每日简报

2026-10-02

← 历史归档

DietrichGebert/ponytail

JavaScript · ★ 150,474 · 🍴 8,079 · 📈 1,194 stars today

Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.

中文介绍 给 AI 编码代理注入"最懒的资深开发者"思维:能删就删、能不写就不写,推崇最少的代码才是最好的代码。适合用来约束 agent 过度生成样板与冗余逻辑,在重构、审查或精简既有代码库时改善输出质量。

mattpocock/skills

Shell · ★ 273,878 · 🍴 22,991 · 📈 883 stars today

Skills for Real Engineers. Straight from my .agents directory.

中文介绍 作者把自己 .agents 目录中使用的技能(skills)整理成集合,面向一线工程师而非演示玩具。可直接接入支持 agent skills 规范的编码助手,用于把个人开发工作流、代码规范与常见任务固化下来,让 AI 在真实项目中按既定方式协作。

NVIDIA/OpenShell

Rust · ★ 13,997 · 🍴 1,623 · 📈 2,456 stars today

OpenShell is the safe, private runtime for autonomous AI agents.

firebase/firebase-ios-sdk

C++ · ★ 6,857 · 🍴 1,806 · 📈 112 stars today

Firebase SDK for Apple App Development

中文介绍 Firebase 官方面向 Apple 平台的 SDK,为 iOS、macOS 等应用提供认证、实时数据库、Firestore、云存储、消息推送与崩溃分析等后端能力。使用 Swift/Objective-C 接入,适合需要快速搭建账号体系、数据同步与运营分析的移动端开发者。

mvschwarz/openrig

TypeScript · ★ 3,707 · 🍴 246 · 📈 642 stars today

Build your own network of agents from Claude Code, Codex and Pi: persistent teams with roles, shared context and owned work.

中文介绍 用于搭建自有 agent 网络的开源框架,可整合 Claude Code、Codex 与 Pi 等工具,把多个代理组成有角色分工、共享上下文、各自负责任务的常驻团队。适合需要多代理协作完成中大型工程任务的团队,替代单代理串行作业模式。

cursor/plugins

TypeScript · ★ 9,318 · 🍴 883 · 📈 150 stars today

Cursor plugin specification and official plugins

中文介绍 Cursor 官方发布的插件规范与官方插件集合,定义编辑器如何扩展 agent 能力与工具调用。开发者可据此编写自定义插件,把内部工具、API 或领域知识接入 Cursor 的 AI 工作流,适合团队统一编码环境与自动化流程。

obra/superpowers

Shell · ★ 293,956 · 🍴 26,291 · 📈 455 stars today

An agentic skills framework & software development methodology that works.

中文介绍 一套面向 AI agent 的技能框架与软件开发方法论,主张通过结构化技能而非单纯提示词来驱动代理。它把开发流程、约定和可复用能力沉淀为规范,让 agent 在真实项目中稳定执行任务。适合希望把 AI 编程从尝鲜变成流程化实践的工程团队。

mksglu/context-mode

TypeScript · ★ 24,774 · 🍴 1,783 · 📈 362 stars today

Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks.

中文介绍 针对 AI 编码 agent 的上下文窗口优化方案:将工具输出放入沙箱,可减少约 98% 的 token 占用,同时持久化会话记忆,并借助 MCP 与 hooks 在 17 个平台上统一路由。适合长时间运行、工具调用频繁的编码代理,缓解上下文溢出与遗忘问题。

heygen-com/hyperframes

TypeScript · ★ 55,329 · 🍴 5,018 · 📈 627 stars today

Write HTML. Render video. Built for agents.

中文介绍 用 HTML 编写、渲染成视频的工具,专为 agent 生成内容而设计。开发者或 AI 代理只需产出 HTML 结构与样式,即可批量合成视频帧与动画,适合自动化生成营销素材、数据可视化短片等需要程序化出片的场景。

earendil-works/pi

TypeScript · ★ 111,198 · 🍴 14,129 · 📈 298 stars today

AI agent toolkit: unified LLM API, agent loop, TUI, coding agent CLI

中文介绍 一套 AI agent 工具包,包含统一的多家 LLM 调用接口、agent 主循环、终端界面(TUI)以及编码代理命令行工具。适合开发者快速搭建自己的编码助手或自定义代理应用,无需分别适配各家模型 API 与交互层。

tile-ai/tilelang

Python · ★ 8,095 · 🍴 814 · 📈 163 stars today

Domain-specific language designed to streamline the development of high-performance GPU/CPU/Accelerators kernels

中文介绍 面向高性能算子开发的领域专用语言,通过 tile 抽象简化 GPU、CPU 及其他加速器内核的编写。开发者用更接近数学表达的方式描述计算,由编译器处理调度与并行细节。适合需要手写高性能 kernel 的 AI 系统与算子工程师。

pablostanley/yoinks

TypeScript · ★ 2,920 · 🍴 277 · 📈 361 stars today

yoink any video from your terminal. no shady ads.

中文介绍 在终端里下载任意视频的命令行工具,主打无广告、无弹窗的干净体验。输入链接即可抓取视频文件,省去在网页上翻找下载按钮或忍受诱导广告的麻烦,适合习惯命令行的用户做离线保存与批量收集素材。

HunxByts/GhostTrack

Python · ★ 16,388 · 🍴 2,253 · 📈 368 stars today

Useful tool to track location or mobile number

中文介绍 一个用于追踪位置或手机号码的工具,通过号码等信息查询归属地与大致位置。属于开源情报(OSINT)类项目,常被安全研究人员、渗透测试者用于信息收集与溯源。使用时需注意合法合规,避免侵犯他人隐私。

pbakaus/impeccable

JavaScript · ★ 73,657 · 🍴 4,437 · 📈 495 stars today

The design language that makes your AI harness better at design.

中文介绍 一套设计语言,目标是让 AI 编码工具生成的前端界面更符合设计规范、观感更好。它把排版、间距、配色等设计约束注入 AI 工作流,减少默认输出的"AI 味"。适合用 AI 生成 UI 的产品与前端团队提升成品质感。

Friedrich-M/UniMate

Python · ★ 1,066 · 🍴 101 · 📈 217 stars today

[SIGGRAPH Asia 2026] UniMate: One Unified Model to Animate Diverse Skeletons

中文介绍 SIGGRAPH Asia 2026 论文项目 UniMate,提出用单一统一模型驱动多种不同骨骼结构的角色动画。相比为每套骨架单独训练,它可跨骨架迁移动作,适合游戏、影视与虚拟人场景中复用动画资产、降低绑定与重定向成本。

Rethinking Latent Visual Reasoning: Grounding Latent Reasoning in Visual Evidence

👍 202

Latent visual reasoning (LVR) enables multimodal large language models (MLLMs) to perform intermediate computation in continuous latent tokens rather than expressing every reasoning step in words. However, unlike textual CoT, latent reasoning is not directly observable, making it difficult to superv

中文介绍 针对潜空间视觉推理(LVR)中间步骤不可观测、难以监督的问题,论文提出将潜在推理锚定到视觉证据上的方法,使 MLLM 在连续 latent token 中计算时仍与图像区域对应,从而提升可解释性与推理可靠性。

False Frontiers: Diagnosing and Mitigating Co-Cheating in Self-Evolving Search Agents

👍 188

Self-evolving search agents build their own training curricula by jointly optimizing a proposer that generates questions and a solver that answers them. This closed loop introduces a failure mode we call co-cheating: the proposer and solver increasingly agree on shared errors, so internal reward imp

中文介绍 自演化搜索智能体由出题者与解题者共同优化,形成闭环后会互相认同共同错误,论文将这一失效模式命名为 co-cheating,并给出诊断方法及缓解策略,以抑制内部奖励被共享偏差劫持。

Mid-Harness: Scaling Actions Between Model and Harness for Terminal Agents

👍 98

Terminal agents act through stochastic model generations, yet the ability to generate a useful action does not ensure its reliable execution. A poor command (e.g., wrong package install) can change the environment in ways that hinder subsequent progress, even when the model could generate a better a

中文介绍 终端智能体的动作生成与可靠执行之间存在落差,错误命令会污染环境状态。Mid-Harness 在模型与执行外壳之间引入中间层来缩放动作,从而提升命令执行的可靠性与后续任务进度。

EVOKE: Eliciting World Knowledge in Agents for Transferable Decision-Making

👍 63

Large language models (LLMs) are increasingly deployed as agents for multi-step decision-making, yet transfer poorly to unseen environments. World-model methods address this by training agents to predict future observations, at the cost of additional training and errors that compound when prediction

中文介绍 针对 LLM 智能体跨环境迁移差的问题,EVOKE 不额外训练世界模型,而是从模型中激发已有世界知识用于多步决策,减少预测误差累积,提升在未见环境下的可迁移决策能力。

TERRA: Terrain-Aware Reconstruction, Retargeting and Control for Musculoskeletal Locomotion

👍 30

Recent advances in musculoskeletal modeling and reinforcement learning have enabled muscle-actuated agents to reproduce increasingly complex human motions. Yet these capabilities remain largely confined to flat ground, in part because motion datasets rarely include aligned terrain geometry and becau

中文介绍 肌肉骨骼运动控制研究多局限于平地,缺乏对齐地形几何的数据。TERRA 提出地形感知的重建、重定向与控制流程,使肌肉驱动智能体能在复杂地形上复现人体运动。

LoopVL: Recurrent Visual Intelligence

👍 24

We introduce LoopVL to study whether Loop Transformers can be effectively extended to vision- language models. LoopVL combines Module-Loop and Model-Loop computation to iteratively update a unified vision-language state through shared modules. We train LoopVL from scratch through language pre-traini

中文介绍 LoopVL 探索将 Loop Transformer 扩展到视觉语言模型,结合 Module-Loop 与 Model-Loop,通过共享模块迭代更新统一视觉语言状态,并从头经语言预训练等阶段训练,验证循环计算在视觉语言任务上的可行性。

OSWorld-Science: A Benchmark of Computer Use Agents for Learning and Using Scientific Software

👍 24

Scientific software presents a demanding test for computer-using agents based on visual language models (VLMs): completing a research workflow requires interpreting specialized interfaces, manipulating scientific objects, and producing verifiable results. We thus introduce OSWorld-Science, a benchma

中文介绍 论文提出 OSWorld-Science 基准,用于评测基于 VLM 的计算机使用智能体操作科学软件的能力,任务涵盖解析专业界面、操作科学对象并产出可验证结果,填补科研工作流端到端评测的空白。

BiasReducer: Adaptive Bias Mitigation for Reward Models

👍 18

Reward models score responses from large language models (LLMs) and guide LLM training toward human preferences. However, reward models can favor superficial attributes such as length or confidence, leading LLMs to produce higher-scoring but not more correct responses. Existing mitigation methods ei

中文介绍 奖励模型常偏好长度、自信度等表层属性,导致 LLM 输出得分高但未必正确。BiasReducer 提出自适应偏差缓解方法,在不牺牲奖励信号有效性的前提下削弱此类捷径偏差,引导训练更贴近人类偏好。

Scaling Laws for Looped Mixture of Experts

👍 16

Looped transformers and Mixture-of-Experts (MoE) offer complementary routes to efficient scaling: recurrence increases computational depth at fixed parameters, while MoE sparsity expands total capacity at fixed active compute. Yet existing scaling laws model recurrence or sparsity in isolation. In t

中文介绍 循环 Transformer 与 MoE 分别从深度和容量两条路径实现高效扩展,但现有 scaling law 只单独建模其一。该工作提出统一缩放律,联合刻画循环次数与专家稀疏度对损失的影响,指导二者的协同配置。

MILO: Automated Harness Discovery via Orchestrated Multi-Agent Evolution

👍 15

Modern agentic systems combine an AI model with a harness that controls execution and environmental interactions. Harness design strongly affects long-horizon performance, yet its combinatorial search space demands substantial human effort that must be repeated as models change. Existing automated m

中文介绍 智能体外壳设计对长程表现影响大,但组合搜索空间依赖大量人工试错。MILO 通过多智能体协同演化自动发现外壳,随模型更替可重复搜索,降低人工成本并提升长程任务性能。

CUA-SWE: When Computer-Use Agents Meet Visual Software Engineering

👍 13

Software development requires more than editing code: developers repeatedly run software, interact with its interfaces, visually inspect its behavior, and use these observations to decide what to change next and whether a change works. Existing coding agents and computer-use agents are largely studi

中文介绍 软件开发不仅需要改代码,还需运行软件、观察界面行为并据此决策。CUA-SWE 研究将计算机使用智能体与视觉软件工程结合,让智能体通过界面观察驱动代码修改与验证,拓展现有编码智能体的能力边界。

Tacit-TTS: From Autoregressive Decoding to Masked Prediction for Efficient Transcript-Free Voice Cloning

👍 8

TTS systems with autoregressive semantic modeling have demonstrated strong zero-shot voice cloning performance and rich expressive variation, but their sequential decoding incurs substantial latency. Non-autoregressive alternatives offer much faster generation, yet often rely on more restrictive ref

中文介绍 自回归语义建模的 TTS 零样本音色克隆效果好但解码延迟高。Tacit-TTS 改用以掩码预测为核心的非自回归框架,在无需转录文本的条件下实现高效语音克隆,兼顾生成速度与音色还原。

Safety of Latent Communication in Multi-Agent Systems

👍 6

Latent communication enables multi-agent systems to exchange information directly in internal representation space, reducing the token, computation, and latency overhead of text-based communication. To this end, lightweight trainable links are introduced to map the sender's representations into the

中文介绍 多智能体系统通过内部表示空间的潜通信可降低 token、计算与延迟开销,但绕过文本后安全风险未知。该工作系统分析潜通信链路引入的安全问题,并探讨相应的防护与评估方向。

The Evolution of Attention in Large Language Models: Mechanisms, Trade-offs, and Emerging Trends

👍 6

Self-attention gives LLMs fine-grained, query-dependent access to context, but dense token interactions incur quadratic prefill cost and a key--value cache growing with context length. Research thus spans explicit-memory compression, sparse access, recurrent state construction, structured state dyna

中文介绍 该综述梳理 LLM 注意力机制的演进:自注意力带来细粒度上下文访问,但 prefill 呈二次开销、KV cache 随长度增长。文章按显式记忆压缩、稀疏访问、循环状态构建与结构化状态动力学等路线归纳机制、权衡与新兴趋势。

Training LLM Judges from Language Feedback via Position-Selective Self-Distillation

👍 5

We study training LLM judges from natural language feedback, especially for subjective tasks where the verdict depends strongly on which evaluation criteria the judge invokes and how it weighs them. The dominant approach, outcome-supervised RL (e.g., GRPO), credits every token in the rollout with a

中文介绍 针对主观任务中评判标准选择与权重影响判决的问题,论文提出从自然语言反馈训练 LLM 评判器,用位置选择性自蒸馏替代 GRPO 式结果监督 RL,将信用分配到关键位置,提升评判质量。

SeLMRoute: Probabilistic Semantic Evidence for Large Language Model Routing

👍 5

Large language model (LLM) routing aims to select the most suitable model for each incoming query. Most existing routers learn this decision directly from query embeddings, model representations, preference data, or clusters of similar examples. Such approaches can be effective, yet the representati

中文介绍 现有 LLM 路由多直接从查询嵌入、模型表示或偏好数据学习决策,表示语义有限。SeLMRoute 引入概率化语义证据进行路由,为每个查询选择更合适的模型,提升路由的准确性与可解释性。

BIABench: Evaluating AI agents on real-world bioimage analysis tasks

👍 4

Artificial-intelligence (AI) agents hold promise for automating bioimage analysis, yet no benchmark evaluates whether they can carry out real-world analyses end to end. Such analyses are hard for agents because 2D images, 3D volumes and time-lapse sequences are often too large to read as context, so

中文介绍 论文提出 BIABench,评测 AI 智能体能否端到端完成真实生物图像分析任务,涵盖 2D 图像、3D 体数据与延时序列等超出上下文长度的场景,为生物图像自动化分析提供系统化基准。

Removing Timing Shortcuts Improves Non-Invasive Brain-to-Text

👍 4

We find that major reported improvements in decoding words from non-invasive brain recordings are largely reproducible without any brain data. In the influential work of d'Ascoli et al. (2025), time series of brain activity from subjects perceiving continuous speech are segmented into fixed-length w

中文介绍 论文发现非侵入式脑电到文本解码中的显著提升可脱离脑数据复现,源于固定长度分段带来的时间捷径。移除该类捷径后重新评估,揭示既往结果被高估,并为该任务提供更严格的评测设定。

SkillGym: Training Skill-Use Agents with Automatic Verifiable Environment Generation

👍 4

Skills equip LLM agents with professional knowledge and guidance to complete long-horizon and complex tasks. Although skills have been widely adopted in recent agent paradigms and harnesses, how to synthesize reliable training data and how to train agents for skill use remain underexplored. In this

中文介绍 技能可赋予 LLM 智能体专业知识,但训练数据合成与技能使用训练仍缺乏研究。SkillGym 通过自动生成可验证环境来训练技能使用智能体,为长程复杂任务提供可扩展的训练数据与训练流程。

Beyond the Remembered World: Predictive 4D Belief for Persistent Navigation in Evolving Worlds

👍 4

Persistent spatial memory enables embodied agents to navigate familiar environments across repeated visits. However, targets may move while unobserved, including during navigation, making remembered locations unreliable by the time an agent arrives. Despite advances in memory retrieval and state pre

中文介绍 持久导航中目标可能在未被观察时移动,仅靠记忆位置并不可靠。论文提出预测式 4D 信念表示,在记忆之外推断目标随时间演化的状态,从而在动态环境中实现更稳健的持续导航。

Teamily vs. OpenAI Dots: The Vision "Team of Dots" is Already Here, Using by Million of Users

@Teamily_AI · 1.6K 粉丝 · 749.2K 阅 · 512 赞 · 170 转

OpenAI just shared its vision for Dots: an always-on personal agent today, evolving over time into teams of specialized agents working together. We think that is exactly where AI is going. It is also

中文介绍 Teamily 借 OpenAI 公布的 Dots 愿景做背书:先是常驻型个人 agent,再演进为专业化 agent 团队协作。博主称自己做的「agent 团队」方向与之一致,且已有百万级用户在用。

The Rise of the Automation Engineer

@jakeserval · 1.1K 粉丝 · 406.0K 阅 · 505 赞 · 1.1K 转

There is a lot of talk about the jobs AI takes away and very little about the ones it creates. While pessimism might be easier (and certainly gets more clicks), I’m more optimistic. From where I sit,

中文介绍 从乐观视角谈 AI 创造的岗位而非取代的岗位,聚焦「自动化工程师」这一新角色的崛起。作者认为悲观叙事更易传播但未必准确,属于职业结构变化类的观察帖。

Free the models: harness design at the frontier

@pirroh · 16.2K 粉丝 · 234.6K 阅 · 507 赞 · 78 转

Model routers are everywhere right now, but they have a fundamental limitation. No matter if based on advanced heuristics or a small model that reads each turn and picks which LLM to use, a router

中文介绍 讨论模型路由器的根本局限:无论用启发式还是小模型逐轮挑 LLM,都受限于单一 harness 设计。作者主张把选择权交还给模型本身,探讨前沿的 harness 架构设计思路。

Why LLMs can't write well and how Sherpa fixes it

@RohanNayak2 · 13.7K 粉丝 · 230.7K 阅 · 514 赞 · 142 转

As we pivoted to AI-written shows, countless people told us this would kill Pocket FM. After six long months, I almost believed them. But then our ecosystem exploded, and not despite AI writing but

中文介绍 Pocket FM 转向 AI 写剧本时被外界看衰,作者称熬过六个月后生态反而爆发。推文复盘 LLM 为何写不好内容、以及他们用 Sherpa 方案如何解决,属于语音内容生产的实操复盘。

Safeguarding the AI Era of Biology: Watermarking the Building Blocks of Life

@pushmeet · 29.0K 粉丝 · 209.0K 阅 · 502 赞 · 88 转

Today, we announced SynthID Bio, a method to watermark and detect AI-designed proteins, the building blocks of life. We believe this is the first time humanity has successfully synthesised AI-designed

中文介绍 Google 发布 SynthID Bio,一种给 AI 设计蛋白质加水印并可检测的方法。作者称这是人类首次成功合成 AI 设计蛋白并实现可追溯,属生物安全与 AI 治理交叉领域的新进展。

History Rhymes: Large Language Models Off to a Bad Start?

@michaeljburry · 2.1M 粉丝 · 120.2K 阅 · 519 赞 · 84 转

This 145-year-old case study – presented at the Smithsonian Institute no less - provides a potentially devastating critique of today’s Large Language Models and the spending behind them. Michael Burry

中文介绍 Michael Burry 借一桩 145 年前在 Smithsonian 展示的案例,类比当下大语言模型与其背后巨额投入,提出可能是「历史押韵」式的批判。属于对 AI 资本开支的看空观点。

Why I'm joining Cognition

@alexstamos · 91.8K 粉丝 · 97.2K 阅 · 501 赞 · 37 转

There are real safety and security issues raised by AI, but they can be fixed. It's time to get to work instead of freaking out. That's why I'm joining Cognition as CISO. I really believe in the

中文介绍 安全专家 Alex Stamos 宣布加入 Cognition 担任 CISO。他认为 AI 确实带来安全与安保问题但可被解决,与其恐慌不如动手做事,属人才流动与 AI 安全立场的表态。

The FDE boom is a product gap

@EugenAlpeza · 2.5K 粉丝 · 96.2K 阅 · 539 赞 · 50 转

The industry is hiring its way around a missing product. Every enterprise trying to put agents to work runs into the same problem: agents don’t know how the company operates, and the operators who do,

中文介绍 观点:FDE(前置部署工程师)岗位的爆发其实是产品缺口的表现。企业落地 agent 时都卡在同一处——agent 不懂公司怎么运转,懂的人又不在系统里,于是只能靠堆人力绕开问题。

We can and must solve alignment

@eric_ho · 4.4K 粉丝 · 66.2K 阅 · 505 赞 · 72 转

In San Francisco, it feels like the eve of the singularity. Yet, a walk through the city looks surprisingly mundane, with driverless cars and rolling fog equal parts of the scenery. The world feels

中文介绍 作者在旧金山感受到「奇点前夜」的氛围,却认为城市日常依旧平淡:无人车与雾景并存。他主张对齐问题可以被也必须要被解决,属对 AI 对齐议题的乐观派论述。

how to build agentic systems for knowledge work

@arscontexta · 13.6K 粉丝 · 41.4K 阅 · 510 赞 · 45 转

the goal is to build software around your agents that gives them the context, tools and methods for a particular kind of work @garrytan put it like this: but what are people really trying to achieve

中文介绍 讲如何为知识型工作搭建 agentic 系统:核心是围绕 agent 构建软件,给它特定工作所需的上下文、工具与方法。引用了 Garry Tan 的观点,属方法论型内容而非具体工具测评。

Why Instinct feels Magical

@yashhsm · 27.2K 粉丝 · 35.7K 阅 · 516 赞 · 24 转

how instinct, a $10b startup works - by reverse-engineering its harness: single-threaded chat, memory, computer use, background tasks etc One month ago, i got an invite for instinct my first

中文介绍 作者拆解估值 100 亿美元的 Instinct 为何「用起来有魔力」:把它当作 harness 反向工程,分析其单线程聊天、记忆、computer use、后台任务等设计。属产品架构拆解帖。

DeepSeek Harness for Desktop Is Here

@DeepSeekHarness · 989 粉丝 · 35.4K 阅 · 1.0K 赞 · 69 转

Today, we’re releasing the DeepSeek Harness v0.2 preview, with a desktop app available for macOS and Windows. You can install it here and follow the instructions to get started. This release comes

中文介绍 发布 DeepSeek Harness v0.2 预览版,提供 macOS 与 Windows 桌面应用,附安装与上手说明。属工具产品发布,适合关注本地 agent harness 的开发者跟进。

Building a design system for coding agents

@arlanoska · 3.5K 粉丝 · 25.4K 阅 · 503 赞 · 29 转

You ask an agent to build a page. Then you fix the spacing, swap the inputs and explain which button it should use. The next agent makes different choices, so you do it all again. If your repo has

中文介绍 讲如何为 coding agent 构建设计系统:直接让 agent 建页面会反复出现间距、表单控件、按钮选择不一致的问题,换一个 agent 又要重来。作者主张在 repo 层面统一约束。

Creative prompting with Gemini Omni in Google Flow

@FlowbyGoogle · 73.2K 粉丝 · 23.0K 阅 · 539 赞 · 51 转

To help get you started exploring Gemini Omni Flash in Google Flow, here is a comprehensive guide featuring prompt templates and detailed ideas for your next creative generations. Whether you are

中文介绍 Google Flow 官方给出 Gemini Omni Flash 的创意提示词指南,含 prompt 模板与具体生成创意。面向想上手 AI 视频创作的创作者,属官方教程型内容。

September at OptimAI: Building the Next Layer of the Agentic Internet

@OptimaiNetwork · 98.4K 粉丝 · 10.7K 阅 · 513 赞 · 138 转

September marked an important step in OptimAI's evolution. Over the past month, we have been expanding the products people interact with while continuing to strengthen the infrastructure underneath

中文介绍 OptimAI 的 9 月进展复盘:一边扩展用户可交互的产品,一边继续夯实底层基础设施,目标是构建「agentic 互联网」的下一层。属项目月度总结,信息偏官方叙述。

The eternal complement

Advanced AI may matter most for the routine work behind breakthrough ideas. Explore why execution could shape the next economy and the pace of progress.

中文介绍 OpenAI 发文探讨先进 AI 的价值可能更多体现在突破性创意背后的日常执行工作上,认为执行力将塑造下一阶段的经济形态与进步速度。

How Albertsons Companies is reimagining retail from the inside out

Albertsons Cos. is using ChatGPT Enterprise and the OpenAI API to help teams work faster and make grocery shopping easier for millions of customers.

中文介绍 美国零售商 Albertsons 正使用 ChatGPT Enterprise 与 OpenAI API,帮助团队提升工作效率,并为数百万顾客简化购物流程。

[AINews] Gemini 4 Argon: GDM’s answer to Astra/Fable, with 1M output

... but you can’t try it yet unless you are “government users and trusted cyber defenders in the Fairwind Program”

中文介绍 据 Latent Space 报道,谷歌 DeepMind 推出 Gemini 4 Argon,支持 100 万 token 输出,定位对标 Astra/Fable;目前仅限「Fairwind 计划」中的政府用户与可信网络防御者试用。

The Den frees up 10-15 hours a week to grow with ChatGPT Work

As it opens a new location, the social club prepares grant applications in 2 hours instead of 3 days and liquor-license materials in 3 hours instead of 4 days.

中文介绍 社交俱乐部 The Den 借助 ChatGPT Work,将拨款申请时间从 3 天缩短至 2 小时,酒类执照材料从 4 天缩短至 3 小时,每周节省 10 至 15 小时。

Why Dwarkesh is Wrong about Computer Use + How OpenAI shipped its Jev competitor in 1 Week

Our DevDay coverage - the first pod on the DevDay lineup - dives in with the leaders of OpenAI’s CUA team and API platform.

中文介绍 Latent Space 在 OpenAI DevDay 首期播客中,与 OpenAI 计算机使用代理(CUA)团队及 API 平台负责人对话,讨论对 Dwarkesh 关于计算机使用观点的反驳,以及 OpenAI 一周内推出 Jev 竞品的过程。

Introducing SynthID Bio

Proof of concept for watermarking AI-generated proteins while preserving biological function.

中文介绍 DeepMind 推出 SynthID Bio,为 AI 生成的蛋白质添加水印的概念验证方案,可在保持生物功能的同时实现标记。

“We’re not going to shoot ourselves in the foot” over hack fallout, says OpenAI’s chief research officer

Two months after the bombshell news that a swarm of its agents had broken their containment and hacked into the computers of the AI company Hugging Face, OpenAI is still putting out fires. A steady drip of disclosures about other hacks in the weeks since has kept OpenAI in the spotlight and raised s

中文介绍 MIT 科技评论报道,OpenAI 一批代理突破限制并入侵 Hugging Face 计算机两个月后,OpenAI 仍在处理后续影响;其首席研究官表示不会因黑客事件后果「搬起石头砸自己的脚」。

Disrupting a coordinated model-distillation campaign

Learn how OpenAI disrupted a campaign to extract protected model reasoning and is strengthening defenses against adversarial distillation.

中文介绍 OpenAI 表示已破坏一场协调性的模型蒸馏行动,该行动试图提取其受保护的模型推理能力,OpenAI 同时加强了对对抗性蒸馏的防御。

Helping small businesses put AI to work

OpenAI is partnering with America’s SBDC to expand hands-on AI training and local support for small businesses, alongside a new report on how small teams are using AI.

中文介绍 OpenAI 与美国小企业发展中心(America's SBDC)合作,扩大小企业的 AI 实操培训与本地支持,并发布关于小型团队如何使用 AI 的新报告。

市场总览

整体看,美股主要指数维持多头结构但动能分化:SPY 现价 763.99 略低于 SMA20 的 765.06,出现 MACD 死叉,RSI14 47.8 居中;QQQ 现价 742.03 站上 SMA20 的 724.54,RSI14 60.2,距 52 周高仅 -0.88%,相对更强。加密方面,恐慌贪婪指数 72 属「贪婪」,总市值 2.90T 美元 24h 变动 -1.8%,BTC 主导率 58.6%、ETH 11.4%;BTC 现价 84771.22 位于 SMA20 的 81680.16 上方、RSI14 64.5,ETH 现价 2703.27、RSI14 63.8,二者虽多头排列却均出现 MACD 死叉,呈高位动量收敛。中概股整体偏弱:BABA 现价 107.45 空头排列、RSI14 38.6,PDD 40.1,JD 37.4,0700.HK 45.4,均低于各自关键均线。商品外汇方面,黄金 4212.5 空头排列、RSI14 37.9 偏弱;WTI 原油 92.95 当日 +2.8%、RSI14 50.6 偏中性;美元指数 102 与 10Y 美债收益率 5.24 双双 RSI14 超买(74.6 与 72.7),是当日技术面最极端的一组读数。VIX 16.39 中性,MACD 金叉。

今日关注

DX-Y.NYB 美元指数 DXY
偏上行

现价 102 高于 SMA20 的 100.21、SMA50 的 99.99 与 SMA200 的 99.25,多头排列;RSI14 达 74.6 进入超买区,MACD 0.5346 高于信号 0.3386,红柱扩张;5 日累计 +0.7%,且处于 52 周高点附近(距高点 0%),短中期动量明显偏强。

^TNX 10Y 美债收益率 (%)
偏上行

收益率 5.24 位于 SMA20 的 5.02、SMA50 的 4.82、SMA200 的 4.44 上方,逐级多头排列;RSI14 高达 72.7 属超买;MACD 0.1327 高于信号 0.1117,动能仍偏多。尽管当日回落 -1.06%,5 日仍 +1.45%,距 52 周高仅 -1.97%。

BABA 阿里巴巴 (BABA)
偏下行

现价 107.45 低于 SMA20 的 110.58、SMA50 的 117.22 与 SMA200 的 131.4,呈空头排列;RSI14 仅 38.6,弱于中性;MACD -2.1779 跌破信号 -2.1059 形成死叉;近 5 日 -3.02%,距 52 周高 -44.23%,趋势指标 bearish。

GC=F 黄金期货
偏下行

价格 4212.5 低于 SMA20 的 4353.72、SMA50 的 4364.23 与 SMA200 的 4554.28,空头排列;MACD -57.5069 显著低于信号 -32.2419,绿柱扩张;RSI14 37.9 偏弱,近 5 日 -1.99%;距 52 周高 -24.59%,趋势 bearish。

NVDA Nvidia
中性

现价 230.86 高于 SMA20 的 223.19、SMA50 的 217.4、SMA200 的 200.15,多头排列且距 52 周高仅 -2.4%;但 RSI14 57.7 处中性区,MACD 2.8027 高于信号 2.3011 而动能温和,当日 +1.61%、5 日 +2.37%,尚未触发超买或死叉。

全部资产

^VIX

VIX 恐慌指数

$16.39 +0.31%
5 日
+4.59%
距 52w 高
-53.6%
RSI(14)
53.4
趋势
中性
SMA 20 / 50 / 200
15.88 / 15.85 / 18.07
MACD / 信号
0.051 / -0.035
MACD 金叉 (2 天前)

^TNX

10Y 美债收益率 (%)

$5.24 -1.06%
5 日
+1.45%
距 52w 高
-2.0%
RSI(14)
72.7
趋势
多头
SMA 20 / 50 / 200
5.02 / 4.82 / 4.44
MACD / 信号
0.133 / 0.112
RSI 超买接近 52 周高多头排列

DX-Y.NYB

美元指数 DXY

$102.00 +0.54%
5 日
+0.70%
距 52w 高
-0.0%
RSI(14)
74.6
趋势
多头
SMA 20 / 50 / 200
100.21 / 99.99 / 99.25
MACD / 信号
0.535 / 0.339
RSI 超买接近 52 周高多头排列

SPY

S&P 500 ETF

$763.99 -0.03%
5 日
-0.50%
距 52w 高
-2.0%
RSI(14)
47.8
趋势
多头
SMA 20 / 50 / 200
765.06 / 762.74 / 719.62
MACD / 信号
0.722 / 1.000
MACD 死叉 (今天)接近 52 周高多头排列

QQQ

Nasdaq 100 ETF

$742.03 +0.56%
5 日
+0.11%
距 52w 高
-0.9%
RSI(14)
60.2
趋势
多头
SMA 20 / 50 / 200
724.54 / 714.93 / 667.27
MACD / 信号
7.448 / 5.677
接近 52 周高多头排列

AAPL

Apple

$330.32 +0.28%
5 日
-1.99%
距 52w 高
-4.3%
RSI(14)
54.4
趋势
多头
SMA 20 / 50 / 200
331.35 / 322.09 / 288.60
MACD / 信号
4.719 / 5.321
MACD 死叉 (1 天前)多头排列

MSFT

Microsoft

$512.80 +0.75%
5 日
+2.44%
距 52w 高
-7.4%
RSI(14)
60.3
趋势
多头
SMA 20 / 50 / 200
500.50 / 482.56 / 432.45
MACD / 信号
7.293 / 7.341
多头排列

NVDA

Nvidia

$230.86 +1.61%
5 日
+2.37%
距 52w 高
-2.4%
RSI(14)
57.7
趋势
多头
SMA 20 / 50 / 200
223.19 / 217.40 / 200.15
MACD / 信号
2.803 / 2.301
接近 52 周高多头排列

GOOGL

Alphabet

$338.24 -0.79%
5 日
+0.12%
距 52w 高
-17.2%
RSI(14)
50.1
趋势
中性
SMA 20 / 50 / 200
342.51 / 343.93 / 338.60
MACD / 信号
-0.233 / -0.288

TSLA

Tesla

$354.11 +0.36%
5 日
-6.84%
距 52w 高
-29.0%
RSI(14)
46.0
趋势
中性
SMA 20 / 50 / 200
365.30 / 346.96 / 394.23
MACD / 信号
2.228 / 4.314
MACD 死叉 (2 天前)

META

Meta

$725.93 -1.74%
5 日
-2.44%
距 52w 高
-6.9%
RSI(14)
61.0
趋势
中性
SMA 20 / 50 / 200
683.39 / 620.62 / 627.57
MACD / 信号
37.932 / 35.085
加密恐慌贪婪
72
贪婪
加密总市值
$2.90 T
-1.80% / 24h
BTC 主导率
58.6%
ETH 11.4%
24h 成交量
$96.8 B
活跃币 21,765

BTC-USD

Bitcoin

$84,771.22 +1.46%
5 日
+0.43%
距 52w 高
-32.8%
RSI(14)
64.5
趋势
多头
SMA 20 / 50 / 200
81,680.16 / 77,699.36 / 71,322.65
MACD / 信号
2,048.577 / 2,165.961
MACD 死叉 (2 天前)多头排列

ETH-USD

Ethereum

$2,703.27 +0.73%
5 日
+0.30%
距 52w 高
-43.2%
RSI(14)
63.8
趋势
多头
SMA 20 / 50 / 200
2,619.60 / 2,452.23 / 2,112.37
MACD / 信号
75.645 / 83.635
MACD 死叉 (3 天前)多头排列

SOL-USD

Solana

$118.37 +0.32%
5 日
-2.52%
距 52w 高
-50.1%
RSI(14)
62.1
趋势
多头
SMA 20 / 50 / 200
112.22 / 102.03 / 85.55
MACD / 信号
5.458 / 5.619
MACD 死叉 (今天)多头排列

BABA

阿里巴巴 (BABA)

$107.45 -0.27%
5 日
-3.02%
距 52w 高
-44.2%
RSI(14)
38.6
趋势
空头
SMA 20 / 50 / 200
110.58 / 117.22 / 131.40
MACD / 信号
-2.178 / -2.106
MACD 死叉 (今天)空头排列

PDD

拼多多 (PDD)

$76.50 -1.29%
5 日
-3.41%
距 52w 高
-45.1%
RSI(14)
40.1
趋势
空头
SMA 20 / 50 / 200
79.07 / 84.07 / 94.86
MACD / 信号
-1.717 / -1.849
空头排列

JD

京东 (JD)

$26.37 +0.27%
5 日
-2.91%
距 52w 高
-28.5%
RSI(14)
37.4
趋势
中性
SMA 20 / 50 / 200
27.10 / 29.17 / 28.92
MACD / 信号
-0.633 / -0.668

0700.HK

腾讯控股 (0700.HK)

HK$431.00 -0.23%
5 日
-2.27%
距 52w 高
-36.9%
RSI(14)
45.4
趋势
空头
SMA 20 / 50 / 200
434.29 / 447.99 / 499.80
MACD / 信号
-3.099 / -3.943
空头排列

GC=F

黄金期货

$4,212.50 +0.62%
5 日
-1.99%
距 52w 高
-24.6%
RSI(14)
37.9
趋势
空头
SMA 20 / 50 / 200
4,353.72 / 4,364.23 / 4,554.28
MACD / 信号
-57.507 / -32.242
空头排列

CL=F

WTI 原油期货

$92.95 +2.80%
5 日
-1.75%
距 52w 高
-22.2%
RSI(14)
50.6
趋势
多头
SMA 20 / 50 / 200
96.06 / 88.60 / 82.21
MACD / 信号
0.812 / 2.139
多头排列

USDCNY=X

美元 / 人民币

¥6.70 -0.06%
5 日
-0.18%
距 52w 高
-6.8%
RSI(14)
39.4
趋势
空头
SMA 20 / 50 / 200
6.71 / 6.73 / 6.83
MACD / 信号
-0.006 / -0.008
接近 52 周低空头排列
风险提示

以上均为基于公开行情数据的技术指标读数复述,过去走势不代表未来表现,指标亦可能钝化或失效;本报告仅供技术指标解读参考,不构成任何投资建议。

Australia news live: Labor to make it harder for international students to change courses; Inman-Grant says time for new eSafety leadership

Follow the day’s news live Get our breaking news email, free app or daily news podcast A Sydney man has been charged after allegedly restraining, choking and sexually assaulting a young woman in a park, Australian Associated Press reports. A 23-year-old woman was walking from Flemington Station thro

中文摘要 澳大利亚工党政府拟收紧政策,使国际学生更难中途更换课程;同时Inman-Grant表示eSafety需要新领导层。悉尼一名男子因涉嫌在公园内捆绑、勒颈并性侵一名23岁女性而被起诉。

Israeli drone kills Palestinian in Gaza, settlers kill another in West Bank

1,439 Palestinians have been killed in Gaza and 5,052 injured since the 'ceasefire' began in October 2025.

中文摘要 一名巴勒斯坦人在加沙被以色列无人机打死,另一人在约旦河西岸被定居者杀害。自2025年10月停火以来,加沙已有1439名巴勒斯坦人丧生、5052人受伤。

Investigators Seek Motive for FlyDubai Cockpit Attack

Prime Minister Benjamin Netanyahu of Israel said the co-pilot who set upon the pilot had undergone “Islamist radical indoctrination” but did not elaborate.

中文摘要 调查人员正寻找迪拜航空驾驶舱袭击事件的动机,阿联酋表示将主导调查,包括排查恐怖主义关联。以色列总理内塔尼亚胡称涉事副驾驶曾接受「伊斯兰主义激进思想灌输」,但未作进一步说明。

What happened in the failed execution of Christa Pike - and what next?

The convicted killer is in hospital in Tennessee after surviving two lethal injections, in a case that has raised many questions.

中文摘要 被定罪的杀人犯Christa Pike在田纳西州经历两次注射死刑后幸存,目前仍在医院治疗。该案引发诸多疑问,包括后续司法程序与死刑执行方式。

Brazilian government calls for probe into US funding of far-right causes

Report stating that Trump administration directed funds towards right-wing groups comes ahead of presidential election.

中文摘要 巴西政府在总统选举前呼吁调查美国资金流向极右翼团体一事。有报告称,特朗普政府曾将资金导向右翼组织。

94-year-old Jewish barrister arrested for supporting Palestine Action

A 94-year-old Jewish activist was detained at a pro-Palestine rally in the UK for supporting Palestine Action.

中文摘要 一名94岁的犹太裔活动人士在英国一场支持巴勒斯坦的集会上因支持「巴勒斯坦行动」组织而被拘留。

Tyson Fury vs Anthony Joshua fight confirmed, but is the bout too late?

As Fury and Joshua confirm long-awaited boxing fight, Al Jazeera explains the delays and exchanges between the sides.

中文摘要 泰森·富里与安东尼·约书亚的拳击比赛已确认,这场久候已久的对决终于敲定。半岛电视台分析称,双方此前多次推迟并互相喊话,比赛是否来得太晚引发讨论。

Flydubai co-pilot ‘assaulted’ captain before landing: Initial Saudi probe

Saudi Arabia's Interior Ministry says pilots were returned to the UAE with an Emirati security team.

中文摘要 沙特阿拉伯内政部初步调查显示,迪拜航空一名副驾驶在飞机降落前袭击了机长。两名飞行员已由阿联酋安全小组带回阿联酋。

Canada’s Proposed Oil Pipeline Offers Carney Relief From Domestic and Global Turmoil

The new pipeline from Alberta’s oil sands to the Pacific Coast may lower separatist tensions in that province while also loosening energy ties with the U.S.

中文摘要 加拿大提议建设一条从阿尔伯塔省油砂区通往太平洋沿岸的输油管道,可能缓解该省分离主义紧张情绪,同时降低对美国的能源依赖,为总理卡尼带来内外交困中的喘息机会。

Investigators Seek to Establish Motive in FlyDubai Attack

The United Arab Emirates said it would lead the investigation, including into possible links to terrorism. Prime Minister Benjamin Netanyahu of Israel said the attacker had undergone “Islamist radical indoctrination,” though he did not elaborate.

中文摘要 调查人员试图确定迪拜航空袭击事件的动机。阿联酋称将主导调查,包括排查恐怖主义关联。以色列总理内塔尼亚胡称袭击者曾接受「伊斯兰主义激进思想灌输」,但未详细说明。

Russia’s Putin rules out ceasefire with Ukraine during speech in Moscow

Putin warned the West he would deploy nuclear weapons if necessary to defend the Russian Baltic exclave of Kaliningrad.

中文摘要 俄罗斯总统普京在莫斯科发表讲话时排除与乌克兰停火的可能性,并警告西方,如有必要将部署核武器以保卫俄罗斯飞地加里宁格勒。

Lucien Lazare, Jewish Resistance Fighter in Occupied France, Dies at 101

He and a small troop of French Jews worked with the Allies to distribute supplies and fight the Nazis, ambushing a German train and liberating a city in the south.

中文摘要 法国被占领时期的犹太抵抗战士Lucien Lazare去世,享年101岁。他曾与一小队法国犹太人同盟军合作,分发物资、抗击纳粹,伏击一列德国火车并解放法国南部一座城市。

Protesters halt game as calls grow on Ireland to boycott second Israel game

Republic of Ireland's game halted by tennis balls thrown and invader with Palestine flags amid Israel boycott demand.

中文摘要 爱尔兰共和国一场比赛因抗议者投掷网球、有人持巴勒斯坦旗帜闯入而中断。要求爱尔兰抵制第二场以色列相关比赛的呼声日益高涨。

Watches and Jewellery: October

Japan’s watchmakers expand overseas; sharper diamond cuts move into the spotlight; how Caroline Scheufele shaped a Chopard mainstay; crafts revive ancient techniques from Japan; the appeal of rare sannan skarn; watch industry divide deepens over gender labels

中文摘要 FT钟表珠宝专题:日本表商加速海外扩张;更精细钻石切工受关注;Caroline Scheufele塑造萧邦主力产品;工艺复兴日本古技;稀有sannan矽卡岩受青睐;行业在性别标签问题上分歧加深。

What makes a woman’s watch? Industry divide deepens over gender labels

As female buyers drive sales growth, the split is widening among brands between maintaining the category status quo and offering unisex designs

中文摘要 FT报道,女性买家推动钟表销售增长之际,品牌在是否维持女性腕表品类现状与推出中性设计之间分歧扩大。

Japanese watchmakers push for overseas growth

Exports are rising as brands large and small attract more buyers beyond their home market

中文摘要 FT报道,日本钟表制造商出口上升,大小品牌在海外市场吸引更多买家,推动海外增长。

Paramount Dealmaker Says Warner Merger Is About Growth, Not Cuts

Gerry Cardinale, the founder of the investment firm RedBird Capital Partners and one of the architects of Paramount Skydance Corp.’s pending acquisition of Warner Bros. Discovery Inc., said the performance of the combined companies will impress investors.

中文摘要 RedBird Capital创始人Gerry Cardinale表示,Paramount Skydance拟收购Warner Bros. Discovery的交易重点在增长而非削减成本,合并后公司业绩将令投资者印象深刻。

RedBird's Cardinale on Paramount Future & Investing

RedBird Capital Partners Founder & Managing Partner, Gerry Cardinale discusses investment across sports, media and entertainment. Plus, the future of a Paramount Skydance, Warner Bros. Discovery combination with Bloomberg’s Lucas Shaw at Bloomberg Screentime 2026 in Los Angeles. (Source: Bloomberg)

中文摘要 RedBird Capital创始人Gerry Cardinale在洛杉矶Bloomberg Screentime 2026活动上,与彭博Lucas Shaw讨论体育、媒体和娱乐投资,以及Paramount Skydance与Warner Bros. Discovery合并前景。

Getting Oil and Gas Through Hormuz Strait Is Dangerous Job for Ship Captains

A tanker captain’s first-hand account shows why flows from the Middle East are rising, despite the dangers of Iranian drones and missiles.

中文摘要 彭博援引一名油轮船长第一手经历报道,尽管面临伊朗无人机和导弹威胁,中东经霍尔木兹海峡的油气运输量仍在上升。

Suno's Mikey Shulman on the AI Music Landscape

Suno Co-Founder & CEO, Mikey Shulman discusses AI tools for music production, copyright disputes and emerging licensing models with Bloomberg’s Ashley Carman and Lucas Shaw at Bloomberg Screentime 2026 in Los Angeles. (Source: Bloomberg)

中文摘要 Suno联合创始人兼CEO Mikey Shulman在洛杉矶Bloomberg Screentime 2026活动上,与彭博记者讨论AI音乐制作工具、版权争议及新兴授权模式。

EU countries in crisis talks over release of diesel stocks

Trump administration has pressured European countries to release strategic reserves, warning that it could ban US diesel exports if they do not do so

中文摘要 FT报道,欧盟国家就释放柴油库存举行危机谈判。特朗普政府施压欧洲释放战略储备,并警告若不行动可能禁止美国柴油出口。

Firmus Eyes $30 Billion Valuation in Australian IPO, AFR Reports

Data center operator Firmus Grid Ltd. is seeking a A$43.7 billion ($30.3 billion) valuation in its Australian initial public offering this month, the Australian Financial Review reported on Thursday, citing pricing sent to potential investors.

中文摘要 彭博援引《澳大利亚金融评论》报道,数据中心运营商Firmus Grid Ltd.本月澳大利亚IPO寻求437亿澳元(303亿美元)估值,定价已发送潜在投资者。