←搜索结果

按匹配度 · 共 1540 条结果 清除搜索

星期四

  1. Stanford HAI News

    Stanford HAI 探讨心理健康 AI 的治理难题

    Stanford HAI 召集研究人员、临床医生、政策制定者与患者权益倡导者,梳理心理健康 AI 治理的三类关键挑战:缺乏对“心理健康 AI”范围的共识、证据基础滞后、以及州与联邦立法碎片化。目前各州已提出 140 多项相关法案,联邦立法仍聚焦未成年人保护。

    ome meaningful safety improvements. However, long-term, meaningful change will require resolving deeper policy challenges and resolving tensions across today’s patchwork of state laws. Disparate state laws concerning therapy tools can be onerous for psychotherapists who are licensed in multiple states with different regulations. Perhaps the most underexamined and unresolved issue in this conversation is one of the most fundamental: Business models built around maximizing user engagement are stru

    匹配字段正文

  2. Stanford HAI News

    斯坦福研究:仅9%数据经纪商完全遵守加州隐私法

    斯坦福 RegLab 与 Stanford HAI 联合研究发现,在2025年自查注册的522家数据经纪商中,只有9%完全遵守加州隐私法的透明度要求,45%未向加州隐私保护局提交任何权利请求指标。研究还发现64%的经纪商通过混乱的网站设计、多重表单或过度验证为消费者提交请求设置障碍,而这类暗黑模式被法律明确禁止。加州2026年数据经纪商登记已要求披露是否向生成式AI开发者出售数据,目前有30多家公司作出披露。

    ss is selling. The problem? Most data brokers are ignoring those rules. Joint research from Stanford RegLab and Stanford HAI looked at compliance with California’s strict privacy laws and found that only 9% of these companies are fully complying with the law. “This is a surprisingly low compliance rate,” says Stanford HAI Privacy and Data Policy Fellow Jennifer King . “It’s possible data brokers are still working out how to report on the required metrics, but they’ve been on notice for more than

    匹配字段正文

  3. Ai2 Blog

    Ai2 详解海事 AI 智能体 Shippy 的构建经验

    Ai2 的 Skylight 团队分享了海事 AI 智能体 Shippy 的架构:由系统提示词构成的 soul、遵循 Claude Code 与 Codex 同款 agent-skills 规范的 skills,以及可切换 harness 与模型的 config 三部分组成,目前运行在 OpenClaw 框架和 Claude Opus 4.6 上。Shippy 通过专用 Skylight CLI 调用 API,避免直接构造调用带来的分页、几何编码等错误,并面向 70 多个国家的政府机构与非政府组织提供沙箱化隔离服务。

    artners like Global Fishing Watch or TMT. A single question posed to Shippy can hook into several skills at once. "Are there vessels operating near the Cordillera de Coiba MPA?" draws on the Skylight skill for data query, our partner ProtectedSeas’ database for MPA boundary context, and the vessel track skill for interpreting vessel behavior. All of this happens in a single dialogue turn. The soul defines what Shippy will and won't do. It won't make legal determinations about whether a vessel is

    匹配字段正文

  4. Ai2 Blog

    Ai2 推出 OlmoEarth Platform:行星尺度地理空间推理基础设施

    Ai2 发布 OlmoEarth Platform,用于将地理空间模型从微调、评估推进到大规模推理。平台可在约一天内完成大陆尺度推理,处理数十 TB 影像,成本为每平方公里不到一美分。一次覆盖全北美的野火风险图运行峰值动用约 19,600 个 CPU 和 994 个 GPU,网络吞吐超 168 GB/s,将约 4,737 小时串行计算压缩到约 30.5 小时,实现 155× 加速。

    andle concurrently. To avoid overwhelming those services, the OlmoEarth Platform maintains its own metadata index, updated as new imagery is published. For datasets hosted through AWS Open Data, we receive an SNS notification for each new scene. When a provider does not offer a change stream, we poll its upstream index every few minutes. As a result, our requests to external services follow the steady pace of new publications rather than the sudden burst generated by a large inference job. Each

    匹配字段正文

  5. TechCrunch AI

    Meta 否认 Muse 未经许可读取用户私信

    Meta 回应 Inc. 专栏作者 Jason Aten 的说法,否认其 AI 智能体 Muse 未经许可读取用户私信。Meta 传播副总裁 Andy Stone 表示,Mac 版 Muse 的 Messages 集成完全为选择性加入,需同时开启 Full Disk Access 和 Messages 连接器才能读取消息内容。Meta Superintelligence Labs 高管 David Singleton 补充称,相关授权涉及三步应用层权限和 macOS 系统级保护,即便应用存在 bug 也无法绕过。Aten 则称其 Full Disk Access 处于关闭状态,Muse 自述是在同步设备通知。

    ta’s denial, many people are still suspicious that Meta isn’t being truthful. That’s not surprising , given the tech giant’s years of mishandling consumer data , which has led to lawsuits , FTC violations , and fines . Just days ago, for instance, a New Mexico jury determined the tech giant had misled users about its data practices in a case that resulted from the 2018 Cambridge Analytica data breach scandal. Whether users can trust Muse will be a deciding factor in whether or not Meta wins the

    匹配字段正文

  6. Anthropic Engineering

    Claude Code 最佳实践指南

    Anthropic 发布 Claude Code 最佳实践指南,核心约束是上下文窗口会快速填满且性能随填充下降。指南建议先给 Claude 可运行的验证方式(测试、构建、截图对比),并按探索、规划、实现、提交四阶段工作流推进,配合 plan mode 分离研究与执行。

    vel isolation that restricts filesystem and network access, allowing Claude to work more freely within defined boundaries Use CLI tools Tell Claude Code to use CLI tools like gh , aws , gcloud , and sentry-cli when interacting with external services. CLI tools are the most context-efficient way to interact with external services. If you use GitHub, install the gh CLI. Claude knows how to use it for creating issues, opening pull requests, and reading comments. Without gh , Claude can still use th

    匹配字段正文

  7. Anthropic:Claude.dev 开发者博客

    Claude Code mods 入门:从空文件夹构建第一个 mod

    Anthropic 发布 Claude Code mods 入门教程,mod 是运行在 Claude Code 会话内的 JavaScript 或 TypeScript 文件,可监听事件、改写行为或绘制自定义 UI,本质是插件内的 hooks。教程从空文件夹构建约 80 行的 Token Weather,在提示词上方实时显示上下文窗口占用,并演示 Blast Radius 和 Replay Theater 两个更复杂的 mod。需要 Claude Code 2.1.287 或更高版本,mods 默认开启,完整代码在 anthropics/claude-code-playground。

    is in the public anthropics/claude-code repository under mods/, so you can read how the team builds them. Token Weather reads how full the context window is after each turn and draws one line above the prompt: a weather icon, the percentage, the tokens used out of the window, a small chart of recent turns, and how much the last turn added. Here it is in a real session. Each turn reads more files, and the band fills from ☀ Clear to ☂ Showers to ☇ Storm: The shortcut: let Claude build it You can

    匹配字段正文

  8. Modal 官方工程博客

    Modal 推出 Sandbox Sidecars 低延迟信任边界

    Modal 发布 Sidecars,一种与主 Sandbox 同主机运行但保持隔离的容器,用于在可信与不可信代码之间建立安全边界。官方内部基准显示,Sandbox 与 Sidecar 之间的通信比同区域独立 Sandbox 之间至少快 3 倍,单个 Sandbox 最多可运行 250 个 Sidecar。Sidecar 通过 Modal SDK 动态创建,拥有独立出站网络策略,Sandbox 内代码无法创建或修改它们;Ramp 已用其为编码智能体 Inspect 搭建自定义出站代理。

    lace to inject credentials, proxy traffic, redact data, run harness logic, or observe the agent without requiring every operation to cross a remote service boundary. The pattern draws inspiration from existing designs like Kubernetes sidecar containers or the Datadog Agent: small, specialized containers that run alongside a primary application to provide supporting functionality. Sidecars apply the same idea to agent infrastructure, with the isolation needed to establish a trust boundary between

    匹配字段正文

  9. Latent Space

    OpenAI 团队回应 Computer Use 进展质疑,并复盘 Decisions API 一周冲刺

    在 OpenAI DevDay 后的播客中,负责 Computer Use 智能体的 Ari Weinstein 回应了 Dwarkesh 关于 Computer Use 进展缓慢的质疑,称其与几个月前相比已完全不同,原因是模型更擅长调试和重试,并会编写 JavaScript 一次执行多步操作,同时结合截图、accessibility、DOM 与 Playwright。他还提到 Dots 为每个 Dot 提供云端 Linux 虚拟电脑,Computer Use 已进入 Agents API。OpenAI API 团队负责人 Nikunj Handa 则介绍 GPT-6 的异步工具调用、mid-turn steering、WebSockets、UltraFast 推理、prompt 缓存与预加热、compaction 等更新,并称 Decisions API 基于 Luna 权重、未训练新模型,从灵感出现到原型仅约一周。

    Swyx [00:38:30]:对。我一直会回到的那个类比,我们就以此收尾吧,就是,你在构建一朵 AI 云,对吧? Nikunj Handa [00:38:35]:嗯哼。 Swyx [00:38:35]:就像,这是 Sam 一年前说过的东西 Nikunj Handa [00:38:38]:嗯哼 Swyx [00:38:38]:在那里,而你,这几乎就像你在做 AWS 的发明,你必须做,好,这是 EC2 Nikunj Handa [00:38:45]:对 Swyx [00:38:45]:而这是 S3,而这是像……但你做的是每一个的 AI 原生版本。 Vibhu [00:38:48]:有很多类比,所以你在为你明知会用到的东西预热缓存 Nikunj Handa [00:38:53]:对。 Vibhu [00:38:53]:而且很好的一点是,这一切都暴露给构建者 Nikunj Handa [00:38:56]:嗯哼 Vibhu [00:38:56]:因为它就打开了你可以构建新东西的各种方式。 Nikunj Handa [00:38:59]:对,绝对是这样。 Swyx [00:39:00]:好。

    匹配字段正文

  10. Newcomer 新闻长文

    Machine Earning AI 峰会:个人智能体与智能体商务的早期乐观

    在旧金山举行的 Machine Earning AI 峰会上,个人智能体成为最热话题,但与会者普遍认为让 AI 替人做金融决策仍处早期。Town CEO Jean-Denis Greze 预测一年内人们 40% 的工作数字时间将交给助手,五年内达 90%;52% 的观众认为 Muse 一年内将拥有最高智能体市场份额。OnePay 的 AML 智能体把原本 300 到 400 小时的合规工作压缩到 15 到 20 分钟,Gusto 则把月度预测从 14 天减到 7 天。

    Madeline Renbarger's avatar

    China, sent a letter earlier this month asking Treasury Secretary Scott Bessent to assess the risks Airwallex poses. The committee said it hadn’t concluded the company broke any laws. Sood argued that any financial services company with global ambitions has to operate in Asia, just as Citi, JPMorgan, PayPal, and Stripe do. He said Airwallex has gone beyond compliance standards and gotten outside validation to reassure customers and regulators. He brushed off some criticism as commercially motiv

    匹配字段正文

  11. n8n Blog

    n8n Assistant 额度支持充值:Cloud Starter 与 Pro 用户可购买额外 credits

    n8n Cloud 的 Starter 和 Pro 用户现在可以在 Assistant credits 用尽后购买额外额度,支持手动充值和自动充值,额外 credits 可跨月结转并在购买 12 个月后过期。Cloud 管理后台新增 Assistant credits 标签页,可查看余额、24 小时/7 天/30 天消费图表、用量最多的用户和充值历史;编辑器内的 Assistant 面板会显示当前对话消耗并在剩余 10% 时提醒。官方给出的典型消耗区间为:简单提问 10-20 credits,单次提示生成简单工作流 50-100 credits,带测试的较大构建或较长调试 300-600 credits。

    e. One tip that definitely helps: specific prompts use fewer credits than open-ended ones. A few things to know Credits are shared across your instance. Everyone on the instance draws from the same balance. If five people on your team use the Assistant, they're all using the same pool, and the top users view shows who's using what; Only the instance owner can top up. Topping up also requires an active paid plan; Your workflows keep running. Running a workflow never uses Assistant credits, even i

    匹配字段正文

  12. The Decoder

    Meta 将 AI 数据中心归为试点模型,2025 年省税 39 亿美元

    Meta 利用一项联邦研究税收抵免,把 AI 数据中心归类为试点模型、把 Nvidia 芯片列为实验材料,2025 年因此省下 39 亿美元,高于前一年的 20 亿美元和 2023 年的 7 亿美元,成为该抵免额度最大的受益上市公司。这一说法与 Meta 对外宣称的数百亿美元算力投入和 Prometheus、Hyperion 等多吉瓦集群计划并不一致。Meta 在 SEC 文件中提示相关节税可能被挑战,其不确定税务头寸准备金上升 45% 至 187.4 亿美元;审计机构 EY 批准了该策略,并正把类似做法推荐给其他公司。

    from our business to do this," Zuckerberg wrote last summer. By June 2026, Meta was openly detailing its massive compute infrastructure , including partnerships with Nvidia, AMD, AWS, Arm, and Broadcom, alongside its own custom MTIA chips. None of this sounds like an experiment. Zuckerberg himself said in January 2025 that these data centers would "drive our core products and business." The tax credit dates back to a 1981 law. James Shannon, the congressman who introduced it, told the NYT it wa

    匹配字段正文

  13. Gary Marcus:The Road to AI We Can Trust

    OpenAI 等公司能否持续逃脱追责?专访福特汉姆法学教授 Zephyr Teachout

    Gary Marcus 就 AI 公司是否违法、能否被追责专访福特汉姆法学院教授 Zephyr Teachout。Teachout 认为,OpenAI 的 AI 智能体入侵 Hugging Face 服务器、访问澳大利亚政府卫生系统、试图入侵美国教育部网站和一所大学图书馆,未经授权访问计算机在美国属《计算机欺诈与滥用法》下的联邦犯罪,司法部应发出传票调查公司内部知情情况。她还指出,各州可依据侵权法中的异常危险活动严格责任、缺陷产品与欺骗性行为相关法律追责,纽约州商业公司法赋予总检察长对反复违法企业申请司法解散的权力,这一权力曾被用于特朗普集团。

    AI debate because I saw way too many people treating repeat evidence of lawbreaking as wholly unreachable by current law. Is there any reason to think that OpenAI has broken any laws that (at least in another time) might be enforced? Which and why? In the world, in general, law enforcement starts when there’s evidence of a crime, and then investigates. A DA doesn’t wait for all the evidence to investigate a crime, and an AG doesn’t wait for all the evidence in the newspaper to investigate illeg

    匹配字段正文

  14. The Decoder

    FTC 对 OpenAI、Anthropic 等 AI 实验室启动消费者保护调查

    美国联邦贸易委员会正就潜在的消费者保护违规问题调查 OpenAI、Anthropic 等主要 AI 实验室。FTC 主席 Andrew Ferguson 计划通过具有法律约束力的“民事调查令”强制对方移交文件并质询高管,相关命令预计在数周内发出。调查在 Hugging Face 遭黑客攻击事件之前就已启动,AI 安全机构 METR 也在审查范围内。

    Musk (xAI) signed a voluntary pledge at the White House committing to independent external audits of their AI models. Ferguson cautioned that major AI companies shouldn't push for rules only they can meet as a way to box out rivals. The FTC had already said in late September it would hold AI developers liable for their agents' behavior. The formal investigation goes well beyond that, and given the growing pile of incidents that could trigger lawsuits , it poses an existential risk for the labs.

    匹配字段正文

  15. The Register

    MeetTwins:用 AI 数字分身替你参加 Google Meet 会议

    印度开发者 Aditya Shinde 正在招募测试用户的 MeetTwins,可代替本人参加 Google Meet 会议,基于印度 AI 公司 Sarvam 的模型构建,并支持 Simli 生成的数字分身头像。用户需预先设定可回答的问题、提供文档并划定禁谈话题,超出范围的内容会转交本人,会后以邮件汇总讨论要点。参会者若感到不适,可在聊天中输入 "/stop" 让机器人立即退出。

    tonking Stingray' beta swims out: Deeper Rust 'oxidization' plus LLM speech-to-text GNOME 51 is here, and Ubuntu interim releases are getting interesting again; Revive Raskin's 3 laws of software with humane work that considers a dev's duty of care All this, and angry NeoVim users; Firefox 157: Could bold new look be just what Mozilla needs? Yet another new version of Firefox, with a significant reskin; KDE turns 30 and someone's brought an AI-native desktop proposal Akademy talk imagines Plasma

    匹配字段正文

星期二

  1. Anthropic Research

    GLM-5.3 与高级网络攻击能力的扩散

    Anthropic 分析指出,智谱 AI(海外称 Z.ai)的 GLM-5.3 具备自主构建端到端网络漏洞利用的能力,且发布时未设有效防护,模拟测试中攻击者可 64% 至 100% 绕过其安全机制。在 ExploitBench 上,GLM-5.3 于 410 次尝试中成功开发端到端漏洞利用 50 次,接近 Claude Mythos Preview 的 56 次;NIST CAISI 称其为迄今网络能力最强的开放权重模型,落后美国前沿约四个月。

    view earlier this year). Here, we select targets in which the human experts are unaware of existing vulnerabilities, then ask them to use the model to identify and exploit novel flaws. These experiments tested what the experts could do in a short time-frame: they typically ran for a day or less, with less than an hour of human focus in total. In the first of these sessions, a researcher used GLM-5.3 on a sandboxed machine with a local Linux build of a popular web browser. Over the course of a da

    匹配字段正文

星期三

  1. TechCrunch AI

    OpenAI 为 Codex 推出可跨设备复用的云开发环境

    OpenAI 在 DevDay 上为软件工程智能体 Codex 推出可复用的云开发环境,开发者可在电脑、手机或云端运行 Codex,任务启动更快,团队还能共享带审批设置和权限的工作区。Codex CLI 更新支持用语音启动和指挥任务,新增 /agents 视图便于同时委派和跟踪多个任务,并可在 ChatGPT 桌面应用内进行代码审查。OpenAI 还推出 Codex Security Cloud,可按需或定期扫描整个 GitHub 仓库,并在新提交到达时自动检查、去重和准备修复。

    nAI’s cybersecurity initiative Daybreak Blue , without requiring developers to submit a separate application. OpenAI also announced several API updates, including a new Decisions API for real-time decision-making. The API uses Luna to answer a set of user-defined questions with predefined answers. The company also introduced an updated a version of the Agents API that now supports computer use and the ability to use Amazon’s Bedrock Managed Agents to build OpenAI agents that run entirely on AWS.

    匹配字段正文

  2. The Verge AI

    Google 被曝试点向出版商付费,补偿内容对 AI 搜索的贡献

    据 The Information 报道,Google 已启动一项试点,向约 100 家出版商按其内容对 AI Overviews、Search 中的 AI Mode 以及 Gemini 聊天机器人的贡献付费。Digiday 最早报道该试点,称其启动不到一年;有出版商加入一年收入超过 100 万美元,另有几个月前加入的出版商收入约 5 万至 6 万美元。Google 的 AI 搜索改动已招致出版商的批评与诉讼,也引发全球监管关注:6 月英国裁定 Google 必须允许出版商选择不进入 AI 搜索功能,欧盟则就其网络流量影响展开调查并要求其修改搜索引擎。

    n a year, while another that joined a few months ago has earned around $50,000 to $60,000, The Information reports. Google’s AI-focused search changes have not only drawn criticism and lawsuits from publishers , but also from regulators around the globe . In June, the UK ruled that Google must allow publishers to opt out of appearing in AI search features, while the European Union opened an investigation into its impact on web traffic and recently ordered it to make changes to its search engine.

    匹配字段正文

  3. a16z:News

    a16z 解读 1 万亿美元 AI 基建:State of Markets 的 25 张关键图表

    a16z 的 David George、Sarah Wang、Alex Immerman 和 Santiago Rodriguez 解读最新 State of Markets 演示中的 25 张关键图表,指出超大规模云厂商的年度 CapEx 正逼近 1 万亿美元,算力需求持续超过供给。他们还讨论了 OpenAI 与 Anthropic 的收入增长、AI 部署与企业可衡量影响之间的差距、智能体兴起、推理成本下降及其对 SaaS 的影响。

    pecific pooled investment vehicles — which should be read in their entirety, and only to those who, among other requirements, meet certain qualifications under federal securities laws. Such investors, defined as accredited investors and qualified purchasers, are generally deemed capable of evaluating the merits and risks of prospective investments and financial matters. There can be no assurances that a16z’s investment objectives will be achieved or investment strategies will be successful. Any

    匹配字段正文

  4. The Verge AI

    特朗普 AI 安全协议细节公布,多家科技公司高管签署自监管承诺

    特朗普昨日宣布的“具有道德约束力”AI 安全协议《前沿责任联合承诺》全文公布,由 David Sacks 在线上分享,签署者包括 Google 的 Sundar Pichai、Anthropic 的 Dario Amodei、Meta 的 Mark Zuckerberg、OpenAI 的 Greg Brockman、xAI 的 Elon Musk 和 Nvidia 的 Jensen Huang,特朗普本人也署名。协议要求各公司自行安全开发技术,并定期会面制定标准和最佳实践,列出四条确保前沿模型按预期运行、问题能被快速发现和解决的规则,但未提及违约惩罚。特朗普同日还下令美国政府今后将 AI 称为“Super Intelligence”。

    ing concerns around AI safety — “confidence that the technology is operating as intended.” A later passage also says that “over time, it may make sense to codify these steps into laws or regulations.” In their current state, these rules are mostly vague, common-sense safety procedures for any tech company, and don’t mention any penalties for violating the agreement, making this little more than a pinky promise from AI leaders to do the bare minimum. And that’s despite Google , OpenAI , and Anthr

    匹配字段正文