TLDR AI
Anthropic 推出 Claude Opus 4.8，在协作与编码方面表现提升，但文档解析出现部分退化；同时 Anthropic 估值达 9650 亿美元，微软也发布了新编码模型。（多家报道）
推荐理由：Opus 4.8 是主要模型迭代，反应行业最新水平，且 Anthropic 估值逼近万亿，投资者与从业者应关注。
2026-05-31（内容获取于 05/31 19:05）
Claude Blog
Claude Code 新增动态工作流功能，旨在提升 AI 编码智能体在复杂任务中的灵活性与可控性。
推荐理由：动态工作流可显著提升开发者使用 AI 编码代理的效率与体验，建议体验。
Anthropic Engineering
随着 AI 智能体能力增强，潜在影响范围扩大；Anthropic 公开了在 Claude.ai、Claude Code 和 Cowork 中构建安全隔离的工程经验。
推荐理由：深度工程复盘，对 AI 安全工程师和产品负责人具有直接借鉴意义。
GitHub Trending
微软推出的 Python 工具，可高效将各类办公文档转换为 Markdown 格式，便于 AI 与文档处理场景。
推荐理由：开源、低门槛，文档处理与 AI 预处理的绝佳实用工具，开发者和数据工作者可直接使用。
GitHub Trending
利用 AI 大模型一键生成高清短视频的开源项目，降低内容创作门槛。
推荐理由：上手即用，适合自媒体创作者快速产出视频内容，工具化价值高。
OpenAI News
波士顿儿童医院借助 OpenAI 技术提升患者护理效率，减少运营负担，并成功帮助诊断超过 40 个罕见病病例。
推荐理由：展示 AI 在医疗领域的深度应用与积极成果，对医疗科技从业者有启发。
Hugging Face Blog
Hugging Face 发布 torch.profiler 初学者教程，帮助开发者分析和优化 PyTorch 模型性能。
推荐理由：实用的性能调优教程，开发者可立即上手提升训练效率。
Anthropic Research
Anthropic 发布经济研究报告，探讨编码智能体在社会科学研究中的应用潜力与风险。
推荐理由：为社科研究者提供 AI 智能体应用的新思路，非技术类读者也能理解。
MIT Tech Review AI
教皇 Leo XIV 发布关于人工智能的新通谕「Magnifica Humanitas」，强调技术非中立，呼吁科技界与政策制定者关注 AI 的伦理维度。
推荐理由：提供 AI 伦理与人文视角，引发深度思考，适合关注 AI 社会影响的读者。
HuggingFace Trending Papers
新论文研究视觉语言模型的空间推理能力，发现其可能基于统计捷径而非真正的 3D 理解。
推荐理由：有助于理解当前模型的局限性，对研究人员具有参考价值。
microsoft/markitdown
Python · ★ 133,615 · 🍴 9,135 · 📈 2,470 stars today
Python tool for converting files and office documents to Markdown.
中文介绍 Python 工具，用于将各类文件（包括 Office 文档）转换为 Markdown 格式。解决非结构化文档向轻量标记语言转化的问题，适用于文档迁移、内容提取和知识整理场景。开发者、数据工作者可快速批量处理 Word、Excel、PPT 等文件。
harry0703/MoneyPrinterTurbo
Python · ★ 72,913 · 🍴 10,416 · 📈 2,768 stars today
利用AI大模型，一键生成高清短视频 Generate short videos with one click using AI LLM.
中文介绍 基于 AI 大模型，一键生成高清短视频。用户只需输入文案，系统自动完成配音、配图、剪辑等流程，大幅降低视频制作门槛。适合内容创作者、营销人员快速产出宣传或科普类短视频。
anthropics/claude-code
Python · ★ 128,599 · 🍴 20,965 · 📈 592 stars today
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.
中文介绍 终端内的智能编码助手，能够理解代码库并执行日常任务、解释复杂代码、处理 git 工作流。集成 Claude 的 agentic 能力，帮助开发者提升编码效率，减少重复劳动。适用于快速代码导航、重构、代码审查等场景。
cursor/plugins
TypeScript · ★ 1,548 · 🍴 122 · 📈 205 stars today
Cursor plugin specification and official plugins
中文介绍 Cursor 编辑器的插件规范与官方插件集合。为基于 AI 的代码编辑器提供扩展能力，允许开发者自定义功能、集成第三方服务。适用于需要增强 Cursor 功能的前端、后端开发者，通过插件生态实现工作流定制。
revfactory/harness
HTML · ★ 4,404 · 🍴 633 · 📈 55 stars today
A meta-skill that designs domain-specific agent teams, defines specialized agents, and generates the skills they use.
中文介绍 元技能框架，用于设计和生成领域专属的 AI agent 团队。用户可定义专项 agent 及其技能组合，实现复杂任务的自动化编排。适用于需要多 agent 协作的场景，如软件开发、数据分析、客户服务等。
EveryInc/compound-engineering-plugin
TypeScript · ★ 18,524 · 🍴 1,399 · 📈 349 stars today
Official Compound Engineering plugin for Claude Code, Codex, Cursor, and more
中文介绍 面向 Claude Code、Codex、Cursor 等AI编码工具的复合工程插件。提供集成化的工程能力扩展，帮助开发者更高效地管理项目、执行复杂开发任务。适合使用多个 AI 编码工具的工程师，统一工作流。
affaan-m/ECC
JavaScript · ★ 199,724 · 🍴 30,661 · 📈 908 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 Agent 性能优化系统，为 Claude Code、Codex、Opencode、Cursor 等工具提供技能、直觉、记忆、安全等增强能力。旨在提升 AI agent 的稳定性和智能水平，支持研究优先的开发模式。适用于高级 AI 开发者和研究者。
OpenBMB/VoxCPM
Python · ★ 23,067 · 🍴 2,694 · 📈 779 stars today
VoxCPM2: Tokenizer-Free TTS for Multilingual Speech Generation, Creative Voice Design, and True-to-Life Cloning
中文介绍 VoxCPM2：无分词器的多语言语音生成 TTS 模型，支持创造性声音设计与真实感克隆。基于 tokenizer-free 架构，生成高质量、高表现力的语音，适合多语言配音、语音助手、数字人声音定制等场景。
galilai-group/stable-worldmodel
Python · ★ 1,527 · 🍴 171 · 📈 318 stars today
A platform for reproducible world model research and evaluation
中文介绍 可复现的世界模型研究评估平台。提供标准化的训练环境、评价指标和数据集，解决世界模型研究中结果难以复现的问题。适用于强化学习、机器人控制等领域的研究者，加速世界模型的设计与对比。
Crosstalk-Solutions/project-nomad
TypeScript · ★ 27,491 · 🍴 2,696 · 📈 469 stars today
Project N.O.M.A.D, is a self-contained, offline survival computer packed with critical tools, knowledge, and AI to keep you informed and empowered—anytime, anywhere.
中文介绍 自包含的离线生存计算机，集成关键工具、知识库和 AI，支持无网络环境下保持信息与运行。适用于极端环境、灾害应急、野外探险等场景，为用户提供独立的计算与智能支持。
run-llama/liteparse
Rust · ★ 8,121 · 🍴 481 · 📈 925 stars today
A fast, helpful, and open-source document parser
中文介绍 快速开源的文档解析器，支持从 PDF、Word 等文件中提取结构化内容。优化速度和易用性，适合需要将文档转化为可处理数据的场景，如 RAG 系统、信息抽取和文档索引。
chen08209/FlClash
Dart · ★ 40,488 · 🍴 2,537 · 📈 187 stars today
A multi-platform proxy client based on ClashMeta,simple and easy to use, open-source and ad-free.
中文介绍 基于 ClashMeta 的多平台代理客户端，界面简洁、易于使用，开源且无广告。支持多种代理协议，适用于需要翻墙或科学上网的用户，提供稳定的网络访问体验。
FareedKhan-dev/train-llm-from-scratch
Jupyter Notebook · ★ 2,514 · 🍴 392 · 📈 327 stars today
A straightforward method for training your LLM, from downloading data to generating text.
中文介绍 从头训练大语言模型的实用教程，涵盖数据下载到文本生成的完整流程。提供清晰的步骤和代码，帮助初学者理解 LLM 训练的核心概念。适合希望入门大模型训练的开发者和研究者。
ruvnet/RuView
Rust · ★ 69,206 · 🍴 9,236 · 📈 655 stars today
π RuView turns commodity WiFi signals into real-time spatial intelligence, vital sign monitoring, and presence detection — all without a single pixel of video.
中文介绍 利用 WiFi 信号实现实时空间感知、生命体征监测和存在检测，无需任何摄像头。通过分析信号反射推断位置和生理状态，适用于隐私敏感场景，如智能家居、老人看护和安防。
DataTalksClub/data-engineering-zoomcamp
Jupyter Notebook · ★ 41,849 · 🍴 8,293 · 📈 274 stars today
Data Engineering Zoomcamp is a free 9-week course on building production-ready data pipelines. The next cohort starts in January 2026. Join the course here 👇🏼
中文介绍 免费 9 周数据工程课程，教授构建生产级数据管道。涵盖数据建模、ETL、编排、云服务等核心技能。适合转行或提升数据工程能力的开发者，通过实践项目掌握全栈数据架构。
OpenMOSS/MOSS-TTS
Python · ★ 2,736 · 🍴 243 · 📈 62 stars today
MOSS‑TTS Family is an open‑source speech and sound generation model family from MOSI.AI and the OpenMOSS team. It is designed for high‑fidelity, high‑expressiveness, and complex real‑world scenarios, covering stable long‑form speech, multi‑speaker dialogue, voice/character design, environmental soun
中文介绍 开源语音和声音生成模型家族，由 MOSI.AI 和 OpenMOSS 团队开发。专为高保真、高表现力、复杂实际场景设计，支持语音合成、声音克隆和音效生成。适合语音助手、内容制作和虚拟角色配音。
dreammis/social-auto-upload
Python · ★ 11,959 · 🍴 2,106 · 📈 73 stars today
自动化上传视频到社交媒体：抖音、小红书、视频号、tiktok、youtube、bilibili
中文介绍 自动化视频上传工具，支持抖音、小红书、视频号、YouTube、Bilibili 等多个平台。批量处理视频发布任务，节省手动操作时间。适合内容创作者、营销团队进行多平台分发。
anthropics/skills
Python · ★ 144,441 · 🍴 17,014 · 📈 454 stars today
Public repository for Agent Skills
中文介绍 Anthropic 官方 agent 技能公共仓库，提供预构建的技能模块供开发者使用。覆盖代码编写、数据分析、任务编排等常见能力，帮助快速构建面向特定领域的 AI agent。适用于扩展 Claude 等 AI 工具的功能。
codecrafters-io/build-your-own-x
Markdown · ★ 508,696 · 🍴 48,263 · 📈 817 stars today
Master programming by recreating your favorite technologies from scratch.
中文介绍 汇总了从零动手复现各种经典技术的教程，涵盖数据库、Git、Docker 等。每篇文章教读者通过编程重建一项技术，深入理解其原理。适合希望提升底层编程能力和系统设计能力的开发者。
Why Far Looks Up: Probing Spatial Representation in Vision-Language Models
👍 38
Vision-language models (VLMs) achieve strong performance on spatial reasoning benchmarks, yet it remains unclear whether this reflects structured 3D understanding or reliance on statistical shortcuts in natural images. We introduce a representation-level analysis framework that constructs minimal co
中文介绍 视觉语言模型在空间推理基准上表现强劲，但研究引入表示层分析框架，探究其是否真正具备结构化3D理解，抑或依赖自然图像中的统计捷径。
DynaFLIP: Rethinking Robotics Perception via Tri-Modal-Dynamics Guided Representation
👍 7
Robot manipulation critically depends on perception that preserves the action-relevant aspects of a scene. Yet most robot learning pipelines are built upon visual encoders pre-trained for static recognition or vision-language alignment, leaving motion understanding to downstream policies. We introdu
中文介绍 机器人操作依赖于保留动作相关场景信息的感知，但现有管道多基于静态识别或视觉语言对齐的视觉编码器。研究提出三模态动态引导表示方法DynaFLIP以改进感知。
Tiny but Trusted: Efficient Vision-Language Reasoning for Time-Series Anomaly Detection
👍 0
Recent advances in Vision-Language Models (VLMs) have achieved impressive performance across many tasks, yet prior studies report unsatisfactory performance when applying large language or multimodal models to finding abnormal patterns in sequential data. Public anomaly detection benchmarks typicall
中文介绍 视觉语言模型在时序异常检测中表现不佳。研究提出Tiny but Trusted方法，实现高效视觉语言推理，用于在序列数据中发现异常模式。
Reducing Political Manipulation with Consistency Training
👍 0
Large language models (LLMs) exhibit systematic political bias across a variety of sensitive contexts. We find that LLMs handle counterpart topics from opposing political sides asymmetrically. We refer to this phenomenon as covert political bias and identify 7 categories of techniques through which
中文介绍 大语言模型在敏感话题上表现出系统性政治偏见。研究识别出7种隐性政治偏见类别，并提出一致性训练方法以减少政治操控。
REPOT: Recoverable Program-of-Thought via Checkpoint Repair
👍 6
One-shot Program-of-Thought (PoT) emits a Python program that prints a primitive-action plan; a single invalid action silently invalidates the trajectory. We introduce RePoT (Recoverable PoT): a deterministic verified replay that walks the plan through the environment to its first invalid transition
中文介绍 单次思维程序（PoT）生成Python代码输出动作计划，但无效动作会失效。研究提出RePoT，通过确定性验证重放与环境交互，实现恢复性计划。
Xetrieval: Mechanistically Explaining Dense Retrieval
👍 17
Explaining why dense retrievers assign high relevance scores remains challenging because retrieval decisions are made through opaque high-dimensional embeddings. Existing explanations often focus on surface signals, such as lexical matches, token alignments, or post-hoc textual rationales, and thus
中文介绍 密集检索器分配相关性分数的机制难以解释。研究提出Xetrieval，通过机制解释揭示高维嵌入中的检索决策，超越表面信号。
CoHyDE: Iterative Co-Training of LLM Rewriter & Dense Encoder for Tool Retrieval
👍 4
Tool retrieval over large API catalogs is a core bottleneck for LLM agents: user queries arrive in colloquial, often underspecified language, while the catalog uses technical API vocabulary that no fixed encoder can bridge on its own. The two dominant training approaches, contrastive encoder fine-tu
中文介绍 工具检索是大模型代理处理大型API目录的核心瓶颈。研究提出CoHyDE，通过迭代共同训练LLM改写器和密集编码器，弥合用户口语与API术语之间的差距。
EarlyTom: Early Token Compression Completes Fast Video Understanding
👍 26
Video large language models (Video-LLMs) have demonstrated strong capabilities in video understanding tasks. However, their practical deployment is still hindered by the inefficiency introduced by processing massive amounts of visual tokens. Although recent approaches achieve extremely low token ret
中文介绍 视频大模型在处理大量视觉令牌时效率低下。研究提出EarlyTom，通过早期令牌压缩实现快速视频理解，加速实际部署。
When Cloud Agents Meet Device Agents: Lessons from Hybrid Multi-Agent Systems
👍 11
The design space of agentic AI inference spans two extremes: frontier large language models (LLMs), typically hosted in the cloud and offering strong performance across a wide range of tasks at substantially high cost, and more cost-efficient small language models (SLMs), which are amenable to on-de
中文介绍 智能体AI推理的设计空间涵盖云端大模型与设备端小模型。研究探讨混合多智能体系统结合两者的经验教训，平衡性能与成本。
Thinking Before Constraining: A Unified Decoding Framework for Large Language Models
👍 6
Natural generation allows Large Language Models (LLMs) to produce free-form responses with rich reasoning, yet the lack of structure makes outputs difficult to verify. Conversely, constrained decoding ensures standardized formats but can inadvertently restrict reasoning capabilities by imposing cons
中文介绍 自然生成缺乏结构导致输出难以验证，而约束解码可能限制推理。研究提出统一解码框架，先思考后约束，兼顾灵活性与可验证性。
PhyGenHOI: Physically-Aware 4D Generation of Dynamic Human-Object Interactions
👍 8
We address the task of generating physically accurate and visually faithful 4D Human-Object Interaction (HOI). Given a static 3D human and target object represented as 3D Gaussian Splats (3DGS), our goal is to synthesize dynamic scenes where the human actively engages with the object through actions
中文介绍 研究解决动态人-物交互的4D生成任务。给定静态3D人体和目标物体，通过3D高斯点云表示，合成人体与物体互动的物理准确动态场景。
UniSteer: Text-Guided Flow Matching in Activation Space for Versatile LLM Steering
👍 21
Activation-based control steers large language models (LLMs) by intervening on their internal representations during inference, and has emerged as an effective paradigm for controlling behaviors such as persona and style. However, existing methods often rely on fixed steering directions or task-spec
中文介绍 基于激活的控制通过干预内部表示引导大语言模型行为，但现有方法依赖固定方向。研究提出UniSteer，利用文本引导的流匹配实现灵活控制。
Discovering Cooperative Pipelines: Autoresearch for Sequential Social Dilemmas
👍 1
We study two-level autoresearch for cooperation: an outer-loop AI agent autonomously redesigns the inner-loop pipeline of an LLM policy-synthesis system for multi-agent Sequential Social Dilemmas (SSDs). A researcher agent R (run as a coding agent) reads the inner-loop source code, edits system prom
Verifiable Rewards Beyond Math and Code: Lightweight Corpus-Grounded Process Supervision for Factual Question Answering
👍 6
Applying reinforcement learning to improve factual accuracy in knowledge-intensive question answering faces a reward design dilemma. Response-level rewards provide only coarse supervision and cannot distinguish correct from incorrect statements within a reasoning trace. Sentence-level alternatives o
Towards Verifiable Multimodal Deep Research: A Multi-Agent Harness for Interleaved Report Generation
👍 10
Large Language Models (LLMs) have advanced autonomous agents from deep search, which retrieves concise factual answers, to deep research, which synthesizes scattered evidence into long-form reports. However, verifiable multimodal deep research remains challenging due to open-ended synthesis without
Colored Noise Diffusion Sampling
👍 18
Diffusion models achieve state-of-the-art image synthesis, with their generative trajectories fundamentally exhibiting a spectral bias, resolving low-frequency global structures early and high-frequency fine details later. Conventional stochastic differential equation (SDE) solvers fail to account f
CausaLab: A Scalable Environment for Interactive Causal Discovery Toward AI Scientists
👍 12
We introduce CausaLab, a scalable environment for evaluating interactive causal discovery by LLM agents. Unlike prior evaluations, CausaLab evaluates both whether an agent can solve a problem using causal evidence and whether its answer is grounded in a faithful recovered causal mechanism. Each epis
Token-Level Generalization in LoRA Adapter Backdoors: Attack Characterization and Behavioral Detection
👍 3
We show that LoRA adapters, the dominant distribution format for fine-tuned LLMs, can be reliably backdoored through training data poisoning while preserving baseline task performance. On a Qwen 2.5 1.5B prompt-injection classifier, a small fraction of poisoned examples drives a clean-accuracy-prese
YoCausal: How Far is Video Generation from World Model? A Causality Perspective
👍 39
As video diffusion models (VDMs) advance toward world models, a key question arises: do they truly understand causality, or merely overfit to statistical temporal patterns? Existing benchmarks mostly rely on synthetic data, limiting real-world generalization due to the sim-to-real gap. We present Yo
PhoneWorld: Scaling Phone-Use Agent Environments
👍 0
A central bottleneck for phone-use agents is that controllable, reproducible environments covering real mobile behavior are hard to build at scale. Existing mobile-agent benchmarks have made important progress on evaluation, but they do not by themselves provide a scalable way to construct many new
WorldMemArena: Evaluating Multimodal Agent Memory Through Action-World Interaction
👍 7
Multimodal large language models are increasingly deployed as long-horizon agents, where memory must do more than recall: it must track an evolving world, revise what has gone stale, and surface the right evidence at decision time. Existing benchmarks measure recall over static dialogue, collapse me
PRISM: A Multi-Dimensional Benchmark for Evaluating LLM Peer Reviewers
👍 10
The rapid growth in submissions to machine learning venues has strained the scientific peer-review system and intensified interest in LLM-based automated peer reviewers. However, how good these systems are actually, especially compared to human reviewers at catching scientific gaps, remains poorly u
RUBRIC-ARROW: Alternating Pointwise Rubric Reward Modeling for LLM Post-training in Non-verifiable Domains
👍 8
Pointwise reward modeling offers critical signals for LLM post-training, yet struggles with absolute scoring in subjective, non-verifiable settings. Rubric-based methods address this by decomposing evaluation into explicit criteria, but existing approaches typically depend on frontier LLMs and suffe
Skill0.5: Joint Skill Internalization and Utilization for Out-of-Distribution Generalization in Agentic Reinforcement Learning
👍 21
Equipping large language models with explicit skills has emerged as a promising paradigm for enabling autonomous agents to solve complex tasks. Agent skills can be inherently divided into general skills for broad cognitive transfer and task-specific skills for dynamic execution. However, existing sk
PANDO: Efficient Multimodal AI Agents via Online Skill Distillation
👍 5
Recent advances in multimodal web agents often rely on increased inference-time computation, including rollout search, verifier passes, offline skill discovery, and specialist model stacks. This raises a central question: can a web agent become more efficient as it accumulates experience, rather tha
Alignment Tampering: How Reinforcement Learning from Human Feedback Is Exploited to Optimize Misaligned Biases
👍 2
Reinforcement Learning from Human Feedback (RLHF) is the standard method to align Large Language Models (LLMs) with human preferences. In this work, we introduce alignment tampering, a potential vulnerability where the LLM undergoing alignment influences the preference dataset, causing RLHF to ampli
CONF-KV: Confidence-Aware KV Cache Eviction with Mixed-Precision Storage for Long-Horizon LLM
👍 4
Long-horizon LLM inference turns the key--value (KV) cache into the dominant GPU memory consumer and makes per-token attention increasingly expensive. Many common eviction policies use static recency windows or historical attention, leaving unused a signal computed on every decoding step: the model'
Learning A Unified Risk Map for Autonomous Driving in Partially Observable Environments
👍 5
Occlusion-aware prediction remains a critical challenge in autonomous driving due to the inherent uncertainty of unobserved regions. Existing approaches either overestimate risk based on reachable states or struggle to predict accurate trajectories under high occlusion uncertainty. To address these
Reflective Prompt Tuning through Language Model Function-Calling
👍 4
Large language models (LLMs) have become increasingly capable of following instructions and complex reasoning, making prompting a flexible interface for adapting models without parameter updates. Yet prompt design remains labor-intensive and highly sensitive to formatting, phrasing, and instruction
ORACLE: Anticipating Scams from Partial Trajectories in Streaming App Usage
👍 0
Smartphone scams are increasingly prevalent and typically manifest as multi-stage, cross-application processes with gradually emerging intent. Effective intervention thus requires anticipating scams before the intent becomes explicit. This is inherently challenging, as decisions must rely on partial
How to Build a Software Factory with Claude Code That Ships Features While You Sleep
@sairahul1 · 106.0K 粉丝 · 2.8M 阅 · 1.5K 赞 · 205 转
I thought I was using AI to code. I was actually just typing faster. Here is the difference — and the 7-agent system that changed everything. Save this. It will save you months. THE PROBLEM NOBODY
中文介绍 分享一个7-agent系统，将Claude Code从简单的打字加速工具变为真正的“软件工厂”，能自主构建功能、处理工作流，实现夜间接单、早晨交付的效果。核心是区分“用AI辅助编码”和“让AI负责编码”。
How to Build a Claude Research Agent That Reads the Internet Every Morning and Briefs You in 5 Mins
@cyrilXBT · 181.7K 粉丝 · 127.8K 阅 · 533 赞 · 80 转
Most people start their day the same way. They open Twitter and spend 20 minutes scrolling through noise looking for the three things that actually matter. They open their email and get pulled into
中文介绍 教用户用Claude构建一个自动研究代理，每天早晨自动浏览网页、做简报，避免刷X和邮件的时间浪费。目标是5分钟内获得当日关键信息，提升效率。
How I Use Cursor
@poteto · 26.6K 粉丝 · 86.5K 阅 · 540 赞 · 48 转
I need to get something off my chest. Before my interview @cursor_ai, I had never actually used Cursor. At Meta, Claude Code was explosively taking off. I even paid for a personal $200 a month plan
中文介绍 自述在Meta时团队多用Claude Code且个人自费使用，但面试Cursor前从未用过Cursor的亲身经历。对比了Cursor与Claude Code的使用体验和差异，强调工具选择的重要性。
Step-By-Step LLM Engineering Projects (2026 Edition)
@TheAhmadOsman · 59.9K 粉丝 · 54.5K 阅 · 512 赞 · 65 转
At some point, reading about LLMs stops being enough. You need to build the stack yourself: Tokenizer first, then embeddings, position, attention, Transformer blocks, objectives, decoding, cache, long
中文介绍 整理了一份LLM工程学习路线图，强调动手实践的重要性。项目从Tokenizer、嵌入层、位置编码、注意力机制、Transformer块、训练目标、解码策略、缓存到长上下文处理，覆盖面全。
THE DIFF THAT CHANGED EVERYTHING
@difflawb · 20.3K 粉丝 · 21.9K 阅 · 1.1K 赞 · 389 转
How a 40-line shell script became infrastructure In August 2024, Andrej Karpathy — co-founder of OpenAI, former AI Director at Tesla — published something unexpectedly small. Not a paper. Not a model.
中文介绍 回顾2024年8月Andrej Karpathy发布的一个仅40行shell脚本如何成为基础设施级工具。这个“微小diff”虽不起眼，却引发了AI开发流程的大变化，强调简单代码的长期影响力。
how i make AI videos (a beginner’s breakdown)
@0xileri · 7.3K 粉丝 · 12.2K 阅 · 533 赞 · 63 转
I’ve been getting a lot of DMs since I started posting AI videos, so I figured I’d just write it all out. Fair warning: I’m still learning too. This is just what’s been working for me. tools
中文介绍 一位AI视频制作者分享自己的入门工作流，包含所用工具和步骤，同时承认自己仍在学习。旨在回答频繁收到的私信问题，适合想尝试AI视频但不知从何起步的初学者。
The Start of the End: AI Replacement Has Begun
@ActionModelAI · 57.1K 粉丝 · 5.8K 阅 · 505 赞 · 344 转
We are witnessing the beginning of the biggest economic shift in modern history. And most people still don’t realize it. AI replacement is no longer some distant sci-fi prediction. It has started.
中文介绍 讨论AI替代人力已非科幻，而是正在发生的经济转型。指出多数人尚未意识到这一巨变，并暗示将带来深远影响。属于对未来趋势的思考与预警。
该源今日无内容。
Browsers Are Dead. Codex Just Replaced Them.
中文介绍 观点认为传统浏览器已过时，Codex 工具可能取代其功能。
AI Agents: Every NEW Claude & Codex Feature Explained
中文介绍 详细解析 Claude 和 Codex 中所有新功能，涵盖 AI Agent 特性。
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
With Codex and GPT 5.5 you can just do things.
Before we ship a Claude model, these teams try to break it.
中文介绍 Claude 模型发布前，会由专门团队进行安全性测试，以发现潜在漏洞。
Embrace long-running tasks with Opus 4.8 and Claude Code
中文介绍 Claude 3.5 Opus 4.8 和 Claude Code 支持长时间运行任务，提升自动化效率。
The Problem Solvers: Michele Catasta at Replit
中文介绍 Michele Catasta 在 Replit 分享关于解决问题的经验。
Ship your first Managed Agent
中文介绍 介绍如何部署第一个托管 AI Agent，提供实操指导。
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
How we Claude Code
Making agentic workflows trustworthy and verifiable with a custom DSL
Before we ship a Claude model, these teams try to break it.
中文介绍 Claude 模型发布前，会由专门团队进行安全性测试，以发现潜在漏洞。
Embrace long-running tasks with Opus 4.8 and Claude Code
中文介绍 Claude 3.5 Opus 4.8 和 Claude Code 支持长时间运行任务，提升自动化效率。
The Problem Solvers: Michele Catasta at Replit
中文介绍 Michele Catasta 在 Replit 分享关于解决问题的经验。
Ship your first Managed Agent
中文介绍 介绍如何部署第一个托管 AI Agent，提供实操指导。
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
How we Claude Code
Making agentic workflows trustworthy and verifiable with a custom DSL
Google DeepMind CEO Loves Hard Questions 🙂
中文介绍 Google DeepMind CEO 表示乐于回答尖锐问题，展示开放态度。
Demis Hassabis On What AI Will Do Next
中文介绍 Demis Hassabis 展望 AI 下一步发展方向，涉及通用人工智能等话题。
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
Sakana AI’s God Simulator Is Brilliant
Solved: The Bug That Haunted AI Video For Years
NVIDIA's New AI Broke My Brain
[AINews] Founders and Forward Deployed Engineers
a quiet day lets us highlight the new AIE WF focuses
中文介绍 本期 AINews 聚焦创始人与前置部署工程师，介绍 AIE 新工作流重点。
Boston Children’s uses AI to unlock new diagnoses
Boston Children’s Hospital uses OpenAI technology to improve patient care, reduce operational burden, and help diagnose more than 40 rare disease cases.
中文介绍 波士顿儿童医院利用 OpenAI 技术改善患者护理、减轻运营负担，并帮助诊断超 40 例罕见病。
How Braintrust turns customer requests into code with Codex
How Braintrust engineers use Codex with GPT-5.5 to run experiments and code faster.
中文介绍 Braintrust 工程团队使用 Codex 与 GPT-5.5 将客户需求转化为代码，加速实验与开发。
How the Pope’s Magnifica Humanitas offers a template for individuals to meet the AI moment
Pope Leo XIV’s new encyclical on artificial intelligence includes a statement that warrants serious attention from technologists and policymakers: “Technology is never neutral.” Magnifica Humanitas (“Magnificent Humanity”) is a clarion call to all people to act with courage and solidarity as we ente
中文介绍 教皇利奥十四世发布通谕《辉煌人性》，强调「技术从来不是中立的」，呼吁个人勇敢面对 AI 时代。
not much happened today
**Anthropic** rolled out **Claude Opus 4.8**, which shows incremental improvements but mixed benchmark results, including better cooperation and coding behavior but some regressions in document parsing. Platform updates include mid-conversation system instructions enhancing long agent sessions, thou
中文介绍 Anthropic 推出 Claude Opus 4.8，带来渐进改进，包括更好的协作与编码行为，但文档解析出现退化。平台支持对话中修改系统指令。
Strengthening societal resilience with Rosalind Biodefense
OpenAI launches Rosalind Biodefense, expanding trusted access to GPT-Rosalind for vetted developers and U.S. government partners advancing biodefense, public health, and pandemic preparedness through frontier AI.
中文介绍 OpenAI 启动 Rosalind 生物防御项目，扩展 GPT-Rosalind 权限，为开发者与美国政府伙伴提供生物防御与公共卫生支持。
[AINews] Anthropic raises $965B Series H, releases Opus 4.8 and Dynamic Workflows/ultracode
Total Anthropic victory!
中文介绍 Anthropic 完成 9650 亿美元 H 轮融资，并发布 Opus 4.8 与动态工作流功能。
A shared playbook for trustworthy third party evaluations
OpenAI shares guidance on third-party AI evaluations, covering how to assess model capabilities, safeguards, and validity for frontier systems.
中文介绍 OpenAI 发布第三方 AI 评估指南，涵盖模型能力、安全措施与有效性评估方法。
Profiling in PyTorch (Part 1): A Beginner's Guide to torch.profiler
中文介绍 PyTorch 官方博客发布入门指南，介绍 torch.profiler 工具用于性能分析。
Opus 4.8 🧠, Anthropic at $965B 💰, Microsoft’s coding model 👨💻
中文介绍 今日 AI 新闻速览：Opus 4.8 发布、Anthropic 估值达 9650 亿美元、微软推出编程模型。
The Age of Async Agents — Cognition's Walden Yan & OpenInspect's Cole Murray
80% Devin Commits, Spec-to-PR Workflows, Full VMs, Agent Memory, and PMs Shipping Code
中文介绍 Cognition 与 OpenInspect 讨论异步智能体：Devin 承担 80% 代码提交、Spec-to-PR 工作流、完整虚拟机、智能体记忆与项目经理编写代码。
How Endava builds an agentic organization with Codex
Learn how Endava uses Codex to build an agentic organization, accelerating software delivery and reducing requirements analysis from weeks to hours.
中文介绍 Endava 使用 Codex 构建智能体组织，将需求分析从数周缩短至数小时，加速软件交付。
The AI Hype Index: AI gets booed in graduation season
It is one thing to say AI will change the world. It is another to expect the class of 2026 to applaud it. In fact, when former Google CEO Eric Schmidt told University of Arizona graduates that their task is to help shape AI, he was met with a resounding chorus of boos. “I can…
中文介绍 AI 炒作指数报道：前谷歌 CEO 埃里克·施密特在毕业演讲中遭嘘声，反映公众对 AI 改变世界的态度复杂。
[AINews] Cognition raises $1B in $26B Series D
coding is an uncapped TAM market
中文介绍 Cognition 以 260 亿美元估值完成 10 亿美元 D 轮融资，定位编程为无限市场。
Anthropic raises $65B in Series H at a $965B post-money valuation, releases Opus 4.8 and Dynamic Workflows
**Anthropic** announced a massive **$65B Series H financing** at a **$965B valuation**, led by **Altimeter, Dragoneer, Greenoaks, and Sequoia**, with run-rate revenue surpassing **$47B**. They launched **Claude Opus 4.8**, an update to Opus 4.7 featuring "sharper judgment," "more honesty," and longe
中文介绍 Anthropic 完成 650 亿美元 H 轮融资，估值达 9650 亿美元，由 Altimeter、Dragoneer 等领投，年化收入超 470 亿美元。同时发布 Claude Opus 4.8，性能有提升。
工作时，你们喜欢打开音乐吗
12 回复 · 程序员 节点
[通知] Codex 明早重置额度，上号，/fast 开蹬了
6 回复 · 程序员 节点
codex 即将刷新，快蹬
12 回复 · 程序员 节点
现在如果要加密备份到云盘的文件，有哪些方案
12 回复 · 程序员 节点
美区 AppleID 有办法用中国教育优惠认证吗？
6 回复 · Apple 节点
AI 原生开发，大伙能分享一下经验吗？
6 回复 · 程序员 节点
AI 编程时代， MVP 思维已经失效了？
26 回复 · 程序员 节点
关于 iPhone air 国行 esim 卡的一个想法
9 回复 · Apple 节点
美区礼品卡被连着砍单 3 次
17 回复 · Apple 节点
计划外，换机请教🙏
6 回复 · Apple 节点
痛心疾首！！！揭露gpt plus日抛骗局
买过gpt plus的佬们注意了，gpt plus 日抛号骗局，openai 官方对于账号的封禁只有一种，直接删你账号，而不会出现账号能用反代不能用的情况，通过 codex2api + 授权的方式是一定可以反向代理出来的，各种渠道的所谓日抛号，没那么恶心人的给你 refresh token，恶心人的给你 access token (当然 openai 提高风控之后，access token 或者 session token 这条路基本堵死了)，然后过了两天，卖家登录 chatgpt 网页端或者直接走 api 刷新 refresh token，原来的 refresh token 失效了，但是实际
掉回二级是l站冲浪生涯中不得不品尝的一环
就我而言,入站时间不算长,来时好像还是为了交流云服务器,ds等技术以及各种小巧思 但意外遇到了O/A大善人轮流开闸,各种顶模轮流登场 佬友们有讨论各个渠道灌水程度的(经典opus樱花女孩),也有讨论协议/古法注册机的,更有科普ccs,skill,mcp,harness这些术语都是什么含义,该如何把握的布道者,属实是一派勃勃生机万物竞发的景象 于是乎我这个长期佛系二级用户也开始对三级心生向往,想看看佬们对渠道前沿的机会如何把握,对驾驭顶模有何心得 作为仓鼠党的我开始把收藏夹里的帖子们一个个打开细读,虽说动机是满足浏览量要求,但也意外得到了很多收获(现在我的工作流就是从那几位大佬的讨论贴以及回复中
真的没有人觉得挖鼻孔很利好思考吗
38 个帖子 - 35 位参与者 阅读完整话题
xAI开始招聘了，可兼职可全职，时薪 $35-65 USD （约250-460 RMB/小时） 6个月合同
XAI开始招聘了，可兼职可全职，帮助训练ai 时薪 $35-65 USD （约250-460 RMB/小时） 6个月合同 远程在家做，有福利 特别招中文母语者！帮训练Grok的语音、多语言能力 标注等。筛选测试通过后收入稳定。 申请链接： job-boards.greenhouse.io AI Tutor - Chinese Remote 21 个帖子 - 17 位参与者 阅读完整话题
史上最强原生PPT Skill，更适合中国宝宝
本帖使用社区开源推广，符合推广要求。我申明并遵循社区要求的以下内容： 我的帖子已经打上 开源推广 标签： 是 我的开源项目完整开源，无未开源部分： 是 我的开源项目已链接认可 LINUX DO 社区： 是 我帖子内的项目介绍，AI生成、润色内容部分已截图发出： 是 以上选择我承诺是永久有效的，接受社区和佬友监督： 是 以下为项目介绍正文内容，AI生成、润色内容已使用截图方式发出 使用我创建的这个Skill，一句话即可生成复杂、豪华、可编辑的PPT文件。试一次，要是生成的效果不让你震惊，你来打我。 几大特色： 使用方法，把以下提示词发给Agent： GitHub - GordenSun/Gord
【CHY公益站】第四波补充额度
本帖使用社区公益推广，符合推广要求。我申明并遵循社区要求的以下内容： 我的项目是免费使用的，无收费（变相收费、赞助）部分： 是 我的帖子已经打上 公益推广 标签： 是 我的项目属于个人项目，与公司或商业机构无关： 是 我的项目不存在QQ、TG等群组引流： 是 我的项目不存在非运营必要的网站引流： 是 我的项目不存在为他人推广、AFF： 是 我的项目无关联的商业项目： 是 我的站点存在登录，并已接入 LINUX DO Connect： 是 我帖子内的项目介绍，AI生成、润色内容部分已截图发出： 是 以上选择我承诺是永久有效的，接受社区和佬友监督： 是 以下为项目介绍正文内容，AI生成、润色内容已
缺乏休息是致命的
很辛苦，现在生物钟已经被学校折磨得不成人形了。哪怕是在家里，一天也不能保证超过 7 小时的睡眠，因为睡不着。也有可能是家长每到八点就要把我房间门打开，根本没有好好睡觉的余地。 当然了，如果是困那还好，但是这种晚睡早起的习惯已经把我的精神摧毁了。精神恍惚还是小的，每当需要注意力时，不能把心思聚焦到一件事上那才叫人崩溃，更何况是上一秒想起来的事情下一秒就忘了。 佬友们好好休息，身体才是最重要的，健康没了什么都没了。 44 个帖子 - 29 位参与者 阅读完整话题
【黑与白公益站】【集换卡解析】我是如何当上余额榜第一的w（心得分享）
[!info] 如图，集换卡一直是黑与白公益站最数值膨胀的项目，没有之一，更别提这期还延期并进一步放松了兑换的次数（前几期上限约为45次） 本期集换卡改动 先简单介绍一下本期对于兑换的改动，本期的前49次获得的额度会每次递增5%，在49次达到峰值，持续至第59次。后面的递减比例没有标出，但是根据计算约为每次递减3% 而每天的抽卡上限在前期没有提升爱心互助起扣线的前提下也得到了提升，来到620抽（含vip20免费） ，基本上在35次兑换才能开始盈利，所以前期的资金压力非常显著，建议在资金不足的情况下不要每天抽满付费额度，否则可能入不敷出 抽卡策略 本期是比较特殊的一期，结束日期向后推迟了两次，但
感谢L站，高中生拿下广东省粤港澳学生科创大赛智能体冠军
感谢L站，高中生拿下广东省粤港澳学生科创大赛智能体冠军 还剪了一个两分多钟的冠军速通第一视角vlog，发在视频号了 channels.weixin.qq.com 视频号 补充 很多佬友想要赛题，我周一回学校整理一下发出来给大家。 最致命的是现场全程离线环境，没有办法vibe coding 53 个帖子 - 47 位参与者 阅读完整话题
Opus 4.8知乎大佬模型编程能力榜分数出炉
省流版：Opus 4.8 跳过了 C 和 D 题是因为Opus 4.6(Nothinking) 已经评级A级不再复测 关于评分与实际体验 模型的扣分不完全体现实际编程体验，因此榜单按实际交互体验对模型进行分档： 档位 定义 A 几乎不犯错，仅出现微小的 UI/交互类问题 B 大概率会错，但描述错误后可在 ≤2 轮内修复 C 大概率会错，需更多轮交互，但模型能自主推进修复，无需人工辅助 D 必须由人工提供大量 log、视觉描述、协助操作等才能修复 F 知识或方法论不足，即便有人帮助也无法完成任务 同档位中，若仅少数轮次出现问题、大部分情况表现良好，升半档，以 B+、C+ 表示。 通关机制：A 评
London's Free Roof Terraces
81 points · 18 comments
The Website Specification
186 points · 63 comments
Telli (YC F24) is hiring in engineering, design, and GTM [Berlin, on-site]
1 points · 0 comments
Avian Visitors
47 points · 5 comments
A pictorial introduction to differential geometry (2017)
56 points · 0 comments
Ahoy, DECmate II the little PDP-8 that could
34 points · 5 comments
Mechanical Pencil: An illustrated celebration of the engineering around us
88 points · 10 comments
Microsoft Office 2019 and 2021 for Mac view-only conversion
871 points · 305 comments
Cheese Paper: a text editor specifically designed for writing
109 points · 29 comments
Shantell Sans (2023)
263 points · 27 comments
The AV2 Video Standard Has Released (Final v1.0 Specification)
218 points · 99 comments
wolfSSL releases a new product; wolfCOSE a zero alloc C embbedded COSE stack
93 points · 23 comments
Domain expertise has always been the real moat
626 points · 376 comments
Show HN: Breathe CLI – Paced resonance breathing in the macOS terminal
I built a terminal app that paces slow breathing at 6 breaths per minute for vagal tone training. It's a single Python file, stdlib only, no dependencies — just run breathe and follow the bar.I'm a cardiology patient (HFrEF). Slow breathing at resonance frequency is one of the few non-phar
Parallel Reconstruction of Lawful TLS Wiretapping
104 points · 44 comments
Jef Raskin, the Visionary Behind the Mac (2013)
105 points · 49 comments
Zig ELF Linker Improvements Devlog
208 points · 74 comments
OpenRouter raises $113M Series B
421 points · 201 comments
Accenture to acquire Ookla
https://www.theverge.com/tech/889234/downdetector-ookla-spee..., https://archive.ph/FR8NDhttps://arstechnica.com/information-technology/2026/03/downd...
Voxel Space (2017)
286 points · 59 comments
Openrsync: An implementation of rsync, by the OpenBSD team
418 points · 155 comments
Pandoc Templates
401 points · 52 comments
One year of Roto, a compiled scripting language for Rust
40 points · 7 comments
Security Envelope Pattern collection – S.E.C.R.E.T
3 points · 0 comments
A Gentle Introduction to Lattice-Based Cryptography [pdf]
91 points · 4 comments
I found a seashell in the middle of the desert
https://shell.hawzen.me/
Associative learning turns DEET from aversive to appetitive in Aedes aegypti
39 points · 16 comments
Soviet 80s supercomputer project "Start"
9 points · 1 comments
Racket v9.2
126 points · 14 comments
Mysteries of the Griffin iMate
14 points · 3 comments