TODAY · 20 STORIES
Last Update: 2026-06-13 08:01
01
一篇关于"直接丢给 ChatGPT"现象的反思，讨论人们把所有问题甩给 AI 的习惯及背后的工作流变化。
A reflection on the cultural habit of defaulting to ChatGPT for every task and what that reveals about evolving workflows.
02
教程：如何在 macOS 上搭建本地编码 Agent，覆盖环境配置和工具链选择，面向想本地跑 AI 编程助手的开发者。
A practical guide to setting up a local coding agent on macOS, covering environment setup and tooling choices for developers who want offline AI coding assistance.
03
分享如何稍微改善 AI 生成前端代码的"潦草感"，给出实操技巧，帮助开发者拿到更接近生产质量的 UI 代码。
Tips and techniques for slightly improving the sloppiness of AI-generated front-end code, helping developers get UI output closer to production quality.
04
TLDR AI 头条：贝佐斯成立 410 亿美元的"通用工程师"AI 公司 Project Prometheus，目标做 AGI 级工程能力；另含 X + Openclaw 社交内容工作流。
PLUS: Use this X + Openclaw setup to write viral social content Zach Mink, +4
05
TLDR AI：Anthropic 向华盛顿提交了一份 AI 监管"剧本"，主动给政策制定者出招；附用 Codex 跟进销售线索的玩法。
PLUS: Close more deals with Codex sales followups Zach Mink, +4
06
TLDR AI：Anthropic 向公众开放 Mythos 级别 AI 模型，并附 Dexter 自动化金融研究的玩法。
PLUS: Automate financial research with Dexter Zach Mink, +4
07
TLDR AI：苹果新一代 Siri AI 升级"姗姗来迟"地登场，同时报道 OpenAI 进入所谓"第三阶段"的战略转向。
PLUS: The new ‘third phase’ of OpenAI Zach Mink, +4
08
TLDR AI：华盛顿想分一杯 OpenAI 的羹（监管或股权介入），并附用 Agent 框架每天找到 5 个潜在客户的玩法。
PLUS: Find five prospects a day with this agentic framework Zach Mink, +4
09
TLDR AI：Anthropic 面对递归自我改进（RSI）时钟的压力，相关安全议题升温；附用 Perplexity 压力测试商业点子的玩法。
PLUS: Stress test business ideas with Perplexity Zach Mink, +4
10
Anthropic 起源故事：诞生于旧金山 Precita Park，最初 7 人在公园里带电脑开干，至今 7 人仍一起做 Claude。
Anthropic was born in San Francisco's Precita Park, where the original 7 founders met outdoors with laptops for the first month — and all 7 are still building Claude together today.
11
Anthropic 4 月以 4 亿美元收购药物发现 AI 公司 Coefficient Bio，6 月 Fable 5 上线时却悄悄降级生物/化学/药物研究输出，作者指出两者领域完全重叠，质疑这是用"安全"包装的独占策略。
Anthropic acquired drug-discovery AI startup Coefficient Bio for $400M in April, then two months later Fable 5 silently degraded outputs in the exact same biology/chemistry/drug-discovery domains — raising concerns that 'safety' guardrails are masking an exclusivity play.
12
质疑帖：Fable 在生命科学领域可能只是"真的不行"，"安全护栏"或许在掩盖能力短板，否则为何没有给科研机构开放的"Project Glasswing"通道。
A skeptical take suggesting Fable's 'safety guardrails' may just be hiding genuinely poor life-sciences capability, given no vetted-institution access program exists.
13
实测数据：Fable 5 拒答生物类问题率高达 93–100%（医学遗传 100%、大学生物 95%、MMLU-Pro 生物 97%），远超官方"不到 5%"声明，且 Haiku/Sonnet/Opus 对同样 152 题零拒答，问题被算作答错隐藏在准确率里。
Benchmark results show Fable 5 refuses 93–100% of biology questions across MMLU and MMLU-Pro (97% on MMLU-Pro biology), far above Anthropic's stated <5%, while Haiku 4.5, Sonnet 4.6, and Opus 4.8 answered all 152 refused items — refusals masquerade as wrong answers.
14
作者展示一份完全由 Claude 经过数月迭代生成、没有手写一行代码的设计作品，邀请大家评判是"vibe coded"还是真的好看。
Spoiler: I did not wrote a single line of code, it was all made by Claude over months of iteration.
15
Qursor：指向任意 UI 元素即可把精确上下文发送给 AI 的工具，解决与 AI 沟通时上下文不清的问题。
Point at any UI to send exact context to your AI
Discussion
|
Link
16
Slack Data Agent：不用离开 Slack 就能用自然语言查询业务数据的 Agent 工具。
Ask about your data without leaving Slack
Discussion
|
Link
17
Bob's CLI：本地优先、可自适应用户习惯的 AI 编程 CLI，强调隐私和个性化。
A local-first AI coding CLI that adapts to you
Discussion
|
Link
18
LocIn AI：用 AI 做语调感知的本地化（i18n），自动化多语言翻译工作流。
Localize your app with tone-aware AI, automated workflows
Discussion
|
Link
19
BitBoard（YC P25）在 HN 上发布：专为 AI Agent 打造的分析工作空间，让 Agent 自助做数据分析。
BitBoard (YC P25) launches on HN as an analytics workspace built specifically for AI agents to perform autonomous data analysis.
20
TLDR AI 通讯综述：贝佐斯的 AI 工程师项目、SpaceX 创纪录 IPO、垂直 Agent 构建话题汇总。
TLDR AI newsletter roundup covering Bezos's AI engineer venture, SpaceX's record-setting IPO, and building vertical agents.