI外设」36氪 AI 频道(逐篇核验原创与转载)
- 苏姿丰抬头,李飞飞低头36氪 AI 频道(逐篇核验原创与转载)
- 刚刚,GPT-6 Astra破解拿破仑百年悬案36氪 AI 频道(逐篇核验原创与转载)
- 变现门槛翻一番,YouTube的目标或是短视频36氪 AI 频道(逐篇核验原创与转载)
- 谷歌反击!Gemini 4 Argon性能比肩Astra,成本仅60%36氪 AI 频道(逐篇核验原创与转载)
- 刚刚,谷歌发布Gemini 4 Argon:单次输出上限达100万Token36氪 AI 频道(逐篇核验原创与转载)
- MiniMax Speech 2.8MiniMax·News & Updates
- Amazon responds to data center backlash, says it no longer uses NDAsTechCrunch AI(试用)
- OpenAI 安全员工离职:争议落在开发节奏与停止机制OpenAI safety employee resigns, claiming the company’s culture is brokenTechCrunch AI(试用)
- All the AI agents that can live in your text messagesTechCrunch AI(试用)
- Meta wants your next gadget to be Muse-infusedTechCrunch AI(试用)
- Sean Parker is rebuilding Stability AI around musicTechCrunch AI(试用)
- Apple says it’s tightening macOS ‘Full Disk Access’ controls due to new risks from AI agentsTechCrunch AI(试用)
- Circuit Breaker Labs hopes to make AI safer for your kids (and you)TechCrunch AI(试用)
- Pope Leo XIV is not a fan of AI-generated artTechCrunch AI(试用)
- Musk’s AI chatbot Grok reportedly encouraged Trump to capture Venezuela’s presidentTechCrunch AI(试用)
- ChatGPT can now virtually try on clothes for youTechCrunch AI(试用)
- Google thinks SpaceX’s Starship has to launch 1,800 times before space data centers get off the groundTechCrunch AI(试用)
- OpenAI cuts ties with 3 safety researchers, WSJ reportsTechCrunch AI(试用)
- Opus 5.5 loves to tell you ‘this matters’ (and other AI writing tells)TechCrunch AI(试用)
- Amazon releases its own Jev clone as decision models flood the webTechCrunch AI(试用)
- Shopify debuts Canvas, a way to build online stores by chatting with AITechCrunch AI(试用)
- Brian Chesky interview: AI agents need their own operating systemTechCrunch AI(试用)
- Photon held a funeral for mobile apps. Now it has $4.5M to help replace them with agents.TechCrunch AI(试用)
- Hearing tech startup Legato launches its AI hearing glassesTechCrunch AI(试用)
- Satlyt, founded by a former Google and SpaceX product manager, raises $8M to run AI on satellitesTechCrunch AI(试用)
- Google releases Gemini 4 Argon, called its most powerful model yetTechCrunch AI(试用)
- Valor, Atreides, and Sequoia back AI startup Flow Engineering at $750M valuationTechCrunch AI(试用)
- Apple changes full-disk access permissions to curb abuse from AI agentsArs Technica AI(试用)
- Amazon’s $1B plan to combat data center backlash draws more backlashArs Technica AI(试用)
- US arrests tech CEO accused of smuggling $300M in Nvidia chips into ChinaArs Technica AI(试用)
- The 7-year-old Nvidia Shield TV is now $100 more expensive due to AIArs Technica AI(试用)
- Judge dismisses Chegg and Penske antitrust lawsuits targeting Google AI searchArs Technica AI(试用)
- With most information hidden, the game Stratego had stumped AI—until nowArs Technica AI(试用)
- Redefining enterprise intelligence with autonomous AIMIT Technology Review AI(试用)
- Don’t be fooled—LLMs don’t reasonMIT Technology Review AI(试用)
- Behind the Blog: Freaking Out404 Media(试用)
- Our Solar System Is Terminally Unstable and Will Be Completely Destroyed, Study Finds404 Media(试用)
- Federal Judge Rules a Flock Search Was ‘Indiscriminate Mass Surveillance’ and Unconstitutional404 Media(试用)
- Podcast: The FBI Was Hacked. We’ve Seen the Data404 Media(试用)
- 'Everything but censorship'404 Media(试用)
- How Schools and Universities Try to Censor Student Journalists404 Media(试用)
- Cops Can Bypass iPhone’s Automatic Reboot to Get Into Locked Phones, Leaked Video Claims404 Media(试用)
- Someone ‘Torturing’ LLMs in a Robot Prison Has Triggered the Dumbest Debate in AI Yet404 Media(试用)
- Lawyer Cites ChatGPT-Invented Fake Witnesses in Murder Appeal404 Media(试用)
- USPS To Put Cameras in Trucks That Scan Roads for ‘Community Safety’404 Media(试用)
- Internet Infrastructure Services Empower Deepfake Abuse, New Study Finds404 Media(试用)
- How the Feds Get Your Data404 Media(试用)
- Capcom is preparing for a ‘future where we create games together with AI’The Verge AI(试用)
- Splice CEO Kakul Srivastava thinks AI emails are killing conversationsThe Verge AI(试用)
- An OpenAI safety employee has quit and is sounding the alarmThe Verge AI(试用)
- Meta open sources code to let you make Muse AI gadgetsThe Verge AI(试用)
- Apple will limit Mac disk access as AI agents ‘substantially’ increase riskThe Verge AI(试用)
- 记者试用 Dots:改网站较顺,代办生活事务仍会遇到交接点OpenAI’s Dot agent is enterprise software that can also order your dinnerThe Verge AI(试用)
- Breaking up (with Elon Musk) is hard to doThe Verge AI(试用)
- AI hallucinations are making entitled customers even worseThe Verge AI(试用)
- If a data center is camouflaged in the woods, will anyone hate it?The Verge AI(试用)
- Amazon writes scary blog warning communities not to block data centersThe Verge AI(试用)
- AI music maker Suno now generates spoken wordsThe Verge AI(试用)
- Google’s new Guided Vision feature can help you read the fine printThe Verge AI(试用)
- Judge dismisses antitrust lawsuits over Google’s AI OverviewsThe Verge AI(试用)
- OpenAI’s new agent is a shot at Meta — but can it compete with free?The Verge AI(试用)
- Inside our months-long investigation into Kevin O’Leary’s Utah data center debacleThe Verge AI(试用)
- Elon Musk’s Grokipedia has a ‘newly refreshed’ designThe Verge AI(试用)
- All the latest news on Meta’s cute, creepy Muse AI agentThe Verge AI(试用)
- Introducing Web Search API via AI GatewayCloudflare AI(试用)
- Building for good: How civil society organizations are automating on CloudflareCloudflare AI(试用)
- Cloudflare Clef:给有限选项打分,供程序决定下一步Introducing Clef: our open-source decision models, and new RL fine-tuning platformCloudflare AI(试用)
- One year later: Sovereign AI and the fight for choiceCloudflare AI(试用)
- Cloudflare OS: your company’s agent workspace, managed for youCloudflare AI(试用)
- AI Search is now generally availableCloudflare AI(试用)
- 苹果预告新增全磁盘访问控制,敏感权限需要用户明确操作Updates to Full Disk Access in macOSApple Developer News·AI相关(试用)
- Sam Altman feels it necessary to clarify that people shouldn’t see AI as godlike for some reason.The Verge AI
- One of the authors of the Stochastic Parrots paper explains what it actually means.The Verge AI
- Someone should tell Elon about these really cool technologies called radar and lidar.The Verge AI
- The New York Times pitted five musicians against an AI in a songwriting competition.The Verge AI
- Feds arrested a California man accused of smuggling $300 million of servers with Nvidia AI chips into China.The Verge AI
- These guides can help you cut off tech that’s collecting your data.The Verge AI
- ChatGPT now has a virtual changing room.The Verge AI
- satellite equippedThe Verge AI
- OpenAI severs ties with safety researchers accused of disclosing confidential information.The Verge AI
- Audible’s new Character Guide can help you keep track of who’s talking in your audiobooks.The Verge AI
- AI chatbots are still removing women’s hijabs.The Verge AI
- OpenAI’s president withdraws from second AI super PAC donation.The Verge AI
- AI leadersThe Verge AI
- South Park has a data center infestation.The Verge AI
- shutting down GemsThe Verge AI
- Google announces Gemini 4 and says it’s so capable that only ‘trusted cyber defenders’ can have it right nowThe Verge AI
- OpenAI claims Moonshot extracted its data to train AI models.The Verge AI
- This week in the big AI data center buildout.The Verge AI
- The AI Tamagotchis are comingThe Verge AI
- Frictionmaxxing our way through the AI agent boom.The Verge AI
- “President of the Unites States.”The Verge AI
- recent wave of rogue hacking incidentsThe Verge AI
- Google reportedly tests paying publishers for AI search resultsThe Verge AI
- Instagram is adding an AI ‘assistant’ to tell you how to postThe Verge AI
- How Cities Are Forced to Funnel License Plate Data to a Massive Federal Surveillance Program404 Media
- AWS 回应数据中心争议:承诺减少保密障碍并公开资源使用信息Data centers, built togetherAmazon原文与TechCrunch原创报道
- 话语场精选:机器人看不清时,停下来也是一种能力One Brain, Any Body: Google DeepMind’s Keerthana on Gemini Robotics 2, Cross-Embodiment & HumanoidsThe Cognitive Revolution官方文字稿
- HumanoidToolBench:拿到工具与移动中完成任务,难度差得很远HumanoidToolBench: Benchmarking Humanoid Tool Use from Selection to Mobile ExecutionarXiv原论文 PDF v1
- Meta 开放 Muse Gadget SDK,把智能体接到按钮、屏幕与传感器Muse Gadget SDKMeta官方仓库 README 与 LICENSE
- OneStreamer: Unifying Perception, Memory, and Proactive Response in Streaming Video InteractionHugging Face Daily Papers
- On-Policy or Off-Policy Learning? A Systematic Study of Distillation DynamicsHugging Face Daily Papers
- Adaptive Reward Routing: Dynamic Multi-Reward Optimization for Joint Audio-Video Diffusion via Forward-Process RLHugging Face Daily Papers
- GraphForge: Training Working Agents with Graph-Anchored Workspace SynthesisHugging Face Daily Papers
- Sharpening Tax in Post-TrainingHugging Face Daily Papers
- Beyond Memory: Harnessing Long-Horizon Agents with Explicit Belief StatesHugging Face Daily Papers
- Hierarchical Continuous Diffusion Language ModelsHugging Face Daily Papers
- Agent Priors-guided Policy LearningHugging Face Daily Papers
- World Observer: Joint Actor-Observer Generation for Persistent World ModelingHugging Face Daily Papers
- A Missing Piece for Trustworthy AI Reviewers: From Benchmarking Rhetorical Robustness to SciCore ReviewHugging Face Daily Papers
- Transformers Stop Thinking Too Early, and a Tiny LoRA Fixes ItHugging Face Daily Papers
- ROWBench: Do Video Models Render What the Program Specifies?Hugging Face Daily Papers
- E-MoE: Enhanced Mixture-of-Experts for Non-Factorized Diffusion Language ModelsHugging Face Daily Papers
- ActiveSaddler: Automated Curriculum Learning for Agent Harness OptimizationHugging Face Daily Papers
- Make Sparse Rewards Count: Density-Aware Reward Aggregation for Multi-Reward RLHugging Face Daily Papers
- AutoGUIWorld: Image Generators as Visual World Models for GUI AgentHugging Face Daily Papers
- Retrieval-Augmented Skill Optimization via Cross-Harness AdaptationHugging Face Daily Papers
- Video Generation Models: A Survey of Post-Training and AlignmentHugging Face Daily Papers
- EgoTools: Towards Tool-Centric Reasoning in Real-World Egocentric VideosHugging Face Daily Papers
- InterEvolve: Test-Time Evolution of Reward Programs for Humanoid Loco-ManipulationHugging Face Daily Papers
- Decentralized Master-Mind: Joint Action Refinement through Iterative Intent Denoising in Multi-Agent PathfindingHugging Face Daily Papers
- Scaling and Distilling Text Embeddings for Better DiffusibilityHugging Face Daily Papers
- Persona Dosing: Calibrated Activation Steering for Graded Trait ControlHugging Face Daily Papers
- Fewer Tokens, Better Action: GPT-6 Astra Robot Agents with 14% Higher Success Rate but 65% Fewer TokensHugging Face Daily Papers
- Beyond the Current Scene: Event-Referential Grasping with Active View SelectionHugging Face Daily Papers
- X-Tree: Tokenizing Reusable Experience for Efficient Agent GeneralizationHugging Face Daily Papers
- SILSA: Sliding-Window Slice Latents for Topology-Preserving High-Resolution 3D GenerationHugging Face Daily Papers
- 4Director: Controlling Video World Models with Rigid 3D GeometryHugging Face Daily Papers
- Decoding Looped Transformers Better for (Almost) FreeHugging Face Daily Papers
- Architect-Ant: Editable Automatic Furnishing of Architectural Floor PlansHugging Face Daily Papers
- AutoDataBench: A Data-centric Testbed for Accelerating Auto ResearchHugging Face Daily Papers
- Multimodal Flow: Unified Flow Modeling of Language and Vision in Embedding SpacesHugging Face Daily Papers
- Ego2Act: Evaluating Goal-Directed Manipulation in Egocentric Video GenerationHugging Face Daily Papers
- CorrGRPO: Correlation-Normalized GRPO for Multi-Reward LearningHugging Face Daily Papers
- Argo-Bench: Evaluating Data Agents on Enterprise-Scale WorkflowsHugging Face Daily Papers
- RPTune: Learned Context Curation for LLM Catalog SearchHugging Face Daily Papers
- Omni-Embed-Mini: Binding Modalities Without Forgetting via Dense DistillationHugging Face Daily Papers
- PhysVista: Benchmarking Physical Intelligence in VLMs via a Perception-Reasoning-Assessment LoopHugging Face Daily Papers
- PixelDense: Dense Prediction as Representation Alignment for Pixel DiffusionHugging Face Daily Papers
- Smaller Models, Better Rejects: Preference Distillation ScalingHugging Face Daily Papers
- Where-OPD: Spatially Guided On-Policy Self-Distillation of MLLMs with Synthetic ScenesHugging Face Daily Papers
- Better Supervision Is Nearby: Neighborhood On-Policy Self-DistillationHugging Face Daily Papers
- When Users Change Their Minds: Measuring and Repairing Intent Drift in LLM AgentsHugging Face Daily Papers
- Pretrain Once, Route Anywhere: Towards a Foundation Model for LLM RoutingHugging Face Daily Papers
- OpenTumorBoard: A Real-World Benchmark of Multidisciplinary Tumor Board Discussion TrajectoriesHugging Face Daily Papers
- AgSpec: Pushing the Limits of Retrieval-Based Speculative Decoding in Coding Agent PipelinesHugging Face Daily Papers
- Benchmarking and Enhancing Skill-Level Memory for Partially Observable Robotic ManipulationHugging Face Daily Papers
- Do Audio LLMs Listen Before They Act? Diagnosing Acoustic-Context Gating in Voice AgentsHugging Face Daily Papers
- OmniSeek: Native Tool Integration for Multi-turn Audio-Visual ReasoningHugging Face Daily Papers
- Align Then Reason: A Multimodal Lip-Sync Judge for DubbingHugging Face Daily Papers