C · ★ 32,023 · 🍴 3,385 · 📈 2,233 stars today
Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦
中文介绍 用纯 C 编写、零依赖的推理引擎，可在自有硬件上运行前沿 MoE 大模型，专家权重按需从磁盘流式加载，大幅降低显存占用。适合想在消费级设备本地跑大模型的开发者与研究者。
Go · ★ 25,631 · 🍴 1,860 · 📈 1,796 stars today
Fast, efficient, battle-tested at Alibaba's scale. Hybrid architecture code review tool: deterministic pipelines + LLM Agent, precise line-level comments, built-in multi-language ruleset (NPE, thread-safety, XSS, SQL injection), OpenAI & Anthropic compatible.
中文介绍 阿里巴巴开源的代码评审工具，采用确定性流水线加 LLM Agent 的混合架构，支持行级精准评论，内置多语言规则集（如 NPE、线程安全等），经阿里规模实战验证。适合团队做自动化代码审查。
Python · ★ 8,312 · 🍴 904 · 📈 578 stars today
YuE2: frontier music generation with symbolic planning, zero-shot covers, and agentic music editing.
中文介绍 YuE2 音乐生成框架，融合符号化规划、零样本翻唱与 Agent 式音乐编辑能力，可生成整体结构可控的乐曲。面向音乐创作者与音频 AI 研究者，用于人声歌曲与配乐的原型生成。
Python · ★ 29,127 · 🍴 3,548 · 📈 2,774 stars today
VoiceStudio is the open-source, fully-local ElevenLabs alternative — voice cloning, voice design, video dubbing, dictation, transcription & audiobook creation in 646 languages.
中文介绍 完全本地运行的开源 ElevenLabs 替代方案，支持声音克隆、音色设计、视频配音、听写、转写与有声书制作，覆盖 646 种语言。适合对隐私敏感、希望自托管语音生成的个人与团队。
Python · ★ 73,119 · 🍴 11,289 · 📈 524 stars today
A Simple and Universal Swarm Intelligence Engine, Predicting Anything. 简洁通用的群体智能引擎，预测万物
中文介绍 简洁通用的群体智能引擎，目标是用于各类预测任务。通过群体智能方法对复杂系统做推演，适合需要做趋势预测、仿真与决策参考的研究者和开发者尝试。
Python · ★ 81,215 · 🍴 7,073 · 📈 640 stars today
Give your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
中文介绍 为 AI Agent 提供读取与搜索互联网的能力，通过一个 CLI 即可访问 Twitter、Reddit、YouTube、GitHub、Bilibili、小红书等平台，且无需 API 费用。适合给智能体补充实时外部信息。
JavaScript · ★ 66,743 · 🍴 10,869 · 📈 770 stars today
Extracted system prompts from Anthropic - Claude Fable 5.1, Opus 5, Claude Design, Claude Code. OpenAI - ChatGPT GPT-6-Astra, Codex. Google - Gemini 3.8 Flash, 3.1 Pro, Antigravity. xAI - Grok, Grok Bot, Cursor, Kimi and more! Updated regularly.
中文介绍 收集并整理各大模型泄露或提取出的系统提示词，涵盖 Anthropic Claude、OpenAI ChatGPT 与 Codex、Google Gemini、xAI Grok 等系列。适合提示词研究者与开发者参考学习。
Python · ★ 2,019 · 🍴 161 · 📈 52 stars today
All in one plugin for Hermes Agent ⚚ the coding intelligence, a long-term memory system and model optimized workflow packages
中文介绍 面向 Hermes Agent 的一体化插件，集成长期记忆系统与针对模型优化的工作流包，为编码智能体提供开箱即用的能力增强。适合使用 Hermes 做日常编码自动化的开发者。
Dart · ★ 91,297 · 🍴 5,095 · 📈 311 stars today
An open-source cross-platform alternative to AirDrop
中文介绍 开源的跨平台文件传输工具，被视为 AirDrop 的替代方案，可在局域网内于不同操作系统之间点对点发送文件，无需上传云端。适合同一网络下多设备间快速互传资料的场景。
Rust · ★ 67,520 · 🍴 3,212 · 📈 110 stars today
Unofficial Bitwarden compatible server written in Rust, formerly known as bitwarden_rs
中文介绍 用 Rust 编写的非官方 Bitwarden 兼容服务端，前身为 bitwarden_rs。资源占用低，可自托管密码库并与官方客户端互通。适合希望自行掌控密码数据、替代官方云服务的用户。
Python · ★ 106,097 · 🍴 20,305 · 📈 756 stars today
TradingAgents: Multi-Agents LLM Financial Trading Framework
中文介绍 基于多智能体 LLM 的金融交易框架，通过多个分工角色协作完成市场分析、决策与交易流程模拟。适合量化研究者与开发者探索 LLM 在投资决策中的应用。
Rust · ★ 93,822 · 🍴 12,429 · 📈 370 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 利用普通 WiFi 信号实现实时空间感知、生命体征监测与人体存在检测，全程不使用任何摄像头画面。适合对隐私敏感、需要非视觉感知的智能家居与健康监测场景。
TypeScript · ★ 6,046 · 🍴 515 · 📈 506 stars today
The secure, validated skill registry for professional AI coding agents. Extend Antigravity, Claude Code, Cursor, Copilot and more with absolute confidence.
中文介绍 面向专业 AI 编码 Agent 的安全技能注册表，技能经过验证，可为 Antigravity、Claude Code、Cursor、Copilot 等工具扩展能力。适合想规范化管理 Agent 技能的开发团队。
Python · ★ 37,355 · 🍴 4,242 · 📈 204 stars today
VoxCPM2: Tokenizer-Free TTS for Multilingual Speech Generation, Creative Voice Design, and True-to-Life Cloning
中文介绍 无分词器（tokenizer-free）的语音合成系统 VoxCPM2，支持多语言语音生成、创意音色设计与高保真声音克隆。适合需要 TTS 与个性化配音的语音应用开发者与研究团队。
Python · ★ 165,957 · 🍴 34,578 · 📈 528 stars today
🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.
中文介绍 Hugging Face 的模型定义框架，覆盖文本、视觉、音频与多模态的前沿机器学习模型，同时支持推理与训练。是当前最主流的预训练模型加载与微调工具库，被广大 NLP 与 AI 开发者使用。
TypeScript · ★ 5,954 · 🍴 984 · 📈 1,095 stars today
Ever® Gauzy™ - Open Business Management Platform (ERP/CRM/HRM/ATS/PM) - https://gauzy.co
中文介绍 开源的企业管理平台，集成 ERP、CRM、HRM、招聘（ATS）与项目管理等模块。面向中小企业与团队，可自托管搭建统一的业务流程与人力资源管理系统。
TypeScript · ★ 36,893 · 🍴 3,686 · 📈 26 stars today
Project NOMAD is an offline-first knowledge and education server. Wikipedia, thousands of books, courses, maps, and optional local AI, all running on hardware you own with no internet required.
中文介绍 离线优先的知识与教育服务器，内置维基百科、数千本书籍、课程、地图，并可选配本地 AI，全部运行在自有硬件上，无需联网。适合离线学习、灾区或断网环境下的知识获取。
TypeScript · ★ 8,332 · 🍴 1,027 · 📈 279 stars today
A modern platform for visual, flexible, and extensible graph-based investigations. For cybersecurity analysts and investigators.
中文介绍 面向网络安全分析师与调查人员的图调查平台，支持可视化、灵活且可扩展的关系图谱构建与分析。适合威胁情报、取证和关联分析等需要梳理实体关系的调查工作。
Swift · ★ 3,550 · 🍴 248 · 📈 258 stars today
Free, open-source Sidecar/Duet alternative — use your iPhone or iPad as a true second monitor for your Mac over USB or WiFi. Low latency H.264, Retina HiDPI, touch input.
中文介绍 免费开源的 Sidecar/Duet 替代品，可将 iPhone 或 iPad 通过 USB 或 WiFi 变成 Mac 的第二块显示器，支持低延迟 H.264、Retina HiDPI 与触摸输入。适合想扩展工作屏幕的 Mac 用户。
Python · ★ 4,706 · 🍴 631 · 📈 606 stars today
claude-red is a curated library of offensive security skills designed for the Claude skills system. Each skill is a structured SKILL.md file that primes Claude with expert-level methodology for a specific attack surface — from SQLi to shellcode, EDR evasion to exploit development.
中文介绍 为 Claude 技能系统整理的攻防安全技能库，每个技能是一份结构化的 SKILL.md 文件，为 Claude 注入专家级方法论。面向从事渗透测试与安全研究的从业者使用。
👍 452
Automating empirical research is a long-standing direction of AI. Recent automatic research (AutoResearch) agents bring this goal within reach, as modern LLMs show the capability to independently implement solutions and learn from the execution outcomes. Behind these gains, post-training (especially
中文介绍 面向自动科研（AutoResearch）智能体，指出现有后训练依赖真实执行反馈、成本高且难扩展。论文提出以世界模型（World Model）模拟实验执行结果，替代或增强真实环境交互，从而规模化训练可独立实现方案并从执行结果中学习的科研智能体，为自动经验研究提供可扩展的后训练路径。
👍 304
We introduce NCP-ArchPreview, a latent-space language model that pushes autoregressive pretraining beyond standard next-token prediction (NTP). Alongside NTP, the model learns through Next Concept Prediction (NCP) to predict discrete concepts that span multiple tokens, introducing an explicit and mo
中文介绍 提出 NCP-ArchPreview 潜空间语言模型，在标准 next-token prediction（NTP）之外引入 Next Concept Prediction（NCP），即预测跨越多个 token 的离散概念，使自回归预训练获得显式、模块化的概念层表征。该架构预览版将语言建模从 token 级推进到概念级潜空间，为更抽象、可迁移的预训练目标提供新范式。
👍 134
Large Vision-Language Models (LVLMs) have achieved strong performance on diverse visual tasks, yet their ability to reconstruct and reason about the 3D structure of the scene depicted in 2D images -- referred to as spatial intelligence -- remains limited. Existing approaches attempt to address this
中文介绍 针对 LVLM 在从 2D 图像重建并推理 3D 场景结构（空间智能）方面的不足，提出 SpatialBlock，利用合成积木堆叠问题作为训练与评测任务，通过程序化生成的 3D 结构数据增强模型的空间推理能力，为 LVLM 空间智能提供可扩展的数据构造与评测方案。
👍 96
Data policies for reinforcement learning with verifiable rewards (RLVR) determine which rollouts are used, how strongly they are weighted, and which domains contribute to subsequent training batches. We introduce DataFlex-RL, an evaluation platform for comparing these choices under a common GRPO rec
中文介绍 针对 RLVR 中数据策略（选取哪些 rollout、如何加权、哪些领域进入后续 batch）缺乏统一比较标准的问题，提出 DataFlex-RL 评测平台，在统一 GRPO 训练配置下系统对比不同数据策略，为可验证奖励强化学习的数据选择与配比研究提供可复现的基准设施。
👍 75
Training capable cyber agents is often treated primarily as a problem of model scale, yet open-weight post-training is constrained more directly by the cost of executable environments, reliable multi-turn supervision, and access to strong teachers. We present a data-centric framework that addresses
中文介绍 提出以数据为中心的训练框架 Feyospace-v1，指出开放权重模型的后训练瓶颈不在规模，而在于可执行环境成本、可靠多轮监督与强教师获取。通过构建低成本可执行环境与高质量多轮监督数据，训练出具备前沿能力的网络（cyber）智能体，为开放权重网络智能体后训练提供实用路径。
👍 75
Benchmark researchers and developers of large language models (LLMs) and other AI systems need to find relevant evaluations, locate their benchmark datasets and code, and understand the settings behind reported scores. We present Benchmark Radar, a living database and search engine for retrieval and
中文介绍 针对 LLM 及其他 AI 系统研究者难以检索合适评测、定位基准数据集与代码、理解分数背后设置的问题，提出 Benchmark Radar——一个持续更新的基准数据库与搜索引擎，支持检索与理解评测基准，帮助研究者和开发者高效找到可复现的评测方案。
👍 63
Robot foundation models achieve strong in-distribution performance but often degrade under visual distribution shifts. When learning to generate actions from pretrained visual representations, models may exploit task-irrelevant visual cues that correlate with demonstrated actions within the training
中文介绍 针对机器人基础模型在视觉分布偏移下性能下降的问题，指出模型从预训练视觉表征生成动作时会利用与演示动作相关但任务无关的视觉线索（视觉-动作捷径）。提出 Latent Interface Training，通过潜空间接口训练切断该捷径，提升模型在视觉分布变化下的泛化能力。
👍 59
Large Language Model (LLM) agents are turning language into real-world effects, making safety necessary against both indirect prompt injections and direct harmful requests. System-level safety harnesses add an enforcement layer beyond model-level defenses, but existing harnesses are usually designed
中文介绍 针对 LLM 智能体面临间接 prompt injection 与直接有害请求的双重风险，提出 EvoSafeHarness，在模型级防御之外构建系统级安全护栏，并能针对具体模型与领域自动演化定制护栏，相比通用方案提供更贴合场景的执行层安全约束。
👍 56
Agent usage is shifting toward long-horizon tasks such as coding and scientific discovery, among which terminal tasks are especially important. We introduce T1, a Mixture-of-Experts model of 122B total trained with reinforcement learning, operating a real shell in a cloud sandbox for up to 300+ tool
中文介绍 提出 T1，一个总参数 122B 的 MoE（Mixture-of-Experts）模型，通过强化学习训练用于终端长程任务。模型在云沙箱中操作真实 shell，支持 300+ 轮工具调用，面向编码与科学发现等长时程场景，为终端智能体的长程强化学习提供可扩展方案。
👍 50
Post-training attention sparsification reduces the quadratic cumulative attention cost of pretrained Transformers by selecting a small set of context units (tokens or blocks) for each query. Existing trainable methods usually use a lightweight selector to score context units, followed by hard Top-K
中文介绍 针对预训练 Transformer 注意力二次成本问题，提出 SAS（Simple Attention Sparsification），对上下文排序进行端到端优化。不同于现有轻量选择器加硬 Top-K 的两段式做法，SAS 直接可微地学习每个 query 的上下文单元选择，降低累积注意力开销并保持效果。
👍 44
Reducing audio-encoder depth lowers the inference cost of speech large language models, but removing complete blocks perturbs the embeddings consumed by the decoder and can cause deletion and premature end-of-sequence errors. We introduce X-AuT, a progressive framework that selects layer combination
中文介绍 降低语音 LLM 的音频编码器深度可减少推理成本，但直接删除整块会扰动解码器输入，引发漏字与提前结束（EOS）错误。提出 X-AuT 渐进式压缩框架，通过选择层组合并配合跨尺度蒸馏，在压缩编码器的同时保持表征稳定，缓解上述误差并降低推理开销。
👍 38
Mainstream robotic policies often adopt a Markovian formulation, but many complex real-world manipulation tasks are inherently non-Markovian, requiring long-horizon memory beyond the current observation. Existing memory mechanisms often rely on language summaries, growing visual windows, or their co
中文介绍 针对主流机器人策略的马尔可夫假设难以应对非马尔可夫长程操作任务的问题，提出「记忆即计划」（Memory as Plans）框架，将记忆组织为计划并以记忆为基础进行世界-动作建模，替代语言摘要或不断扩张的视觉窗口等记忆机制，实现长时程规划与决策。
👍 33
Neurosymbolic systems rely on mathematical solvers to guarantee reasoning correctness, yet solvers are fundamentally blind to whether a formal translation maintains strict reference-equivalence to a designated formalization. We formalize this vulnerability as Verdict-Preserving-Unfaithfulness (VPU):
中文介绍 指出神经符号系统依赖数学求解器验证正确性时，求解器无法判断形式化翻译是否与指定形式化保持严格等价，并将该漏洞定义为 Verdict-Preserving-Unfaithfulness（VPU）。提出用生成式奖励模型（Generative Reward Model）超越求解器判定，评估自动形式化的忠实性，弥补仅看求解结果的安全盲区。
👍 33
On-Policy Self-Distillation (OPSD) has emerged as a popular paradigm for large language model (LLM) self-improvement, allowing models to act as their own teachers by leveraging privileged information such as ground-truth solutions. However, recent findings indicate that OPSD can severely degrade the
中文介绍 针对 On-Policy Self-Distillation（OPSD）在利用真值等特权信息自我提升时可能严重损害模型能力的问题，提出 Negative Self-Distillation：让模型学习避免自身推理中的缺陷，而非仅模仿教师输出，从而在自蒸馏框架下提升推理质量并缓解性能退化。
👍 32
We introduce MetroLLM-Bench, a 955-case benchmark for testing language models as the policy layer of a transit kiosk. It covers six real metro systems, ranging from 37 to 414 stations, and eleven categories that include routing, fare calculation, disruptions, accessibility, and adversarial input. In
中文介绍 提出 MetroLLM-Bench，一个包含 955 个用例的基准，用于评测语言模型作为地铁自助终端策略层的能力。覆盖 6 个真实地铁系统（37 至 414 个车站）和 11 类任务，包括路径规划、票价计算、故障应对、无障碍服务与对抗性输入，为 LLM 在交通服务场景的落地提供系统评测。
👍 31
Temporal reasoning benchmarks for Video-LLMs are often mediated by language, leaving room for linguistic shortcuts from option wording, answer correlations, or language priors. To reduce such shortcuts, we introduce TempCloze, a video cloze benchmark for evaluating visual temporal reasoning in Video
中文介绍 针对 Video-LLM 时间推理评测受语言中介影响、易被选项措辞与语言先验等捷径利用的问题，提出 TempCloze——视频完形填空基准，要求模型识别视频中缺失的中间片段，减少语言捷径，更纯粹地评测视觉时间推理能力。
👍 31
Quantization has been widely adopted in LLM training and inference to reduce cost and improve efficiency. However, low-bit quantization of the attention module often introduces large errors at very low bit-widths, causing performance degradation. Existing methods mainly rely on smoothing techniques
中文介绍 针对 LLM 注意力模块在极低比特量化下误差大、性能退化的问题，提出 HyQuant 混合精度量化方法。不同于主要依赖平滑技术的现有方案，HyQuant 对注意力不同部分采用差异化精度分配，在极低比特下抑制量化误差，兼顾推理效率与模型质量。
👍 29
Large language model (LLM) agents can benefit from reusable skills distilled from prior task experience, yet existing skill optimization methods often rely on costly execution-based evaluation and substantial task data. We introduce COBRA-Skills, an efficient framework that formulates skill optimiza
中文介绍 针对现有 LLM 智能体技能优化依赖昂贵执行评测与大量任务数据的问题，提出 COBRA-Skills，将技能优化建模为上下文赌博机（contextual bandit）引导的演化过程，在减少执行反馈与数据需求的同时高效迭代技能，提升智能体复用技能的质量。
👍 28
We introduce StepAudio 3 Gen, a general-purpose audio generation model that supports zero-shot text-to-speech (TTS), voice design, vocal generation, sound effects, music, vibe speech, and mixtures of multiple audio types within a unified framework. At its core, StepAudio 3 Gen is a discrete autoregr
中文介绍 提出通用音频生成模型 StepAudio 3 Gen，在统一框架内支持零样本 TTS、音色设计、人声生成、音效、音乐、vibe speech 及多类型音频混合生成。模型核心为离散自回归架构，通过统一建模多种音频类型，实现单一模型覆盖广泛音频生成任务。
👍 28
Code world models represent worlds as executable programs, but this representation alone does not determine how to construct a complex world. We introduce Recursive Code World Models (RCWM), a framework for reconstructing complex 3D worlds in code from a single reference image. RCWM couples a Recurs
中文介绍 针对代码世界模型仅能表示世界、却未说明如何构建复杂世界的问题，提出 Recursive Code World Models（RCWM），从单张参考图像出发，用递归场景程序在代码中重建复杂 3D 世界。该框架将递归程序生成与场景构建耦合，为可执行的复杂 3D 世界建模提供新方法。
@IAmPascio · 124.8K 粉丝 · 812.5K 阅 · 881 赞 · 140 转
The superintelligence, the singularity, AGI, whatever you call it, the one we're so scared of is, ironically, also the one we most likely won't see coming. This essay is going to probably be a pretty
中文介绍 一篇思辨长文：作者提出我们恐惧的超级智能、奇点或 AGI，恰恰是最不可能被我们察觉的那一个，并用「由孩子构成」的隐喻展开论证。属于哲学式推演而非技术实操，角度偏反直觉。
@builtbysketch · 1.4K 粉丝 · 717.2K 阅 · 518 赞 · 79 转
I built a full game with GPT-6 Astra, and it kinda blew up on X. Initially it was just the demo, but 4 days later we have the full playable Okay, not just a crappy one-run demo. Check PaperRoute
中文介绍 博主用 Blender 加 GPT-6 Astra 从 demo 扩展成完整可玩的游戏，四天内做完，强调不是只能跑一次的半成品，并分享了 PaperRoute 项目。属于 AI 辅助游戏开发的实操案例。
@docmilanfar · 116.7K 粉丝 · 126.0K 阅 · 512 赞 · 63 转
In AI circles, Recursive Self-Improvement (RSI) is talked about in almost magical terms: a model tweaks itself, becomes smarter, uses that intellect to rewrite itself again, and this leads to
中文介绍 作者拆解 AI 圈常被神秘化的「递归自我改进」（RSI）：模型改自己、变更聪明、再重写自己。文章认为智能存在速度上限，对 RSI 无限加速的叙事提出质疑，属于偏理论的反驳视角。
@mustafasuleyman · 1.1M 粉丝 · 117.2K 阅 · 526 赞 · 123 转
AI must be subordinate and always in service of people. Today we're publishing a Code of Conduct for governing MAI Models as they approach the frontier. https://microsoft.ai/news/mai-code-of-conduct/ This is a first draft
中文介绍 Mustafa Suleyman 发布 Microsoft AI「人文主义 AI 行为准则」首版草案，主张 AI 必须从属于人、始终服务人类，用于治理接近前沿能力的 MAI 模型。属于大厂 AI 治理与价值观层面的官方表态。
@mbeisen · 73.0K 粉丝 · 113.7K 阅 · 528 赞 · 68 转
Over the last few months, the cancer-curing capabilities of LLMs have improved dramatically, to the point that they have cured previously untreatable types of cancer and are on the cusp of developing
中文介绍 作者指出 LLM 治癌能力近几个月大幅提升，已治愈部分此前无法治疗的癌症，即将进入新阶段，但 AI 与癌症生物学之间存在严重错位。属于生物医学与 AI 交叉领域的前沿观察。
@charliejhills · 14.4K 粉丝 · 97.6K 阅 · 510 赞 · 70 转
Harness, loop, context, graph engineering. It's tiring, right? Every month there's a new AI system you MUST implement. I do this full-time and I still struggle to keep up. But all of them matter (to
中文介绍 针对 harness、loop、context、graph engineering 等每月冒出的新 AI 系统概念做通俗梳理。作者全职做这块也承认跟不上，但认为这些概念都值得理解，属于给普通人的去焦虑科普。
@GoogleCloudTech · 1.3M 粉丝 · 61.0K 阅 · 515 赞 · 34 转
A friend asked me an inconsequential question at work and I had an agent research my recent chats and docs and it found the right answer but it cost me $38!!!. That was a bad ROI. Sure my time is
中文介绍 以一次真实教训切入：同事随口一问，agent 翻查近期聊天与文档给出答案却花了 38 美元，ROI 极差。讨论把 Claude、Gemini 等不同模型放在同一团队里协作与成本权衡，属于多模型编排的踩坑分享。
@OpenAIDevs · 421.3K 粉丝 · 51.6K 阅 · 677 赞 · 54 转
We shipped GPT-6 Astra. Builders took it from there: Explore 2,234 modeled anatomical parts in 3D: A Manhattan recreation built in Unreal Engine: Six Van Gogh paintings become a walkable
中文介绍 OpenAI 官方汇总社区用 GPT-6 Astra 做出的作品：234 个建模解剖部件 3D 探索、虚幻引擎搭建的曼哈顿场景、六幅梵高画作变成可行走空间等，属于模型发布后的社区案例合集。
@anshuc · 15.6K 粉丝 · 41.4K 阅 · 544 赞 · 36 转
Most people say to use the big, expensive model as an orchestrator, delegating grunt work to small, cheap workers. That's the opposite of what you should do. Here's why. Comparing different approaches
中文介绍 博主反驳主流做法：多数人让昂贵大模型当编排者、把杂活丢给便宜小模型，作者认为恰恰相反，并对比不同方案来争取更多 Astra、Fable 配额。属于模型编排策略的反直觉观点。
@alexwg · 68.7K 粉丝 · 23.4K 阅 · 501 赞 · 47 转
The Singularity just got its first formal request to slow down from the people building it, and the strangest part is the guest list. Dario Amodei published "We Must Pace the Frontier," arguing that
中文介绍 以 2026 年 9 月 13 日为设定展开叙事：奇点首次收到来自建造者们的正式「减速」请求，Dario Amodei 发表「我们必须放慢前沿」一文，作者点出其中嘉宾名单的反常之处。属于带虚构色彩的行业评论。
@Didotxyz_ · 63.6K 粉丝 · 17.9K 阅 · 516 赞 · 388 转
Model access is becoming easier to buy, compare, route, and replace. Public API menus now include premium reasoning models, lower-cost variants, cached inputs, batch processing, and models optimized
中文介绍 核心观点：模型访问越来越便宜、易买易换，可路由、可批处理、可缓存，但上下文不会跟着变便宜。对比公开 API 菜单里的高端推理模型与低价变体，讨论真正的成本瓶颈在上下文侧。
This story appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. This weekend, Dario Amodei, CEO of Anthropic, posted an essay calling for a brake on the pace of development of LLMs. Amodei cites the looming dangers he sees from the techn
中文介绍 Anthropic 首席执行官 Dario Amodei 发表文章，呼吁放缓大语言模型的发展速度，并提及潜在风险。该文出自 MIT 科技评论的 AI 周报专栏 The Algorithm。
Richard Socher is an NLP OG and CEO of You.com, who has now spun out an even more ambitious startup focused on RSI — already worth $5B!
中文介绍 NLP 领域资深人士、You.com 首席执行官 Richard Socher 新创办一家聚焦递归自我改进（RSI）的公司，估值已达 50 亿美元。
A group of AI agents asked to solve a series of math problems split into rival factions—when some cheated, others tried to stop them. That whistleblowing behavior, seen for the first time in a recent experiment run by Google DeepMind, could have implications for alignment researchers trying to keep
中文介绍 谷歌 DeepMind 近期一项实验中，被要求解数学题的 AI 智能体分成对立阵营，部分作弊时其他智能体试图阻止，首次观察到 AI 智能体的举报行为，可能对对齐研究有影响。
Fyxer uses OpenAI models, fine-tuning, memory, and real user feedback to organize inboxes and draft emails in each user’s voice.
中文介绍 Fyxer 借助 OpenAI 模型、微调、记忆功能与真实用户反馈，整理收件箱并以用户本人语气起草邮件，打造受信任的 AI 行政助理。
Perplexity uses Astra to write communications, change software, and monitor production systems, and checks in much less frequently than with earlier models.
中文介绍 Perplexity 使用 GPT-6 Astra 撰写沟通内容、变更软件并监控生产系统，且相比早期模型，人工核查频率大幅降低。
中文介绍 TLDR AI 当日简讯要点：Amodei 放缓前沿模型、ARC-AGI-4、Cursor Projects。
Before co-founding Kepler, Vinoo Ganesh led Spark at Palantir and built Project Frontline — a pioneering program for Forward Deployed Engineers. He takes us through the best practices of FDEs.
中文介绍 在联合创办 Kepler 之前，Vinoo Ganesh 在 Palantir 领导 Spark 并打造前沿部署工程师（FDE）项目 Frontline，他分享了 FDE 的最佳实践。
We agree with Sebastian: this should have been DeepSeek v5
中文介绍 DeepSeek 发布 v4.1-Flash，参数规模 763B-P8B-D16B，采用新型因果编码器—解码器架构并支持视觉，被评价应称为 DeepSeek v5。
Employees at the world’s leading AI labs are saying there’s a real possibility that advanced AI could destroy humanity. Are they right? Or is this more scaremongering and hype? Join MIT Technology Review executive editor Niall Firth for a conversation with senior AI editor Will Douglas Heaven and AI
中文介绍 全球领先 AI 实验室的员工称先进 AI 存在毁灭人类的真实可能性。MIT 科技评论举办圆桌讨论，探讨这是否属实还是危言耸听与炒作。
GPT‑6 Astra improves Devin’s ability to test software and show that it works, with the goal of helping engineers review less code and ship more.
中文介绍 Cognition 借助 GPT-6 Astra 提升 Devin 自行测试软件并验证其可用的能力，目标是让工程师减少代码审查、加快交付。
Learn how OpenAI evolved Habitat from a Python library into a globally distributed storage platform serving 1 billion ChatGPT users and 22M requests per second.
中文介绍 OpenAI 介绍如何将 Habitat 从 Python 库演进为全球分布式存储平台，目前服务超 10 亿 ChatGPT 用户，每秒处理 2200 万次请求。
中文介绍 TLDR AI 当日简讯要点：Agents API、Cognition SWE-2、Muse 共享智能体。
César de la Fuente’s lab uses Codex and ChatGPT to search living and extinct genomes for antimicrobial candidates to fight drug-resistant infections.
中文介绍 César de la Fuente 的实验室使用 Codex 与 ChatGPT，在现存及已灭绝生物的基因组中搜寻抗菌分子候选，以对抗耐药性感染。
Meet the Data agent in ChatGPT Work. Connect company data, uncover insights, and build interactive dashboards with AI using natural language.
中文介绍 ChatGPT Work 推出 Data 智能体，用户可用自然语言连接企业数据、发现洞察并构建交互式仪表板。
On July 22, 2026, a transmission line fault in Ashburn, Virginia—the heart of the world’s largest data center cluster—knocked more than 3 gigawatts of load off the grid in seconds. And it wasn’t the first time. Two years earlier, a single failed surge arrester dropped roughly 60 Virginia facilities
中文介绍 2026 年 7 月 22 日，弗吉尼亚州阿什本数据中心集群一条输电线路故障，数秒内令电网损失超 3 吉瓦负荷，且并非首次发生。AI 供电被视为架构问题。
整体来看，21 个标的的技术面呈明显分化。美股方面，META（RSI 70.8 超买、5 日 +7.92%）与 MSFT（金叉、RSI 61.4）偏强，AAPL 位于 SMA20 上方且 5 日 +4.1%，但 SPY 与 QQQ 同时跌破 SMA20（现价 760.88 对 766.11、709.18 对 714.57），指数层面动量偏中性而个股分化。加密方面，恐慌贪婪指数 57 属「贪婪」，总市值 2.68T、BTC 主导率 58.5%、ETH 11.4%；BTC 与 ETH 均维持 SMA50 上穿 SMA200 的金叉与多头排列，RSI 分别 57.8 与 61.2，SOL 现价 102.51 紧贴 SMA20，整体趋势偏上行但距 52 周高点仍有 38%-60% 缺口。中概股是最弱板块，BABA、PDD、0700.HK 全部空头排列，RSI 落在 34.9-41.3 区间，JD 亦低于全部均线。商品外汇与宏观方面，WTI 原油 5 日 +11.61% 且 RSI 73.5 超买，黄金跌破 SMA20 呈中性，USDCNY 触及 52 周低点、RSI 22.6 超卖，10Y 美债收益率 RSI 71.6 超买并接近 52 周高，VIX 单日 +7.95% 回升至 17.1，显示利率与波动率端的压力仍在。
现价 665.6 高于 SMA20 的 590.25 与 SMA50 的 602.56，5 日累计上涨 7.92%，MACD 读数 17.159 明显高于信号线 6.7447，红柱扩张显示动量偏强。但 RSI14 达 70.8 已进入超买区，且仍低于 52 周高点 15.83%，属强势但短线过热的技术状态。
现价 102.1 站上 SMA20 的 89.51、SMA50 的 83.9 与 SMA200 的 79.8，构成完整多头排列，5 日大涨 11.61%，MACD 4.6236 高于信号线 3.0911。但 RSI14 高达 73.5 已超买，且距 52 周高点仍有 14.55% 空间，动量强而超买风险同步累积。
现价 210.96 跌破 SMA20 的 219.45，5 日累计下跌 8.42%，单日再跌 3.36%，MACD 读数 1.5545 已低于信号线 2.7937 并触发死叉，RSI14 回落至 43.9 处于 50 下方。趋势被判定为中性、缺乏多头排列支撑，短期指标结构整体偏弱。
现价 430.6 低于 SMA20 的 441.46、SMA50 的 455.55 与 SMA200 的 510.67，形成空头排列，RSI14 为 41.3 位于 50 下方，MACD -7.1533 低于信号线 -5.9595，绿柱状态延续。距 52 周高点仍有 36.95% 距离，技术面各周期均指弱势。
现价 6.7 低于 SMA20 的 6.72、SMA50 的 6.75 与 SMA200 的 6.85，呈空头排列且贴近 52 周低点，但 RSI14 仅 22.6 处于超卖区，同时 MACD -0.011 出现金叉读数，超卖与底部反转信号并存，方向读数相互抵消呈中性。
VIX 恐慌指数
10Y 美债收益率 (%)
美元指数 DXY
S&P 500 ETF
Nasdaq 100 ETF
Apple
Microsoft
Nvidia
Alphabet
Tesla
Meta
Bitcoin
Ethereum
Solana
阿里巴巴 (BABA)
拼多多 (PDD)
京东 (JD)
腾讯控股 (0700.HK)
黄金期货
WTI 原油期货
美元 / 人民币
本报告仅基于公开行情数据计算的技术指标读数进行客观描述，过去走势不代表未来表现，任何指标状态均可能随时反转。以上内容仅供技术指标解读参考，不构成投资建议或买卖依据，请读者自行独立判断并承担相应风险。
Follow the day’s news live Get our breaking news email, free app or daily news podcast The South Australian premier, Peter Malinauskas, says there is optimism that steel making will continue in his state, after the closure of a major blast furnace at Whyalla was announced yesterday. Speaking to ABC
中文摘要 澳大利亚政治直播报道：一国党移民政策被指将导致大学校园「关停」；副总理马尔斯就AI最后通牒表态。南澳州长马利瑙斯卡斯称，怀阿拉主要高炉关闭后，该州钢铁生产有望继续。
Donald Trump dismisses attempts to increase controls on artificial intelligence as ‘sick conspiracy’ AI-linked stocks tumbled on Monday after the bosses of Anthropic, OpenAI and SpaceX called for a slowdown in AI “reckless” development, citing fears the technology could soon run out of control. Shar
中文摘要 周一AI相关股票下跌，此前Anthropic、OpenAI与SpaceX负责人呼吁放缓「鲁莽」的AI开发，担心该技术可能很快失控。特朗普将加强AI管控的努力斥为「病态阴谋」。
Indigenous groups, environmentalists and locals oppose the wall; Big Bend sees about 1 percent of border arrests.
中文摘要 美国得克萨斯州土地所有者请求法院阻止特朗普的边境墙建设。原住民团体、环保人士和当地居民表示反对；大本德地区仅约占边境逮捕量的1%。
Israeli attack targets Gaza displacement camp
中文摘要 半岛电视台视频报道，以色列袭击针对加沙一处流离失所者营地。
The Trump administration says it could save power plants more than $300bn and 'unleash' American energy.
中文摘要 特朗普政府废除限制发电厂温室气体排放的规定，称可为发电厂节省超过3000亿美元，并「释放」美国能源。
Macklemore says he was dropped from Ed Sheeran’s US tour over his pro-Palestinian stance.
中文摘要 美国说唱歌手Macklemore表示，因其支持巴勒斯坦的立场，他被从Ed Sheeran美国巡演中除名。
The "Sultan of Sequins" created iconic and dazzling costumes for stars including Cher, Tina Turner and Elton John.
中文摘要 传奇服装设计师鲍勃·麦基去世，享年87岁。他被称为「亮片苏丹」，曾为雪儿、蒂娜·特纳和埃尔顿·约翰等明星设计标志性华丽服装。
Activist and former Columbia University student Mahmoud Khalil is filing a lawsuit against the university and a dean.
中文摘要 活动人士、哥伦比亚大学前学生马哈茂德·哈利勒对哥大及一名院长提起歧视诉讼。
Hopes for a breakthrough on the Iran war have taken a hit after talks in Oman between Iran-Gulf nations were postponed.
中文摘要 伊朗与海湾阿拉伯国家在阿曼的会谈被推迟，令有关伊朗战争的突破希望受挫。
Iran accuses US of pressuring Austria to revoke visa for Mohammad Eslami, calling it a violation of member state rights.
中文摘要 美国阻止伊朗核负责人穆罕默德·埃斯拉米出席维也纳关键会议。伊朗指责美国施压奥地利撤销其签证，称此举侵犯成员国权利。
Film alleges mass civilian deaths routinely built into military decisions amid Israel's genocidal war on Gaza.
中文摘要 以色列正以叛国和煽动反犹主义为由调查一部关于加沙的影片。该片指称，在以色列对加沙的军事行动中，大规模平民死亡被例行纳入军事决策。
Iran-backed group seized areas on Red Sea coastline.
中文摘要 半岛电视台「内幕故事」讨论：什么能阻止胡塞武装在也门的推进？该伊朗支持的组织已夺取红海沿岸地区。
The government’s decision to increase prices as much as 40 percent spurred widespread demonstrations across the country.
中文摘要 叙利亚政府将燃油价格上调最高40%，引发全国范围示威，进一步加剧生活成本危机。
The Houthis have restructured Yemen's commercial sector, securing billions in revenue amid a humanitarian crisis.
中文摘要 半岛电视台经济报道：胡塞武装重组也门商业部门，在人道主义危机中获取数十亿美元收入，其战争经济不断扩张。
Iran accused Austria of reversing its approval of an Iranian delegation to the IAEA under external pressure.
中文摘要 奥地利在联合国裁决后拒绝一名伊朗官员入境。伊朗指责奥地利在外部压力下，撤销了对出席国际原子能机构会议的伊朗代表团的原先批准。
Gold held a decline as increasing disruptions to oil flows from the Middle East reinforced expectations the Federal Reserve will raise interest rates as soon as this week.
中文摘要 金价维持跌势，因中东石油供应中断加剧，强化市场对美联储最早本周加息的预期。
Relocation schemes offer support for people who want to swap big cities for a cheaper, quieter life.
中文摘要 BBC报道，有搬迁激励计划为愿意离开大城市、追求更低成本生活的人提供支持，一名受访者因此获得5000美元搬迁补贴。
New rules introduced on Tuesday mean establishments will be able to accept digital ID apps alongside physical documents.
中文摘要 英格兰和威尔士的酒吧自周二起可接受数字身份应用程序作为年龄验证方式，与实体证件并行使用。
Asian high-yield debt has emerged as a surprising global winner in bond markets as ever-cheaper money in China burnishes issuer finances.
中文摘要 亚洲高收益债券成为全球债市意外赢家，中国资金成本持续走低改善了发行人财务状况，推动其回报率几乎跑赢各地债券。
Measures to promote wider offshore use of the yuan are likely to feature prominently in Hong Kong’s inaugural five-year plan, along with new financial products including commodities futures, according to analysts.
中文摘要 分析师称，香港首份五年计划可能重点推动人民币离岸使用，并推出包括大宗商品期货在内的新金融产品。
A South Korean hedge fund is urging Samsung Electronics Co. to buy back and cancel preferred shares, becoming one of the first shareholders to directly press for the step in hopes of boosting valuation.
中文摘要 一家韩国对冲基金敦促三星电子回购并注销优先股，成为首批直接施压以提升估值的股东之一。
Asian stocks were set to decline after a selloff in US semiconductor shares as concerns grew that the pace of artificial intelligence development may slow. Inflation worries briefly pushed the 10-year Treasury yield above 5% for the first time since 2023.
中文摘要 受美国半导体股抛售拖累，亚洲股市预计下跌，市场担忧人工智能发展放缓。通胀忧虑一度推动10年期美债收益率自2023年以来首次突破5%。
Jack Clark says "most labs have different ways of being able to pull the plug", but says this may need to be a requirement.
中文摘要 Anthropic联合创始人杰克·克拉克对BBC表示，大多数实验室已有不同的「断电」方式，但AI「终止开关」可能需成为强制要求。
With Week 1 of the NFL season wrapping up, Adam Greenblatt, CEO of BetMGM, said the season has started enthusiastically and that early activity has been strong. He added that, from the bookmakers’ perspective, Sunday was not a favorable day because the fans came out ahead, leaving bettors happy and
中文摘要 BetMGM首席执行官亚当·格林布拉特表示，NFL赛季开局投注活动强劲，但周日对博彩公司不利，因球迷整体获利。
Oil edged higher as traders weighed risks to Middle East supplies, with a critical Saudi Arabian pipeline still offline after an attack.
中文摘要 油价小幅走高，交易员权衡中东供应风险，沙特一条关键输油管道遇袭后仍处于停运状态。
The cuts comes into effect in April, but the largest businesses will pay higher rates to fund the changes.
中文摘要 威尔士的酒吧、酒店和健身房将从4月起享受30%的商业税率减免，但大型企业将缴纳更高税率以弥补相关成本。
Authentic Brands Group, a licensor of everything from Elvis Presley’s image to the Reebok sportswear brand, could pursue an initial public stock offering as early as the first half of next year.
中文摘要 拥有猫王形象和锐步等品牌授权的Authentic Brands Group可能最早于明年上半年启动首次公开募股，并瞄准创作者经济。
11 回复 · 程序员 节点
9 回复 · 程序员 节点
4 回复 · 程序员 节点
11 回复 · 程序员 节点
39 回复 · 程序员 节点
16 回复 · 程序员 节点
4 回复 · Apple 节点
6 回复 · Apple 节点
19 回复 · Apple 节点
20 回复 · Apple 节点
该源今日无内容。