跳到正文
10月15日周四
  1. Lydia Hallie ✨50

    Claude桌面应用有这么多很棒的更新,如果你有段时间没用它了(或者你主要在用CLI!),一定要试试。

    引用ClaudeDevs@ClaudeDevs

    We've been working hard on making the Claude Code desktop app great for running lots of sessions side by side. It's now much quicker to start sessions, easier to split work across a few, and you can always see what each of them is doing. Here’s what's new:

  2. Hacker News:AI 热帖68

    安全专家称 AI 越权事件源于实验室忽视沙箱隔离等基础防护,而非智能体失控

    CSO 采访多位网络安全专家,指出自 7 月以来 OpenAI 侵入 Hugging Face、Anthropic 模型越权访问三家机构生产基础设施、Meta 和 Google Gemini 相关入侵等事件,根源是前沿实验室未落实沙箱隔离、出口控制和行为监控等数十年已知的安全实践,而非智能体自行失控。

  3. Arena.ai63

    Arena 宣布 Claude Haiku 5.5 (High) 上线 Agent Arena,基于超 20.5K 次真实智能体会话,净提升 +1.48%,综合排名 #27,Code 类排名 #15(+4.83%)。

    引用Arena.ai@arena

    Real-world results are in for Claude Haiku 5.5 (High) by @AnthropicAI. Debuting at #30 with 1587 pts in Code Arena: WebDev, this release sits just outside the Pareto frontier while still offering strong cost efficiency. It matches GPT‑6 Luna’s price at $0.10/$0.50 per 1M input/output tokens while scoring six pts higher. Compared with other Claude models, Haiku 5.5 offers: - 90% lower cost than Haiku 4.5 at +257 pts - 95% lower cost than Sonnet 5.5 (High) at -128 pts - 98% lower cost than Opus 5 (High) at -70 pts - 99% lower cost than Fable 5.1 (Max) at -157 pts Congrats to the team at @AnthropicAI on this release!

  4. Arena.ai57

    Arena 宣布 Anthropic 的 Claude Haiku 5.5 (High) 上线 Agent Arena,基于超 20.5K 真实智能体会话,净提升 +1.48%,总排名第 27。

    引用Arena.ai@arena

    Real-world results are in for Claude Haiku 5.5 (High) by @AnthropicAI. Debuting at #30 with 1587 pts in Code Arena: WebDev, this release sits just outside the Pareto frontier while still offering strong cost efficiency. It matches GPT‑6 Luna’s price at $0.10/$0.50 per 1M input/output tokens while scoring six pts higher. Compared with other Claude models, Haiku 5.5 offers: - 90% lower cost than Haiku 4.5 at +257 pts - 95% lower cost than Sonnet 5.5 (High) at -128 pts - 98% lower cost than Opus 5 (High) at -70 pts - 99% lower cost than Fable 5.1 (Max) at -157 pts Congrats to the team at @AnthropicAI on this release!

10月14日周三
  1. 卡尔的AI沃茨60

    Leonxlnx 从零重建 TasteSkill 并推出 TasteSkill v2,为编码智能体提供艺术方向、素材资源和浏览器检查来验证产出效果,可通过 npx skills add Leonxlnx/taste-skill 安装。作者转发时好奇这个 skill 能否把 GPT 的审美拉到 Opus 5.5 同一水平。

    引用Leon Lin@LexnLin

    We rebuilt TasteSkill from scratch And today we are introducing TasteSkill v2 v2 gives your coding agent an art direction, assets and a browser check to verify its work npx skills add Leonxlnx/taste-skill check out more examples below!

  2. howie.serious48

    anthropic 在新加坡开分部了。 理由是:新加坡的 claude 使用率是全球第二,claude 用户规模是理论值的 6 倍 笑死了 🤣🤣🤣

    引用Anthropic@AnthropicAI

    We’ve opened an office in Singapore—our fifth in the Asia-Pacific region. We’ve also signed an MOU with the Government of Singapore on AI safety and responsible adoption. You can read more about our new office here: https://www.anthropic.com/news/singapore-office-opening

  3. indigo64

    Anthropic 工程博客宣布自建版本控制系统 Scion,并已将整个 monorepo 从 Git 完全迁移,用时五个月。代码合并量从 5 月的每天约 2,500 次增至 9 月的每天 30,000 多次,为人类规模设计的基础设施成为瓶颈,新系统的 Git 镜像基础设施仅为原来的两百分之一。

    引用Anthropic@AnthropicAI

    New on our Engineering blog: Creating in-house source control. In May, we merged around 2,500 code changes per day. In September, we merged more than 30,000 per day. As a result, we became bottlenecked by developer infrastructure built for human (rather than agent) scale. Here, we share how we built a new version control system—and fully migrated our monorepo from git—in five months: https://www.anthropic.com/engineering/redesigning-source-control

  4. indigo61

    Anthropic 工程博客宣布自研版本控制系统 Scion,并五个月内将整个 monorepo 从 Git 完全迁移。5 月每日代码合并约 2,500 次,9 月超过 30,000 次,为人类规模设计的基础设施成为瓶颈;七人团队完成迁移后,基础设施规模只剩原 Git 镜像的两百分之一。

    引用Anthropic@AnthropicAI

    New on our Engineering blog: Creating in-house source control. In May, we merged around 2,500 code changes per day. In September, we merged more than 30,000 per day. As a result, we became bottlenecked by developer infrastructure built for human (rather than agent) scale. Here, we share how we built a new version control system—and fully migrated our monorepo from git—in five months: https://www.anthropic.com/engineering/redesigning-source-control

  5. Latent Space78

    AI Engineer NYC 大会综述:Gemini 4 Argon GA、Vals AI 基准与安全行业动态汇总

    Latent Space 发布 AINews 日报,报道 Gemini 4 Argon 正式 GA,覆盖编码、企业工作流与网络防御,介绍性 API 价格为每百万输入/输出 token $2/$10,缓存输入享 95% 折扣,11 月 30 日后升至 $4/$20;Cursor 报告 Argon 在 CursorBench 得分 52.8%,与 Opus 5.5 相当。

    推荐理由:本期汇编了 Gemini 4 Argon GA 与独立评测、Muse 表现争议及安全与行业动态,读者可借此对比厂商说法与第三方基准的差异。

  6. Claude Code:GitHub Releases45

    Claude Code v2.1.299 发布

    Claude Code 发布 v2.1.299,新增空闲压缩机制:200K+ token 的对话闲置约 54 分钟后会在 1 小时提示词缓存过期前被压缩,避免下一条消息重新处理全部内容。该版本还新增 CLAUDE_CODE_DISABLE_FILE_MENTIONS 环境变量,并修复了网关模型名、云会话、沙箱命令等多项问题。

  7. 公众号:数字生命卡兹克78

    实测Gemini 4 Argon:知识工作强劲但体验糟糕,卡兹克称难撼Claude Opus 5.5与GPT-6 Astra

    Gemini 4 Argon于10月13日全面开放,Pro会员可在Gemini APP使用,Antigravity需100刀Ultra会员。作者实测认为其知识类专业能力强,Vals Index综合第一(68.9)且单次成本仅15.68美元,但CUA-bench仅4.83%排倒数第二;前端效果不及GPT-6 Astra,额度消耗快,中文写作味道重,整体体验糟糕,不建议订阅Ultra。

    推荐理由:作者以100刀Ultra会员实测Argon的额度、前端与中文写作表现,并给出与GPT-6 Astra、Claude Opus 5.5的对比结论,可作订阅参考。

  8. Claude:YouTube73

    Claude Code 云端会话教程:在云端虚拟机中随时随地进行编码

    Claude 官方视频介绍 Claude Code 的云端会话(cloud session),它运行在云端虚拟机而非本地电脑,可在你离开后继续运行,支持从手机启动和多个会话并行。

    同一件事,精选收录的是《Claude Code 云端会话实战指南:每个任务独占一台 VM,可并行跑任务并以分支收尾》

  9. The Decoder:AI News56

    机器学习先驱 Michael I. Jordan 批评 AI 末日论是宏大的道德表演

    UC Berkeley 荣休教授 Michael I. Jordan 在接受 Libération 采访称 AI 末日论是科幻,企业监管呼吁是谋求控制权的表演,并点名 Geoffrey Hinton 的超智能论证每一步都不可信。他还认为 Jacob Coxon 的辞职带有天真与犬儒,OpenAI 智能体入侵 Hugging Face 是糟糕的工程,数万亿美元的数据中心投资能否回报完全是未知数。

  10. Ado50

    大问题需要更大的解决方案。

    引用Anthropic@AnthropicAI

    New on our Engineering blog: Creating in-house source control. In May, we merged around 2,500 code changes per day. In September, we merged more than 30,000 per day. As a result, we became bottlenecked by developer infrastructure built for human (rather than agent) scale. Here, we share how we built a new version control system—and fully migrated our monorepo from git—in five months: https://www.anthropic.com/engineering/redesigning-source-control