Latest
Termany significantly improves developer productivity by enabling parallel development across multiple projects simultaneously. The traffic light status system helps users track agent performance without switching contexts, making it ideal for developers managing complex workflows. Compared to traditional terminals like Codex, Termany offers superior organization and multi-agent capabilities.
有人拿 Opus 5.5 把一小时播客直接变成白板讲解视频，只花了 2% 额度，同场景下 Astra 连剪辑都做不利索。
有个叫 ai-new-tab 的 Chrome 扩展，让 Claude 扫一遍你的浏览历史，把新标签页变成只给你看的主页，代码在 GitHub 上开源，想折腾的可以试试。
流畅阅读更新了，Chrome 开源翻译插件里网站翻译做得最顺，YouTube 字幕能直接下载丢给 Agent 总结，平时看外文内容挺省事。
宝玉把 Claude Slides 的编辑器逆向出来了，还顺手用 Claude Code 5.5 重写了一遍。想研究别人家 Slides 怎么做的，可以照这个流程走。
手里有Grok Bot、Muse、Hark这些工具却不知道干啥的，去这个站抄作业就行。1040多个案例按27类Agent分好，每个都带现成prompt，直接复制。
Skills are becoming portable assets rather than per-tool settings, so a single registry reduces the manual copying that breaks when a tool changes its directory layout. Developers running several coding assistants across more than one machine can use it to keep skill versions aligned, though the project is community-built and its compatibility with each of the 50-plus tools is not independently verified.
Hyung Won Chung 讲怎么把 Bitter Lesson 落到日常工程决策里：先加结构推进，条件变了就拆掉。做模型和系统设计的人可以看看。
Keras 作者 Chollet 在 TPU 上拿 KerasHub+JAX 玩 LLM 后训练，说可能开源，想跟进后训练工具链的可以蹲一下。
Simulation awareness is the practical risk here: if an agent behaves differently once it detects a test harness, eval scores stop predicting production behaviour, so teams need to verify that a simulation reproduces real traces before trusting it. Engineers building agent evaluation pipelines can use the chapter order as a checklist, starting with replaying production traces rather than writing fresh input/output tests.
Mole 作者在问默认清理策略：老对话是默认 120 天清掉，还是永不清理。你要是天天用 Cursor、Claude Code 这类 CLI，可以去投一票。
不会 Blender 也能做 3D 分身，丢一张立绘进去，Codex 加 Tripo 加 Blender 自动跑完建模和骨骼，7 个 Skills 卡在关键点让你拍板。
博主实测说 Grok Bot 后端换成 Opus 5.5 了，还晒了上传自己照片直接生成配音视频的效果，想玩视频生成的可以看看他的提示词写法。
做产品的朋友可以试试这个 Claude 工作流：一条指令找出最活跃的 10 个用户，自动约 15 分钟访谈，比手动筛数据快多了。
归藏的 Next Token 第 5 期上线了，聊 Dots 为什么难用、国内做 personal agent 靠不靠谱，还有 AI 复刻软件和小硬件的机会，5 期就 7000 订阅。
magpie 这个新功能会拿你自己的真实调用在本地跑回放，按 agent 和模型算出最省 Token 的那组配置，不用再瞎猜压缩阈值该调多少。
PewDiePie 拍了期用 env 做后训练的视频，说自己老陷在生成任务数量不够的循环里，Hugging Face 的 merve 也共鸣了，做 RL 训练的可以看看这条经验。
op7418 分享了个 GrokBot 实操：把 Keenan Crane 那串 27kb GLSL 隐式曲面建模的长推串直接拉全、补注释、排成中文文档，手机上就能读。
有人拿 Claude Motion 一条提示词就做出了 6 个广告和讲解视频，对比 After Effects 的手动制作流程，做产品宣传的可以看看这批案例。
LlamaIndex 创始人 Jerry Liu 拿教皇谈'智能要靠记忆'的推文给 RAG 站台， standpoint 很有意思，做 RAG 的可以看看。
有人拿 Notion 视频试了个新玩法：不用视频模型，让 Opus 5.5 用 Codex 生关键帧加代码做动画，对比视频贴出来了
karminski 拿一个还没发布的匿名模型写 three.js，挂绳动效纯代码搞定，以前得写几天，现在几小时，猜猜是哪家的？
The framing clarifies why reasoning-style models changed developer workflows: test-time computation replaces next-token completion, which explains the leap in coding-agent reliability. Engineers evaluating agent frameworks or choosing between reasoning and standard models can use this lens to judge where test-time reasoning pays off. The original chart on X condenses the argument into a single visual.
The most useful part is the practical diagnosis layer: entropy and advantage tracking, clip ratios and KL penalties shown with real training logs rather than equations alone, which is what most GRPO implementations get wrong. Researchers and engineers training reasoning models with open-source models can apply the stabilisation tricks directly to their own RLVR runs. Code and checkpoints are provided, so it works as a follow-along exercise rather than passive viewing.
Knowing which phase a given optimisation targets — faster GPUs for prefill, smaller weights or faster memory for decode — turns technique lists like quantisation and speculative decoding into a coherent decision framework. Practitioners running local inference or serving stacks can use pp128 and tg64 benchmarks separately instead of a single speed number. The guide is hosted on llama.app and centres on llama.cpp tooling.
让 agent 直接在屏幕上画箭头告诉你点哪个按钮，不用再对着终端文字自己找，写 agent 工具的可以看看这个思路。
Garry Tan 晒了用 Capy 给自己的 AI Agent 检索指标做自动评测优化，16 小时没人管也在跑，做 Agent 的可以看看这套玩法。
Only the latest 2000 items are listed; for older ones see the daily archive