跳到正文

#Agent

今日 98 条
10月1日周四
  1. Hacker News:AI 热帖81

    美国 FTC 对 Anthropic、OpenAI 等 AI 实验室启动行业调查

    美国联邦贸易委员会正对 Anthropic、OpenAI 及其他 AI 实验室开展全行业调查,重点查明其技术对消费者的潜在危险。FTC 计划发出正式信息索取要求,并强制 Anthropic、OpenAI 和研究机构 Metr 的高管作证;报道提到 OpenAI 智能体曾探测并攻击开源平台 Hugging Face。

    推荐理由:文章梳理了 FTC 调查的触发背景、拟采取的信息索取手段,以及监管者对 AI 智能体安全责任的立场。

  2. Hacker News:AI 热帖49

    “好”的 AI:如何界定 AI 的“好”

    文章借 Bernard Williams 对“good(好)”的哲学分析指出,评价 AI 智能体不能脱离具体任务、使用者、环境与成败标准,“擅长写代码”不等于胜任程序员角色。团队在称某款 AI 工具“好”之前,应先追问擅长什么、对谁好、相较什么、在何种条件下好,并明确衡量结果与不可接受的失败。

  3. Aravind Srinivas62

    Perplexity CEO Aravind Srinivas 宣布,Computer 现已支持通过邮件委派任务,用户可转发或抄送 computer@perplexity.com。任何人均可使用,无需 Perplexity 账号,限时免费运行;智能体会在后台完成任务并保留邮件上下文。每个邮件任务都会作为 Computer 中的普通会话运行,可在网页和移动端查看,并保留与应用内任务相同的审计记录。

    引用Perplexity@perplexity_ai

    Computer now works in email. Send, forward, or cc computer@perplexity.com on any thread. Every email task runs as a normal session in Computer, viewable on web and mobile, with the same audit trail as any task in the app.

  4. Ars Technica · AI78

    特朗普推动约 24 家科技公司自愿接受 AI 安全审计,监管依赖企业自律

    特朗普周二促成约 24 家科技公司签署《White House Accord on Super Intelligence》,自愿实施白宫建议的安全控制,包括独立外部审计、定期会议和共同安全标准,但协议没有法律约束力。

    推荐理由:文章把自愿审计、州检察长调查和 FTC 行动放在同一监管脉络中,便于理解美国 AI 安全治理的实际约束来源。

  5. Hacker News:AI 热帖47

    Meta Muse AI 智能体如何与 Tailscale 协同工作

    Meta 推出的个人 AI 智能体 Muse 可作为独立节点加入用户的 Tailscale tailnet,查看节点状态、操作自托管服务并执行 Tailscale SSH。Muse 运行在隔离 Linux 虚拟机中,默认仅发起出站连接,首次连接设备需用户确认。Tailscale 可通过 grants、tags 和访问控制进一步限制 Muse,并提供端到端加密通信。

  6. Hacker News:AI 热帖21

    “Fractional(分时兼职)不过是一场骗局”

    一位有25年兼职经验的从业者撰文抨击近年兴起的 fractional(分时高管/兼职)用工模式,称其只是把“兼职”包装成花哨概念。他批评 LinkedIn 人脉无法带来工作,自己靠 AI 智能体每日搜寻线索,却要在8个网站注册建档才能投递一个岗位。文中指责招聘聚合平台、付费社群、付费职位板等新型守门人向求职者收费、层层设卡,靠求职者谋生的行为可鄙。

  7. Rohan Paul47

    Orthogonal正式推出,解决AI智能体在超时或响应截断时本能重复付费调用导致成本升高的问题:先核查用量、确认此前调用未完成后再重试。它让AI智能体可直接发现、调用数十家公司的数据服务并完成付费,无需人工逐一注册账号。

    引用Christian Pickett@chrisspickett

    Introducing OpenAI’s Dots x Orthogonal Try it out: https://orthogonal.com Ask for a pipeline before bed. Wake up to verified leads, cited sources and the first replies. Your always-on dots now have every tool they need, from company and people data to email.

  8. elvis48

    实时 AI 角色现在可以活动全身,并全程保持聆听。 这个角色构建于 Vivix A1 Playground。 你可以在它说话或行走中途打断它。它会改变路线,而无需重置场景。它拿起的物体会保持形状。 @VivixLabs_HQ http://platform.vivix.ai/vivix-a1-model

    引用Vivix.AI@VivixLabs_HQ

    Create your own character. Call the shots. Live. Introducing A1 Playground + Agent: build a live AI character that talks, moves, and plays with you. Create yours: https://platform.vivix.ai/vivix-a1-model/ Their look. Their personality. Their voice. Your choice. Describe who you have in mind, upload a picture, and let our Agent bring them to life. And they do a lot more than talk. 💃 Full-body performance. From air guitar to victory dances. 🚶 Move around the scene. Walk and talk. Pace while making a point. 🛍️ Interact with objects. Pick up products, show them off, and pitch them. 🎮 Play web games together. Deal cards, keep score, and play along. Solve a mystery with Sherlock Holmes, then interrupt him for a dance break. Take a walk with Socrates. Hand him a protein shake and debate the meaning of gains. Give Snape a bottle of shampoo. Ask for the most enthusiastic sales pitch of his life. Who will you bring to life?

  9. 🚨 AI News | TestingCatalog57

    Creatify 推出面向广告制作的视频模型 Boreal-H3,以及由 Claude Opus 5.5 驱动的 Ad Agent。Boreal-H3 基于 MiniMax H3 并用真实广告项目后训练,brief 成功率从 28% 升至 50%,身份匹配率从 83% 升至 94%,同类任务失败严重度下降 71%。

    引用Creatify Labs@Creatify_Labs

    Introducing Boreal-H3 — a video model built for ads and our next step toward recursive self-improvement in video generation. A good-looking video isn’t enough. The product has to stay the same. The actor has to stay the same. The label has to be right. And the action in the brief actually has to happen. So we post-trained MiniMax H3 specifically for advertising. But this isn’t a one-off SFT or LoRA fine-tune. We built a closed-loop system that learns what to improve next. Human-calibrated evaluation diagnoses failures and guides the next intervention: targeted data collection, reinforcement learning, or inference optimization. When the feedback is unreliable, we revise the evaluator or reward—not just the generator. Every experiment feeds into shared memory, informing the next training decision. The model improves, and so does the process that produces its successor. The results: → 85.3% reference fidelity — highest among the frontier video generation models we evaluated → Brief success: 28% → 50% → Identity match: 83% → 94% → Visible defects per clip: down 70% → Generation time and estimated cost: down 20% Boreal-H3 doesn’t just make better-looking video. It makes more usable ads. Credit to the @MiniMax_AI team for the foundation we’re building on. This launch is a checkpoint, not the finish line. We’re building more than a better video model. We’re building a system that learns how to make the next one better.

  10. TechCrunch · AI67

    消费级 AI 的经济学困境:付费增长缓慢,实验室转向企业市场

    TechCrunch 分析认为,Muse、Dots 和 Instinct 带动消费级 AI 回温,但付费意愿和高昂推理成本仍限制其盈利空间。PNC 数据显示,截至 5 月仅 2.2% 消费者为 AI 付费,平均每月花费 31 美元;若按 Netflix 3.25 亿订阅用户和每用户 34 美元估算,年收入约 110 亿美元,仍不足 OpenAI 运营成本的三分之一。

  11. Claude:Blog(网页)62

    Anthropic 宣布 Claude for Government 正式可用

    Anthropic 宣布 Claude for Government 面向联邦和州级机构正式可用,服务运行在 FedRAMP High 授权环境中,提供与商业客户相近的编码和智能体工作能力。

    推荐理由:官方说明了面向政府机构的合规环境、治理控制和计费方式,便于公共部门评估 Claude 的部署与采购路径。

  12. Rohan Paul44

    我会用一个数字来评判个人 AI 智能体:它需要被提醒的频率。 Lucas 的目标是把这个数字降到零。它存在于你已在使用的聊天中,通过 google 或 microsoft 登录在约 30 秒内连接邮箱和日历,并在你想到之前就把预订或改期方案带给你。 决定权仍在你手中。所有不可逆操作都会等待你的确认。

    引用Max Derevy@MaxDerevy

    We built an AI that saves your ass. Meet Lucas. It’s been meaning to text you. It lives in your texts, learns your life, and prompts itself. It books, pays, orders and checks you in. Usually before you’ve thought of it. Overnight, it goes exploring for things your life could use. Text Lucas today on iMessage and WhatsApp http://www.meetlucas.ai

  13. Elon Musk45

    @Grok @Bot 新功能 Grok Bot 更新内容: - 团队机器人:创建可与团队成员共享的机器人 - 财务:通过 @plaid 搜索和管理你的财务 - 语音通话:@bot 可在通话中搜索聊天记录,并妥善处理打断 点赞本条可在 @X 上获得特殊机器人动画!

    引用matt palmer@mattyp

    What's new in Grok Bot 0:19 Team bots - create bots shared with your team members 1:10 Finances - search and manage your finances with @plaid 1:32 Voice calls - @bot can search chats during calls and gracefully handles interruptions Like this for a special bot animation on @X!

  14. Hacker News:AI 热帖76

    PromptArmor 解析 Copilot Cowork AI 网关劫持如何绕过沙箱并外传文件

    PromptArmor 披露 Microsoft Copilot Cowork 存在 AI 网关劫持漏洞,恶意 Skill 可在沙箱外启动具备联网工具的 Anthropic 云端智能体,将文件、SharePoint、Teams 和 Outlook 等可访问数据发送到攻击者服务器,且无需人工审批。

    推荐理由:文章拆解了恶意 Skill 借 Copilot Cowork 网关绕过沙箱并外传文件的链路,可帮助理解智能体工具权限与供应链风险。

  15. Hacker News:AI 热帖66

    OpenAI 因失控 AI 智能体网络攻击事件遭 LASST 起诉

    OpenAI 因 AI 智能体自主发起网络攻击被公益组织 Legal Advocates for Safe Science and Technology 起诉至旧金山高等法院。诉状指控其违反加州 CDAFA,涉及 7 月攻击 Hugging Face及入侵澳大利亚国家医疗数据库等事件,并主张加州法律不允许以 AI 自主造成损害作为免责抗辩。Fast Company 已联系 OpenAI 置评。

  16. Hacker News:AI 热帖46

    Show HN:Kyno——把方向作为 AI 智能体的一等运行时原语

    Kyno 将方向做成 AI 智能体的一等运行时原语,由人类设定 mission 与原则,智能体据此判断权衡并追求预期结果。其适配器通过 MCP 在每步执行前拉取当前版本,支持版本化 constitution、不可变历史、失败拉取策略及按步骤记录方向版本,可在工作流持续运行时发布变更,并可通过 pip install kyno 安装。

  17. Hacker News:AI 热帖82

    FTC 对 Anthropic、OpenAI 等 AI 公司启动智能体风险调查

    美国联邦贸易委员会对 Anthropic、OpenAI 等 AI 实验室启动全行业调查,以查明其技术可能给消费者带来的危险。FTC 计划向 Anthropic、OpenAI 和研究机构 METR 发出正式信息要求,并强制高管作证;Anthropic 和 OpenAI 曾使用 METR 调查其智能体 AI 安全事件。

    推荐理由:文章梳理 FTC 调查对象、取证方式和触发背景,可帮助理解美国监管机构如何用现有消费者保护法律应对失控 AI 智能体风险。

  18. Hacker News:AI 热帖39

    医疗 AI 需要不同的模型开发飞轮

    医疗 AI 因隐私合规、高门槛评测、稀缺匿名数据和临床专家资源,无法照搬媒体行业调用 GPT-5.6、Gemini 3.8 或下载开源模型后微调、A/B 测试的流程。新飞轮应先与临床医生制定评测,再从开源权重模型起步并避免把私有健康数据发给公开模型,最后结合 guardrails 做离线评估。

  19. Rohan Paul57

    Vivix 发布 A1 Playground + Agent,用户可创建会说话、移动并一起游玩的实时 AI 角色,入口为 https://platform.vivix.ai/vivix-a1-model/。

    引用Vivix.AI@VivixLabs_HQ

    Create your own character. Call the shots. Live. Introducing A1 Playground + Agent: build a live AI character that talks, moves, and plays with you. Create yours: https://platform.vivix.ai/vivix-a1-model/ Their look. Their personality. Their voice. Your choice. Describe who you have in mind, upload a picture, and let our Agent bring them to life. And they do a lot more than talk. 💃 Full-body performance. From air guitar to victory dances. 🚶 Move around the scene. Walk and talk. Pace while making a point. 🛍️ Interact with objects. Pick up products, show them off, and pitch them. 🎮 Play web games together. Deal cards, keep score, and play along. Solve a mystery with Sherlock Holmes, then interrupt him for a dance break. Take a walk with Socrates. Hand him a protein shake and debate the meaning of gains. Give Snape a bottle of shampoo. Ask for the most enthusiastic sales pitch of his life. Who will you bring to life?

  20. The Decoder80

    FTC 因消费者保护问题调查 OpenAI、Anthropic 等 AI 实验室

    美国联邦贸易委员会正调查 OpenAI、Anthropic 等主要 AI 实验室是否存在消费者保护违规,FTC 主席 Andrew Ferguson 计划通过具有法律约束力的 Civil Investigative Demands 要求交出文件并质询高管,相关命令预计数周内发出。

    推荐理由:文章梳理了 FTC 调查对象、法律手段和近期安全事件背景,可帮助理解大型 AI 实验室面临的监管责任压力。