本期精选:Karpathy 宣布加入 Anthropic 回归 LLM 研究一线;Claude Code 本周限速重置 + 限额提升 + Agent SDK Credit 即将上线;Sam Altman 宣布 AI 首次解决重大数学开放问题;Aaron Levie 提出「CEO AI 幻觉」——他们看不到 Agent 落地的最后一公里;Google I/O 推出 Gemini Spark、Gemini Omni 和 Project Genie 全球开放;Garry Tan 论 Agent 要先把无聊的事做好。来自 @karpathy、@sama、@levie、@garrytan、@alexalbert__ 等 15+ 位核心人物的当期推文精华。
claude -p 调用及第三方 SDK 应用。4claude agents 控制台——在终端打开后可统一管理多路 Claude Code 会话 5;以及 Claude Security 进入公测,直接嵌入在 Claude Code Web 端,可对仓库做漏洞扫描并就地修复。6implementation-notes.html,记录规范之外的决策、权衡与改动原因 7——他把这称为「HTML 时代」,认为让 AI 用 HTML 输出比纯文字或 Markdown 信息更丰富。CEOs are uniquely prone to AI psychosis because they're sufficiently distant from the last mile of work that still has to happen to generate most value with AI.
Everyone building AI agents is focusing on building the prefrontal cortex... But also, a reframe: there is value in building the cerebellum. Most agent frameworks will fail because they treat all cognition as high cognition. The winners will nail the boring stuff first.
Processed 1400 replies ◾ OpenAI is catching up to Anthropic ◾ 'Codex' got more mentions than 'Claude Code' ◾ However, by model mentions, A\ is mogging
- 纯文本 → 2) Markdown → 3) HTML → … → n) 由神经网络直接生成的交互式视频/模拟
implementation-notes.html 方案可以相互印证。autotriage 规则,读取仓库 VISION.md 来判断哪些 Issue/PR 符合项目方向、代码可推断性高、可自动验证修复效果,然后让 Codex 自主处理这些任务,他只负责最终 review 建议。Codex 可通过 VM + 计算机视觉验证修复结果,无需打断他的工作流程。cmux——一个 tmux-like 的工具,可以同时运行多个 Codex CLI 会话,收获 3657 收藏。23The AI jobpocalypse is not a thing. The mass unemployment thing that AI lab CEOs are talking about—that's not coming.
在 AI 原生团队中,IC 应该开始像 Manager 一样思考(如何向 Agent 下达任务、如何设定标准和验证输出),Manager 应该开始像 IC 一样思考(亲手构建,而不只是管人)。
x.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
github.com
x.com
x.com
x.com
x.com
x.com
x.com
x.com
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.