Claude Blog
Claude 博客探讨如何以 AI 原生方式运行工程组织，涵盖工具、流程与文化调整。
推荐理由：对技术管理者极具启发，直接提供可借鉴的组织运行实践和思考框架。
2026-06-03（内容获取于 06/03 21:23）
X 推文 (AttentionVC)
Claude Code 发布动态工作流：Claude 可按需编写自定义 harness 来执行任务，不再局限于默认编码流程。
推荐理由：提升 Claude Code 灵活性和任务适应性，对日常使用 Claude Code 的开发者是实用更新。
X 创作者 (AttentionVC)
同上事件（同一来源，合并处理）：Claude Code 可针对任务自定义执行框架，增强 agent 能力。
推荐理由：展示 AI agent 工程化前沿能力，对构建复杂 agent 系统的开发者有重要参考值。
GitHub Trending
微软开源 Python 工具 markitdown，可将各类办公文档和文件批量转换为 Markdown 格式。
推荐理由：开箱即用，非常适合需要将文档内容提供给 LLM 处理的场景，能显著简化预处理流程。
Smol AI News
微软发布 MAI-Thinking-1 模型（35B MoE，256K 上下文），AIME 2025 达 97%，超越 Sonnet 4.6；另有 MAI 系列模型产品。
推荐理由：微软新模型性能卓越，可能改变开发者选择，值得关注其技术细节和对市场的影响。
GitHub Trending
headroom 可压缩工具输出、日志、文件及 RAG 块 60-95% 的 token，保持回复质量，提供库、代理和 MCP 服务器模式。
推荐理由：可直接部署使用，大幅降低 LLM 推理成本和延迟，对开发者极具实用价值。
OpenAI News
Travelers 保险基于 OpenAI 构建 AI 理赔助手，提供 24/7 引导式理赔提交和高峰期间运维扩展。
推荐理由：企业级 AI 落地标杆案例，展示大模型在传统行业中的实际应用和规模扩展路径。
Anthropic Engineering
Anthropic 工程师分享如何为 claude.ai、Claude Code 和 Cowork 构建安全隔离机制，限制 agent 爆破半径。
推荐理由：安全关键设计参考，尤其对构建自主 agent 系统的团队有直接借鉴意义。
Hugging Face Blog
博客探讨 Direct Preference Optimization 在非对话场景（如代码生成、推理）中的应用和适配方法。
推荐理由：对研究者和高级开发者有启发，介绍了 DPO 的边界拓展，但需要一定专业知识消化。
HuggingFace Trending Papers
提出 Value-Aware Stochastic KV Cache Eviction 方法，在推理模型长思维链场景下减少内存和计算瓶颈。
推荐理由：学术进展，对推理模型效率优化有参考价值，适合有相关背景的读者了解。
chopratejas/headroom
Python · ★ 8,460 · 🍴 563 · 📈 3,528 stars today
Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 60-95% fewer tokens, same answers. Library, proxy, MCP server.
中文介绍 Headroom 是一款压缩工具，能在 LLM 接收前减少 60-95% 的 token 数量，支持压缩工具输出、日志、文件和 RAG 块。提供函数库、代理和 MCP 服务器多种使用方式，适用于需要降低 LLM 调用成本、提高推理效率的开发者。
microsoft/markitdown
Python · ★ 142,311 · 🍴 9,706 · 📈 2,006 stars today
Python tool for converting files and office documents to Markdown.
中文介绍 Microsoft 开发的 Python 工具，可将各种文件格式和 Office 文档转换为 Markdown。适用于需要将文档内容标准化为轻量标记格式的场景，如文档迁移、内容提取或知识库构建。
affaan-m/ECC
JavaScript · ★ 205,009 · 🍴 31,457 · 📈 2,147 stars today
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
中文介绍 ECC 是面向 AI Agent 的性能优化系统，提供技能、本能、记忆、安全和研究优先的开发框架。主要支持 Claude Code、Codex、OpenCode、Cursor 等编码代理，旨在提升 agent 的自主性和可靠性。
D4Vinci/Scrapling
Python · ★ 59,776 · 🍴 5,769 · 📈 1,078 stars today
🕷️ An adaptive Web Scraping framework that handles everything from a single request to a full-scale crawl!
中文介绍 Scrapling 是一个自适应的网页抓取框架，能处理从单次请求到全站爬取的各种任务。内置反爬虫对抗机制，简化了复杂网站的爬取流程，适合数据采集、信息监测和自动化测试。
nesquena/hermes-webui
Python · ★ 12,909 · 🍴 1,578 · 📈 734 stars today
Hermes WebUI: The best way to use Hermes Agent from the web or from your phone!
中文介绍 Hermes WebUI 是专为 Hermes Agent 设计的 Web 界面，支持从浏览器和手机端使用 agent 功能。提供便捷的远程访问和交互方式，适合需要随时随地调用 AI agent 的用户。
reconurge/flowsint
TypeScript · ★ 4,787 · 🍴 604 · 📈 509 stars today
A modern platform for visual, flexible, and extensible graph-based investigations. For cybersecurity analysts and investigators.
中文介绍 flowsint 是一个现代化的图形化调查平台，支持可视化、灵活可扩展的图数据库分析。专为网络安全分析师和调查人员设计，用于追踪攻击链路、分析威胁情报和进行复杂关系推理。
OpenBMB/VoxCPM
Python · ★ 25,467 · 🍴 2,906 · 📈 713 stars today
VoxCPM2: Tokenizer-Free TTS for Multilingual Speech Generation, Creative Voice Design, and True-to-Life Cloning
中文介绍 VoxCPM2 是无 tokenizer 的 TTS 模型，支持多语言语音生成、创意声音设计和高质量声音克隆。无需传统文本标记化，直接生成自然语音，适用于语音合成、虚拟角色配音和个性化语音应用。
stefan-jansen/machine-learning-for-trading
Jupyter Notebook · ★ 18,876 · 🍴 5,259 · 📈 716 stars today
Code for Machine Learning for Algorithmic Trading, 2nd edition.
中文介绍 《Machine Learning for Algorithmic Trading》第二版的配套代码库，提供完整的量化交易机器学习实现。涵盖从因子工程到回测的全部流程，适合金融从业者和量化研究者学习与实践。
jamwithai/production-agentic-rag-course
Python · ★ 6,605 · 🍴 1,504 · 📈 372 stars today
中文介绍 面向生产的智能体 RAG 课程代码仓库，讲解如何构建可靠的企业级 RAG 系统。涵盖检索增强生成的最佳实践、agent 编排和部署策略，适合希望将 RAG 方案落地的 AI 工程师。
supermemoryai/supermemory
TypeScript · ★ 24,958 · 🍴 2,197 · 📈 601 stars today
Memory engine and app that is extremely fast, scalable. The Memory API for the AI era.
中文介绍 Supermemory 是一个高速可扩展的记忆引擎和应用，提供面向 AI 时代的记忆 API。支持持久化存储和快速检索上下文，适用于需要跨会话记忆的 AI 助手、聊天机器人和个性化推荐系统。
Open-LLM-VTuber/Open-LLM-VTuber
Python · ★ 8,747 · 🍴 1,096 · 📈 702 stars today
Talk to any LLM with hands-free voice interaction, voice interruption, and Live2D taking face running locally across platforms
中文介绍 开源 VTuber 方案，支持与任意 LLM 进行免提语音交互、语音打断，并集成 Live2D 面部动画。全平台本地运行，适合直播、虚拟角色互动或语音助手场景。
Value-Aware Stochastic KV Cache Eviction for Reasoning Models
👍 3
Reasoning models improve accuracy through extended chains of thought, but their long outputs create a memory and compute bottleneck. KV cache eviction methods reduce this cost by evicting unimportant key-value pairs from the cache, yet they often yield worse accuracy than selection-based sparse atte
中文介绍 推理模型通过长思维链提升准确性，但导致内存和计算瓶颈。KV缓存淘汰方法通过移除不重要键值对来降低成本，但准确率常低于选择性保留。论文提出价值感知随机KV缓存淘汰方法。
World Models Meet Language Models: On the Complementarity of Concrete and Abstract Reasoning
👍 16
World models and multimodal large language models (MLLMs) provide complementary capabilities for predicting future outcomes from static visual observations. World models can generate concrete visual rollouts of possible futures, while MLLMs can reason abstractly over questions, goals, and rules. How
中文介绍 世界模型与多模态大语言模型提供互补能力：世界模型可生成未来具体视觉推演，多模态大语言模型则能进行抽象推理。二者结合可更好地从静态视觉观测预测未来。
Small RL Controller, Large Language Model: RL-Guided Adaptive Sampling for Test-Time Scaling
👍 6
Test-time scaling improves the reasoning performance of large language models but incurs substantial cost in both total computation and latency. Existing adaptive sampling methods partially mitigate this issue by dynamically deciding when to stop sampling, yet they typically rely on heuristic rules
中文介绍 测试时缩放提升大语言模型推理性能但代价高昂。现有自适应采样方法部分缓解该问题但仍依赖启发式。论文提出小RL控制器引导大语言模型进行自适应采样。
PaddleOCR-VL-1.6: Expanding the Frontier of Document Parsing with Under-Optimized Region Refinement and Progressive Post-Training
👍 5
We introduce PaddleOCR-VL-1.6, an upgraded compact document parsing model built upon PaddleOCR-VL-1.5. Although PaddleOCR-VL-1.5 establishes a strong 0.9B baseline, its remaining errors concentrate in under-optimized regions where model behavior is unstable, data coverage is sparse, or supervision i
中文介绍 PaddleOCR-VL-1.6是基于1.5版本的升级版紧凑文档解析模型。通过优化不充分区域并渐进后训练，在0.9B参数基线上进一步提升文档解析性能。
NVIDIA OmniDreams: Real-Time Generative World Model for Closed-Loop Autonomous Vehicle Simulation
👍 5
As autonomous vehicle capabilities advance, the safe evaluation of driving policies in long-tail scenarios remains a critical bottleneck. In closed-loop simulation, the driving policy model actively interacts with the environment, where its actions dynamically update the simulator state and directly
中文介绍 NVIDIA推出OmniDreams实时生成式世界模型，用于闭环自动驾驶仿真。该模型动态更新驾驶策略与环境的交互，实现长尾场景安全评估。
Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories
👍 9
The past few decades have witnessed significant advances in the design of machine learning algorithms, from early studies on task-specific shallow models to more general deep Large Language Models (LLMs). Despite showing promising results in tasks that require instant prediction or in-context learni
中文介绍 研究提出语言模型需要类似睡眠的机制：通过自我修改与记忆巩固来提升学习能力。该方法受生物记忆处理启发，使模型能更好地整合长期知识。
Benchmarking Visual State Tracking in Multimodal Video Understanding
👍 4
Understanding a video requires more than recognizing isolated moments, as humans continuously track entities, states, and events over time. This capacity for visual state tracking is fundamental to video understanding, yet remains underexplored in current evaluations of Multimodal Large Language Mod
中文介绍 视频理解需持续追踪实体、状态与事件。论文提出基准用于评估多模态视频理解中的视觉状态追踪能力，指出该能力在当前评估中研究不足。
Ψ-Bench: Evaluating Persona-Sensitive Influencing in Persuasive Dialogues
👍 9
Personalization is a crucial capability of modern language agents. However, current research primarily positions personalized agents as passive responders to user preferences, limiting their ability to interact with users and provide suggestions or guidance proactively. To systematically evaluate su
中文介绍 Ψ-Bench评估对话智能体在说服场景中根据用户个性进行个性化影响的能力，弥补了现有研究将个性化代理仅作为被动回应者的不足。
TRON: Targeted Rule-Verifiable Online Environments for Visual Reasoning RL
👍 10
Reinforcement learning (RL) for visual reasoning needs scalable, verifiable, and controllable training signals. Existing visual RL post-training trains on static curated datasets, with fixed image-question-answer samples bounded by their collection budget. In this work, we introduce TRON (Targeted,
中文介绍 TRON提出面向视觉推理强化学习的可验证在线环境。相比静态数据集，该方法提供可扩展、可验证、可控的训练信号，提升视觉RL后训练效果。
AutoMedBench: Towards Medical AutoResearch with Agentic AI Models
👍 15
Autonomous agents are increasingly expected to support end-to-end medical-AI research workflows, moving beyond isolated prediction tasks or short-form clinical question answering. However, existing medical agent benchmarks primarily evaluate final outputs, providing limited visibility into agent beh
中文介绍 AutoMedBench评估自主AI智能体在端到端医学AI研究中的能力，超越孤立预测或简短问答，提供对完整研究流程的可见性。
A Local Perturbation Theory for Cross-Domain Interference and Recovery in Multi-Domain RL
👍 19
Reinforcement learning (RL) post-training improves large language models (LLMs) on individual domains such as mathematical reasoning, code generation, question answering, and creative writing (CW), but training on one domain often degrades performance on others. Existing explanations based on catast
中文介绍 多领域强化学习后训练提升某一领域性能时通常损害其他领域。论文提出局部扰动理论解释跨域干扰，并探索恢复方法。
Decentralized Instruction Tuning: Conflict-Aware Splitting and Weight Merging
👍 7
Instruction tuning aligns large language models, including multimodal ones, with diverse user intents, but scaling to heterogeneous mixtures is hindered by gradient interference and bandwidth-heavy synchronization. We ask whether these two bottlenecks can be addressed jointly by training parts of th
中文介绍 去中心化指令微调通过冲突感知分割与权重合并，解决混合异构数据时梯度干扰与带宽同步瓶颈，提升多模态大语言模型对齐效果。
PlatonicNav: Unveiling Semantic Correspondence in Navigation with Platonic Topological Maps
👍 5
Embodied visual navigation, where an agent perceives a complex environment and acts to reach a goal from raw sensory input, underpins a wide range of applications such as household service robotics, assistive robotics, and large-scale autonomous exploration. However, recent attempts to unify vision-
DOT-MoE: Differentiable Optimal Transport for MoEfication
👍 0
The scaling of Large Language Models (LLMs) has driven significant performance gains but created substantial challenges in inference efficiency. While Mixture of Experts (MoEs) architectures address this by decoupling model size from inference cost, training MoEs from scratch is often unstable and c
TVIR: Building Deep Research Agents Towards Text--Visual Interleaved Report Generation
👍 3
Deep Research Agents have shown strong capability in multi-step information retrieval, reasoning, and long-form report generation, but existing benchmarks and systems remain predominantly text-centric, with limited evaluation of whether visual elements are factually reliable and well aligned with th
Harness-1: Reinforcement Learning for Search Agents with State-Externalizing Harnesses
👍 34
Search agents are often trained as policies over growing transcripts: the model must decide how to search while also remembering what it has seen, which evidence is useful, which constraints remain open, and which claims have actually been checked. We argue that this formulation puts too much routin
Policy and World Modeling Co-Training for Language Agents
👍 8
Reinforcement learning (RL) improves large language model (LLM) agents by teaching them which actions lead to high rewards, but provides little supervision on what those actions do to the environment. World modeling (WM) can fill this gap, yet existing approaches often require separate simulators, e
ClawHub Security Signals: When VirusTotal, Static Analysis, and SkillSpector Disagree
👍 2
Agent skills extend AI agents with reusable instructions, tools, scripts, references, and workflows, establishing a security boundary distinct from both model safety and traditional package-malware detection. ClawHub Security Signals is a sanitized dataset of 67,453 latest public OpenClaw skill vers
Trust Region On-Policy Distillation
👍 29
On-Policy Distillation (OPD) is a fundamental technique for efficient post-training of large language models (LLMs), with broad applications in agent learning, multi-task enhancement, and model compression. However, OPD training becomes unstable when the teacher and student distributions differ subs
OCC-RAG: Optimal Cognitive Core for Faithful Question Answering
👍 14
Recent progress in the development of language models has been defined by scale, with each generation absorbing more of the world's knowledge into its weights. However, many practical applications benefit more from robust reasoning than from extensive parametric knowledge. In this setting, task-spec
MindZero: Learning Online Mental Reasoning With Zero Annotations
👍 2
Effective real-world assistance requires AI agents with robust Theory of Mind (ToM): inferring human mental states from their behavior. Despite recent advances, several key challenges remain, including (1) online inference with robust uncertainty updates over multiple hypotheses; (2) efficient reaso
SVI-Bench: A Dynamic Microworld for Strategic Video Intelligence
👍 4
True video intelligence demands more than recognizing what is visible: it requires reasoning about why events unfold, predicting what would change under different conditions, and deciding what to do next. We refer to this progression, from perception through causal reasoning and simulation to strate
Model-Based Quality Assessment for Massively Multilingual Parallel Data
👍 1
Large-scale multilingual bitext often contains two distinct problems: non-parallel sentence pairs and low-quality translations. We decompose model-based assessment for such data into two independent components: parallelism assessment with multilingual embeddings and reference-free quality estimation
A Multi-AI-agent Framework Enabling End-to-end Finite Element Analysis for Solid Mechanics Problems
👍 1
Finite element analysis (FEA) is the most important numerical approach for solid mechanics. Challenges of FEA include a steep learning curve for entry-level users and potential false simulations due to incorrect definitions of key simulation components, such as boundary conditions, load cases, and s
Diagnosing Harmful Continuation in Answer-Correct Long-CoT Training Traces
👍 6
Long chain-of-thought (CoT) traces are widely used as supervision for reasoning-oriented LLM SFT, yet answer-correct traces can still lead to markedly different fine-tuning outcomes. We study post-conclusion continuation in answer-correct long-CoT data: a continuation where the answer appears suffic
Pressure-Testing Deception Probes in LLMs: Scaling, Robustness, and the Geometry of Deceptive Representations
👍 0
Linear probes trained on LLM activations are increasingly proposed as deception-detection metrics, yet report AUROC exceeding 0.96 on clean benchmarks while collapsing under distributional shift. This paper systematically pressure-tests probe-based metrics across the Gemma 3 model family (1B-27B par
Same Question, Different Source, Different Answer: Auditing Source-Dependence in Medical Multi-Source RAG
👍 1
A retrieval-augmented generation (RAG) system deployed over a multi-author institutional corpus can give a different answer to the same question depending on which source it retrieves -- a failure mode the dominant single-gold-answer paradigm cannot diagnose. We argue that source-dependence is a mis
The Chain Holds, the Answer Folds: Trace-Answer Dissociation in Reasoning Models Under Adversarial Pressure
👍 1
Reasoning models are evaluated on single-turn benchmarks but deployed in multi-turn dialogue, where users push back on correct answers. Under sustained adversarial pressure we find a previously undocumented failure mode: the chain-of-thought stays factually correct from first turn to last while the
MERIT: Learning Disentangled Music Representations for Audio Similarity
👍 5
Current music similarity models typically compute a single, monolithic score, entangling distinct musical dimensions like melody, rhythm, and timbre. This limits user control and interpretability, making it impossible to execute nuanced queries. We introduce MERIT, a framework for learning disentang
From Activation to Causality: Discovery of Causal Visual Representations in the Human Brain
👍 13
Identifying which brain regions represent a visual concept in the human brain is a central challenge in neuroscience. Existing approaches have localized coarse functional regions (e.g., faces, places) through activation maximization, identifying regions that activate strongly for a target concept re
HOW ONE $2,999 NVIDIA BOX MADE ME $22,000 IN A YEAR
@w1nklerr · 44.2K 粉丝 · 17.7M 阅 · 1.4K 赞 · 161 转
Nobody told me about this for months. I'm telling you now so you don't lose the year I lost. Let me start with the number that made me angry. Last quarter my cloud GPU spend was sitting at $1,900 a
中文介绍 用 2999 美元自购 NVIDIA 设备替代云 GPU，一年省下 22000 美元。博主揭示云 GPU 隐性成本，分享自建算力的真实收益与经验，适合高算力需求者参考。
Context as Topology: Why Your Agent's Memory Forgets, and How Structure Escapes It
@elpresidank · 116 粉丝 · 2.9M 阅 · 543 赞 · 35 转
Most AI agent memory is built on embeddings. And there's now a proof that this entire class of system is going to forget what you stored in it — and confidently make up things you never stored at all.
中文介绍 论证当前基于 embedding 的 AI 记忆系统存在先天缺陷——会遗忘并捏造信息。提出用拓扑结构替代线性存储来突破记忆瓶颈，为 Agent 记忆设计提供新思路。
Range and Depth on Demand
@1salman · 363 粉丝 · 2.0M 阅 · 682 赞 · 45 转
Everyone keeps asking whether AI favors specialists or generalists. I think that is the wrong question. AI does not pick a side. It changes the tradeoff. The old world forced a choice. You could go
中文介绍 AI 并未偏袒专才或通才，而是改变了「深度与广度」的权衡方式。过去被迫二选一，现在 AI 让人可以按需同时获得范围与深度。
How to build a 4-agent team, that ships a feature while you sleep (Exact Setup Inside)
@zodchiii · 20.0K 粉丝 · 743.3K 阅 · 509 赞 · 55 转
Four AI agents can ship a feature while you sleep. Most people never wire them up. They fire a reviewer here, a test generator there, by hand, one at a time, each forgetting what the last one did.
中文介绍 分享一个 4 个子 Agent 编写、审查、测试、部署的自动化团队搭建教程。从零串联协作流程，让 Agent 在你睡觉时完成功能交付，附完整设置指南。
30 Obsidian Workflows, Plugins, and Setups That Most Users Don't Know
@eng_khairallah1 · 61.9K 粉丝 · 693.5K 阅 · 511 赞 · 71 转
Obsidian has 2,700+ community plugins. Over 100 of them are AI-related. Save this :) And the CEO of Obsidian personally published official Claude Skills for the platform - 12,900+ GitHub stars in
中文介绍 整理 30 个鲜为人知的 Obsidian 工作流、插件与配置，包括 AI 相关插件和 CEO 官方发布的 Claude Skills（获 12900+ GitHub star），提升笔记与 AI 整合效率。
What an Enterprise Context Layer Actually Is
@prukalpa · 23.1K 粉丝 · 583.2K 阅 · 506 赞 · 80 转
A field guide to what it is, what it is not, and where it fits in your AI architecture. I have had some version of the same conversation with a CIO almost every day this year. Their team has read
中文介绍 定义企业 AI 架构中的「上下文层」：它是什么、不是什么、放在哪。针对 CIO 常见困惑，提供实际部署指南，区分上下文层与向量数据库、RAG 等概念的边界。
I Searched the Whole Claude Skills Ecosystem - These Are the Ones That Matter [Full GitHub Links]
@polydao · 18.1K 粉丝 · 559.5K 阅 · 505 赞 · 55 转
Most people are still using Claude like a smarter chatbot That is not the game anymore You’re competing against people who treat Claude like an operating system > While you’re typing one-off
中文介绍 筛选整个 Claude Skills 生态中有价值的技能，给出 GitHub 链接。强调 Claude 应作为操作系统而非聊天机器人使用，提供核心 Skills 清单帮助用户切换工作模式。
hacking pewdiepie's AI agent harness using an evil cocomelon website (then helping protect it)
@theonejvo · 22.1K 粉丝 · 504.3K 阅 · 861 赞 · 1 转
Over the past year, @pewdiepie, has been turning into one of the most visible champions of private, self-hosted computing, and it has been a genuine pleasure to watch. What began in late 2025 as an
中文介绍 通过恶意 Cocomelon 网站攻破了 @pewdiepie 自建 AI Agent 的安全防护，并在之后协助其修复漏洞。案例展示自托管 Agent 存在的攻击面及防御实践。
Avoiding Death on the Yellow Brick Road
@joeschmidtiv · 11.8K 粉丝 · 487.0K 阅 · 592 赞 · 59 转
Why The App Layer Isn't Dead The question I keep getting from founders and prospective employees: is there any AI application layer left to build, or are OpenAI and Anthropic going to kill
中文介绍 回应「AI 应用层已死」论，认为底层模型 vs 应用的旧竞争框架已过时。应用层仍有生存空间，关键在于避开被巨头驱赶的路线，找到差异化价值。
Claude Code + NotebookLM + Obsidian: Research Monster That Gets Smarter Every Time You Use It
@monokern · 1.2K 粉丝 · 263.1K 阅 · 505 赞 · 72 转
Most people treat research as a manual task. You open 10 tabs. You watch videos. You read articles. You take notes somewhere. An hour later you have a pile of information you're not sure what to do
中文介绍 整合 Claude Code + NotebookLM + Obsidian，搭建一个越用越智能的研究工作流。突破传统手动搜索、记笔记的低效模式，实现半自动化深度研究。
Stop building Foxconn factories for your agents
@garrytan · 853.3K 粉丝 · 180.6K 阅 · 503 赞 · 43 转
In January I got back into coding and I built Garry's List. Over five hundred thousand lines of Rails and the tests to police it. I was proud of it. I shouldn't have been. The thing worth being proud
中文介绍 Garry Tan 分享亲身教训：不该为 Agent 搭建庞大工厂（Foxconn 模式），而应专注更巧妙的架构。以自己 50 万行 Rails 项目为例，说明过度工程反而是负担。
The Agentic Economy Is Here
@base · 1.3M 粉丝 · 97.3K 阅 · 519 赞 · 74 转
TL;DR: Agents are becoming the internet’s newest paying customers, and the economy serving them is moving fast. On Base, agents already use wallets and stablecoins to pay for inference, live search,
中文介绍 宣布 Agent 正成为互联网新一代付费客户。在 Base 链上，Agent 已能用钱包和稳定币支付推理、搜索等费用，经济体系正在向 Agent 服务化迁移。
🥇Top AI Papers of the Week
@dair_ai · 124.6K 粉丝 · 84.0K 阅 · 504 赞 · 83 转
1. SkillOpt Microsoft Research treats a compact natural-language skill document as the trainable state of a frozen agent, then learns that document through rollouts, reflection, and bounded edits
中文介绍 本周最佳 AI 论文综述：1) 微软 SkillOpt 以自然语言技能文档作为可训练状态，通过回滚与反思优化 Agent；其余论文偏向 Agent 学习与系统设计。
My Agent Stack For Automating My Personal Life
@nicbstme · 23.7K 粉丝 · 84.0K 阅 · 530 赞 · 35 转
My agent manages my emails, SMS, Whatsapp, Telegram and pretty much everything to automate my personal life. People keep asking me how I use agents in real life. I mean the actual boring things that
中文介绍 分享个人 Agent 自动化栈：管理邮件、短信、WhatsApp、Telegram 等日常通信，真正把 Agent 用于生活琐事，而非炫技。提供实际搭建经验。
Robotics: The Next AI Frontier
@ParadisLabs · 48.9K 粉丝 · 82.0K 阅 · 501 赞 · 60 转
AI's next frontier will be Robotics and Humanoids. The past decade has seen rapid AI adoption in the structured digital world. Those LLM breakthroughs now enable more general-purpose learning and more
中文介绍 AI 下一前沿是机器人和人形机器人。LLM 突破带来了通用学习能力的跃升，从结构化数字世界延伸到物理世界，解读技术路径与投资方向。
How to build your own agent harness???
@mfpiccolo · 7.4K 粉丝 · 81.9K 阅 · 607 赞 · 56 转
Most agent teams don't build a harness. They adopt one. LangChain, LangGraph, OpenAI Agents SDK, Anthropic SDK, CrewAI, AutoGen, the loop, the tools, the memory, and the orchestration are picked off
中文介绍 多数 Agent 团队直接拿现成框架（LangChain、OpenAI SDK、CrewAI 等），但博主主张自建 Agent 束缚层（harness）。对比主流框架优缺点，教你怎么从零搭建自制 harness。
A harness for every task: dynamic workflows in Claude Code
@trq212 · 263.1K 粉丝 · 75.7K 阅 · 542 赞 · 36 转
Last week, we released dynamic workflows in Claude Code. Claude can now write its own harness on the fly, custom-built for the task at hand. While the default Claude Code harness is built for coding,
中文介绍 Claude Code 推出新动态工作流功能：Claude 能为不同任务自动编写专属 harness，无需硬编码。对比默认 harness，灵活性和任务适配度大幅提升。
A harness for every task: dynamic workflows in Claude Code
@trq212 · 263.1K 粉丝 · 75.7K 阅 · 7d 曝光 75.7K
A harness for every task: dynamic workflows in Claude Code
Every Agentic Engineering Hack I Know (June 2026)
@mvanhorn · 27.6K 粉丝 · 54.5K 阅 · 7d 曝光 54.5K
Every Agentic Engineering Hack I Know (June 2026)
X Just Got Its TikTok Moment. It's Called Commentary.
@subahwadhwani · 5.1K 粉丝 · 355.9K 阅 · 7d 曝光 355.9K
X Just Got Its TikTok Moment. It's Called Commentary.
Context as Topology: Why Your Agent's Memory Forgets, and How Structure Escapes It
@elpresidank · 116 粉丝 · 2.9M 阅 · 7d 曝光 2.9M
Context as Topology: Why Your Agent's Memory Forgets, and How Structure Escapes It
中文介绍 论证当前基于 embedding 的 AI 记忆系统存在先天缺陷——会遗忘并捏造信息。提出用拓扑结构替代线性存储来突破记忆瓶颈，为 Agent 记忆设计提供新思路。
What an Enterprise Context Layer Actually Is
@prukalpa · 23.1K 粉丝 · 583.2K 阅 · 7d 曝光 583.2K
What an Enterprise Context Layer Actually Is
中文介绍 定义企业 AI 架构中的「上下文层」：它是什么、不是什么、放在哪。针对 CIO 常见困惑，提供实际部署指南，区分上下文层与向量数据库、RAG 等概念的边界。
hacking pewdiepie's AI agent harness using an evil cocomelon website (then helping protect it)
@theonejvo · 22.1K 粉丝 · 504.3K 阅 · 7d 曝光 504.3K
hacking pewdiepie's AI agent harness using an evil cocomelon website (then helping protect it)
Stop building Foxconn factories for your agents
@garrytan · 853.3K 粉丝 · 180.6K 阅 · 7d 曝光 180.6K
Stop building Foxconn factories for your agents
把一本书做成 AI skill，挂闲鱼 ¥19.9、小红书 ¥99~199——保姆级教程，全部开源直接抄！
@AYi_AInotes · 48.5K 粉丝 · 196.7K 阅 · 7d 曝光 196.7K
把一本书做成 AI skill，挂闲鱼 ¥19.9、小红书 ¥99~199——保姆级教程，全部开源直接抄！
6 Workflows, 6 Lessons, 60 Days with Hermes Analyst
@0xJeff · 80.5K 粉丝 · 47.0K 阅 · 7d 曝光 47.0K
6 Workflows, 6 Lessons, 60 Days with Hermes Analyst
Range and Depth on Demand
@1salman · 363 粉丝 · 2.0M 阅 · 7d 曝光 2.0M
Range and Depth on Demand
中文介绍 AI 并未偏袒专才或通才，而是改变了「深度与广度」的权衡方式。过去被迫二选一，现在 AI 让人可以按需同时获得范围与深度。
🥇Top AI Papers of the Week
@dair_ai · 124.6K 粉丝 · 84.0K 阅 · 7d 曝光 84.0K
🥇Top AI Papers of the Week
The Latest Codex Updates and The Truth about Opus 4.8
中文介绍 介绍 Codex 的最新更新内容，并讨论 Opus 4.8 的实际表现与真相。
Browsers Are Dead. Codex Just Replaced Them.
Every NEW Claude & Codex Feature Explained
Codex: Build Your Full AI Marketing Team (Agents + Skills)
OpenAI just released Codex Mobile
Codex Mobile App Released (Complete Setup Guide)
Vibe Coding for Beginners (Full Course 2026)
Codex is The NEW Best AI Coding Tool (Here's Why)
7 Tools That Make Codex 10x MORE Powerful
Learn 95% of Codex in 30 minutes
Team thinking, visualized by Claude
中文介绍 Claude 展示了团队思考过程的可视化演示，呈现 AI 如何协助团队协作与思维整理。
The Problem Solvers | Max Junestrand at Legora
中文介绍 Legora 的 Max Junestrand 在节目中探讨问题解决的方法与经验，分享个人见解。
Before we ship a Claude model, these teams try to break it.
中文介绍 Claude 团队在发布模型前会安排专门团队对其进行压力测试和漏洞检测。
Embrace long-running tasks with Opus 4.8 and Claude Code
中文介绍 Claude 推出 Opus 4.8 和 Claude Code，支持长时间运行任务，优化工作流程。
The Problem Solvers: Michele Catasta at Replit
Ship your first Managed Agent
Tool, skill, or subagent? Decomposing an agent that outgrew its prompt
Agent Battle: Mine the most diamonds in 45 minutes
Evals for taste: Hill-climbing a slide-generation agent
Agents that remember
Team thinking, visualized by Claude
中文介绍 Claude 展示了团队思考过程的可视化演示，呈现 AI 如何协助团队协作与思维整理。
The Problem Solvers | Max Junestrand at Legora
中文介绍 Legora 的 Max Junestrand 在节目中探讨问题解决的方法与经验，分享个人见解。
Before we ship a Claude model, these teams try to break it.
中文介绍 Claude 团队在发布模型前会安排专门团队对其进行压力测试和漏洞检测。
Embrace long-running tasks with Opus 4.8 and Claude Code
中文介绍 Claude 推出 Opus 4.8 和 Claude Code，支持长时间运行任务，优化工作流程。
The Problem Solvers: Michele Catasta at Replit
Ship your first Managed Agent
Tool, skill, or subagent? Decomposing an agent that outgrew its prompt
Agent Battle: Mine the most diamonds in 45 minutes
Evals for taste: Hill-climbing a slide-generation agent
Agents that remember
A Second Nobel Prize for AlphaFold? 🧬🏆 #alphafold #deepmind #nobelprize #science #ai
中文介绍 讨论 AlphaFold 是否可能因其在蛋白质结构预测方面的贡献再次获得诺贝尔奖。
What Happens After A 1,000,000x AI Compute Leap? | Jeff Dean
中文介绍 Jeff Dean 探讨 AI 计算能力提升百万倍后可能带来的变革和未来发展。
Einstein vs Feynman, Who Wins? 🧠🤔 #physics #ai #science #feynman #research
中文介绍 比较爱因斯坦和费曼在物理学领域的贡献与智慧，探讨谁更胜一筹。
Google DeepMind CEO Loves Hard Questions 🙂
Demis Hassabis On What AI Will Do Next
DeepSeek’s New AI Is A Game Changer
NVIDIA New AI Is An Efficiency Monster
OpenAI's ChatGPT 5.5 Instant: The Good, The Bad And The Insane
DeepSeek V4 AI Beats Billion Dollar Systems…For Free
NVIDIA's New AI Turns One Photo Into A World That Never Breaks
Direct Preference Optimization Beyond Chatbots
中文介绍 文章探讨直接偏好优化（DPO）在聊天机器人外的应用，强调其在训练语言模型时对齐人类偏好的扩展潜力。
[AINews] Microsoft Build: MAI-Thinking-1 and MAI Family models
Microsoft Build recap, and new MAI model technical details
中文介绍 微软Build大会发布MAI-Thinking-1及MAI系列模型，细节尚未完全公开，但标志其AI模型布局重要进展。
Running an AI-native engineering org
Running an AI-native engineering org
中文介绍 探讨如何运行AI原生工程组织，分享管理经验与最佳实践。
Codex Sites 💻, Microsoft models 🤖, Anthropic cost backlash 💸
中文介绍 报道Codex网站、微软新模型、Anthropic成本争议等AI领域最新动态。
GitHub's plan for Agents — Kyle Daigle, GitHub
GitHub pioneered the modern AI coding era with Copilot, and the resulting explosion in agentic coding has led to notable strains on the most popular developer platform in the world. Here's the plan.
中文介绍 GitHub高管Kyle Daigle分享GitHub在AI编码时代推动智能体计划的战略，应对开发平台压力。
Holo3.1: Fast & Local Computer Use Agents
中文介绍 介绍Holo3.1，一个快速且本地的计算机使用智能体模型。
Travelers deploys AI-powered claims countrywide with OpenAI
Travelers built an AI-powered Claim Assistant with OpenAI to guide customers through filing claims, provide 24/7 support, and scale operations during peak demand.
中文介绍 Travelers使用OpenAI构建AI理赔助手，提供全天候指导，提升高峰期的运营扩展能力。
Rehumanizing global health care with agentic AI
The global health care sector is under increasing strain. Decades of chronic underinvestment and constraints in recruitment have coincided with a surge in demand for services for aging populations. Gaps in provision are already taking a toll, with fragmented access to care and high rates of stress a
中文介绍 全球医疗体系面临压力，文章探讨利用智能体AI改善医疗可及性和人性化服务，应对人员短缺。
Codex for every role, tool, and workflow
Discover new Codex plugins, sites, and annotations that help analysts, marketers, designers, investors, and other teams get more done with AI.
中文介绍 OpenAI发布Codex新插件、网站和注释功能，面向分析师、设计师等角色，帮助提升AI驱动的工作效率。
How small businesses can leverage AI
This article is from Making AI Work, MIT Technology Review’s limited-run newsletter examining how to apply LLMs across industries. To receive it in your inbox,sign up here. From accounting to design to market research and product development, there’s a staggering breadth of skills needed to run a bu
中文介绍 文章介绍小企业如何利用AI，涵盖会计、设计、市场调研等领域，强调应用广度。
Advancing youth safety and opportunity through global leadership
OpenAI calls for global action on youth AI safety, proposing an international institute to strengthen safeguards, standards, and opportunities for young people.
中文介绍 OpenAI呼吁全球行动加强青年AI安全，提议建立国际机构以制定保障标准和机会。
Microsoft Build: MAI-Thinking-1 and MAI Family models, Surface RTX Spark Dev Box, and OpenClaw in Windows
**Microsoft** introduced **MAI-Thinking-1**, a **35B parameter MoE model** with **256K context**, achieving **97% on AIME 2025** and outperforming **Sonnet 4.6** in human preference tests. The broader **7-model MAI family** spans reasoning, code, image, speech, and voice, with third-party availabili
中文介绍 微软发布MAI-Thinking-1，35B参数MoE模型，256K上下文，AIME 2025达97%，超越Sonnet 4.6。MAI家族覆盖推理、代码、图像等领域。
[AINews] NVIDIA Cosmos 3, Nemotron 3 Ultra, and RTX Spark
Jensen scores a huge win.
中文介绍 NVIDIA发布Cosmos 3、Nemotron 3 Ultra及RTX Spark，Jensen Huang取得重大胜利。
Codex is becoming a productivity tool for everyone
The Next Era of Knowledge Work report explores how Codex is transforming productivity through AI-powered research, data analysis, workflow automation, and content creation.
中文介绍 OpenAI发布报告《知识工作的下一时代》，探讨Codex通过AI研究、数据分析和自动化提升生产力。
A harness for every task: dynamic workflows in Claude Code
A harness for every task: dynamic workflows in Claude Code
中文介绍 介绍Claude Code中动态工作流，适配多种任务，提升开发效率。
求助帖：搭车干净🪜机场
6 回复 · 程序员 节点
Github 仓库被轰炸
7 回复 · 程序员 节点
要领大礼包了，看看工作发现有家公司给 V2EX 写到了 jd 里
6 回复 · 程序员 节点
RackNerd 什么情况??
7 回复 · 程序员 节点
请教下 mac 上指定窗口高度和大小的软件
10 回复 · Apple 节点
Copilot 更改计价模式后的编码模型选择
12 回复 · 程序员 节点
卧槽，用了 10 多年的 github 账号突然被封了，各位老哥怎么办？
19 回复 · 程序员 节点
vibe coding 了一个高仿的 mac 端的 paste ，完全免费。
7 回复 · Apple 节点
还有几天就是苹果 WWDC26 全球开发者大会了，怎么没看到相关新闻，还有下一代 iOS27 更新了什么内容，好像也没见媒体报道了？还是说因为我关注的少了不给我推了？
23 回复 · Apple 节点
玄学？为啥土区支付方式不能改为 None？
17 回复 · Apple 节点
情侣旅行记忆地图（地图 + 时间线 + 回忆）
Map of Intl 是一个属于两个人的旅行记录应用。 最开始只是想解决一个很简单的问题： 旅行结束后，照片越来越多，但回忆却越来越难整理。 于是我尝试用 AI 和 Vibe Coding 的方式，做了一个能够记录共同生活的小程序。 它将地图、时间线、旅行计划与回忆整理融合在一起，把一起去过的城市、走过的路线、拍下的照片和重要纪念日串联成完整的故事。 随着记录越来越多，地图上的足迹会不断增长，时间线也会越来越长。 后来发现，它记录的不只是旅行。 而是两个人一起走过的生活。 希望多年以后再次打开时，依然能够看见那些共同经历过的风景和时光。 可自由编辑路线 可查看旅行日记 接入高德地图API 3
claude这是你不仁
这是你不仁别怪我不义。我是你忠实粉丝，就这样回报我。你彻底成功的让我对你失望透顶 我这么纯都不行 既然你很喜欢发验证码是吧，那我就写个脚本让你全国发，不要停 28 个帖子 - 25 位参与者 阅读完整话题
codex 自动化分析 A 股收盘日报
如果有特别想关注的，可以在提示词中说的更清楚一些，codex拿到的数据还是蛮完善的，我现在就是每天看一下，太忙，最近都空仓着，没咋关注这些。 提示词（可根据需求自行修改，我样式上做了一些细微的调整，不好总结出来了[token 我没算过，我开的pro20x 会员，这个具体会跑多少token，我没细看过，记得注意哈，万一撑不住就不好了]）： 你的任务是：在每个 A 股交易日收盘后，自动生成一份完整的《A股收盘日报 HTML 页面》。 生成时间建议：北京时间每个交易日 15:30 - 18:30。 重要要求： 最终结果必须是一个完整可打开的 HTML 文件。 HTML 页面必须包含完整 CSS 样式
君の公益 回来吧，天才程序员们~
gpt 恢复供应 大家尽量不要给我发私信，有问题评论就行，无力回复，太多了 不要以任何理由索要额度/寻求解封 用不了就是号池在维护 不要去星辰的售后群去问我的客服为啥君の公益用不了，没有别的意思，就是客服可能会比较崩溃 我不会建任何有关君の公益的qq群/tg群，我也不喜欢看到有人以我的名义建群 君の公益唯一的解释权在我这里，我不喜欢看到有人代替我解释什么，希望大家尽量避免 签到额度调整为每天52刀 已经安排佬友帮我查分发了，已经封了一批，后面不定期也会安排检查 285 个帖子 - 253 位参与者 阅读完整话题
RawChat的codex公益站，注册就送100刀额度
RawChat公益，codex每人每天赠送100刀额度 https://new.sharedchat.cc ccstwich配置方式： 公益站有非常完善的风控机制，请勿重复注册，否则可能会被连坐封号 qq群：1075430606（有问题群内反馈回复会快点） 56 个帖子 - 48 位参与者 阅读完整话题
我成为领导眼中成了啥都会的人，我是该笑还是该哭呢？
今天我的一个同事突然来找我，给我看了一个PPT说老板让你把这个ppt做成视频，我就是菜鸡，哪里会做视频啊？求教万能的论坛大佬们，我需要怎么做啊？ 105 个帖子 - 73 位参与者 阅读完整话题
visa卡了办了，Giffgaff卡了买了，居然卡在这个地方了
打算买土区的gpt plus会员，所有都准备好了，手机版本太低，下载不了chatgpt，这不天塌了嘛！我得是苹果13，版本16.0.3，这个版本挺省电的，更新不炸了吗？难道还要换手机？ 114 个帖子 - 67 位参与者 阅读完整话题
记录首次AI漫剧的创作
刚刚完成了一个1分多种的AI动漫短剧，更见坚定了我更愿意成为一个打磨脚本的创作者的想法！！ 第二步：生成人物的三视图，是为了放崩溃 第三步：撰写AI故事脚本 第四步：生成比较火的九宫格图片（我生成的是带首尾帧，但实际上并不需要，反而到seedance2.0会被限制发挥，我建议直出） 第五步：因为seedance2.0最多只能生成15s视频，直接参考九宫格图，不好约束，所以这让我有了新的思考，也许九宫格图更适合15s直出的案例，我经验有限，请佬指正 第六步：最初我选择seedance2.0选择首尾帧模式，因为之前生成的都是带首尾帧的图片，是为了更好的约束AI,我@引用图片的左侧是首帧、右侧是尾帧
技术转销售做了三周，辞职了，给大家说点自己的体验
今天上午提的离职，现在已经在家里重新开始了，先介绍一下我自己的一个情况。 我23年双非本科毕业，在广州做了3年的游戏，最后因为项目没起来，加上一些自己在公司陷入了一点办公室政治，所以辞职了。以为自己可以动力满满开发一个自己的游戏，但是发现自己搞的话没人可以当面交流灵感，也没人监工提醒自己，然后就从2月浑浑噩噩到了5月左右。 因为自己是byd粉丝，一直很想买byd的宋pro，从大学看到现在了，然后就想既然做不动游戏，不如买个车去环一下国内旅游，然后找个班上吧。 然后买了车之后，发现自己的存款还不足以旅游，只能坚持3-4月的开销，虽然也有一点家里支持，但是我这个人就是不想让家里出钱的，因为家里给的
恭请佬友们前来观礼
早岁入站知规艰，仍许潜龙灌水间。 43 个帖子 - 35 位参与者 阅读完整话题
Which sparkling water is the best?
9 points · 1 comments
32GB of DDR5 now costs $375 – AI shortage continues to squeeze PC building
28 points · 10 comments
Meta workers can opt out of being tracked at work up to 30 min
13 points · 1 comments
Uber's $1,500/Month AI Limit Is a Useful Signal for AI Tool Pricing
19 points · 11 comments
Piramidal (YC W24) – Software Engineers – NYC Onsite
We're looking for senior software engineers - frontend, backend, and infra - who are excited about solving difficult problems at the boundary of Foundational AI and Neuroscience. Our core backend and infra stack is Python, Go, and Terraform deployed to various cloud and on-prem, while our front
Every Byte Matters
91 points · 21 comments
Hacking your PC using your speaker without ever touching it
212 points · 42 comments
What I've learned about the trombone
14 points · 9 comments
PlayStation Architecture
67 points · 10 comments
Leiden Declaration on Artificial Intelligence and Mathematics
43 points · 13 comments
The Unreasonable Redundancy of Nature's Protein Folds
126 points · 36 comments
U of T researchers demonstrate AI worm could target any online device
77 points · 20 comments
Capstone – multi-platform, multi-architecture disassembly framework
76 points · 3 comments
Roku LT Operating System open source distribution
94 points · 31 comments
AI outperforms law professors in Stanford Law study
https://law.stanford.edu/wp-content/uploads/2026/06/salinas_...
Pluto.jl 1.0 release – reactive notebook for Julia
170 points · 22 comments
Use your Nvidia GPU's VRAM as swap space on Linux
366 points · 98 comments
CT scans of BYD car parts
433 points · 285 comments
My thoughts after using Clojure for about a month
251 points · 124 comments
HP re-releases classic computer science calculator: The HP-16C
189 points · 118 comments
Show HN: Edsger – A handwritten Clojure REPL for the reMarkable 2
115 points · 23 comments
MAI-Code-1-Flash
https://microsoft.ai/models/mai-code-1-flash/https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDFLaunching seven new MAI models: https://microsoft.ai/news/building-a-hillclimbing-machine-la...
How we index images for RAG
172 points · 23 comments
1-Click GitHub Token Stealing via a VSCode Bug
512 points · 73 comments
Turkey Hacked the Hair Transplant Industry
26 points · 13 comments
Nabokov's pale fire: the lost 'father of all hypertext demos'? (2011)
41 points · 5 comments
Words of Type
96 points · 12 comments
Writing Portable ARM64 Assembly (2023)
40 points · 15 comments
DIY Bipedal Robot Used Pneumatic "Air-Muscles" Instead of Motors
43 points · 11 comments
Show HN: I reverse-engineered the world maps of Test Drive III (1990 DOS game)
150 points · 38 comments