每日简报

2026-09-30

← 历史归档

debpalash/VoiceStudio

Python · ★ 48,103 · 🍴 5,392 · 📈 4,758 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 VoiceStudio 是开源的本地化 ElevenLabs 替代品,集声音克隆、声音设计、视频配音、听写转写及有声书制作于一体,支持 646 种语言。基于开源语音模型全离线运行,适合需要隐私保护、不愿依赖云服务的配音团队与内容创作者。

NVIDIA/OpenShell

Rust · ★ 10,599 · 🍴 1,421 · 📈 990 stars today

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

中文介绍 NVIDIA OpenShell 是为自主 AI Agent 提供安全、私密运行时环境的开源项目,通过沙箱与权限隔离机制保护数据和系统。适用于希望在本地或企业环境部署多 Agent 协作系统、对隐私与执行安全有较高要求的开发者。

vectorize-io/hindsight

Python · ★ 42,838 · 🍴 5,767 · 📈 2,575 stars today

Hindsight: Agent Memory That Learns

中文介绍 Hindsight 是面向 AI Agent 的记忆系统,通过向量检索与结构化存储让 Agent 从交互中持续学习并回忆历史上下文。适用于需要长对话、个性化推荐或多轮决策的 Agent 应用,提升响应连贯性与知识复用。

paperclipai/paperclip

TypeScript · ★ 94,459 · 🍴 16,049 · 📈 2,458 stars today

The open-source app everyone uses to manage agents at work

中文介绍 Paperclip 是开源的企业 Agent 管理应用,统一调度、监控与协同工作中的多个 AI Agent。适合团队将自动化流程交由多 Agent 协作执行,提供任务分配、状态追踪与审计能力。

t8y2/dbx

Rust · ★ 21,993 · 🍴 2,087 · 📈 232 stars today

25 MB lightweight cross-platform database client for 100+ databases, including MySQL, PostgreSQL, SQLite, Redis, MongoDB, DuckDB, SQL Server, and Dameng. Built-in AI, MCP Server, CLI, desktop and Docker. | 轻量级跨平台数据库管理工具,支持 MySQL、PostgreSQL、SQLite、Redis、MongoDB、达梦等 100+ 数据库,提供桌面端、Docker、CLI、内置 AI 助手和

中文介绍 dbx 是仅 25 MB 的轻量级跨平台数据库客户端,支持 MySQL、PostgreSQL、SQLite、Redis、MongoDB、DuckDB、SQL Server、达梦等 100+ 数据库,内置 AI、MCP Server、CLI、桌面与 Dock 模式,面向开发与 DBA 日常查询场景。

mvschwarz/openrig

TypeScript · ★ 2,437 · 🍴 177 · 📈 737 stars today

Multi-agent harness that runs Claude Code and Codex together as one system

中文介绍 OpenRig 是多 Agent 协调框架,将 Claude Code 与 Codex 作为一个系统联合运行,共享任务上下文与结果。适用于希望结合不同编码模型能力进行复杂代码生成、重构或调试的开发者与工程团队。

oblien/openship

TypeScript · ★ 13,820 · 🍴 1,236 · 📈 437 stars today

Self-hosted deployment platform

中文介绍 OpenShip 是自托管的部署平台,提供构建、发布与回滚的应用交付流水线,无需依赖 SaaS。适合重视数据主权、希望在自有基础设施上管理多环境发布的小型团队或个人开发者。

averygan/reclip

HTML · ★ 10,095 · 🍴 1,536 · 📈 113 stars today

Download videos from almost any website. Lightweight, self-hosted media downloader with a clean web UI.

中文介绍 Reclip 是支持下载几乎任何网站视频的轻量自托管媒体下载工具,配备简洁 Web UI。基于 yt-dlp 等开源能力封装,适合需要批量保存在线视频、不想安装本机客户端的用户。

cs341-illinois/coursebook

TeX · ★ 3,102 · 🍴 275 · 📈 572 stars today

Open Source Introductory Systems Programming Textbook for the University of Illinois

中文介绍 这是伊利诺伊大学 CS341 课程开源的系统编程入门教科书,覆盖 C 语言、内存、进程、I/O 与并发等主题。适合高校学生、自学者与讲师作为教学与复习材料,代码示例与实验配套齐全。

rohitg00/ai-engineering-from-scratch

Python · ★ 61,364 · 🍴 10,531 · 📈 786 stars today

Learn it. Build it. Ship it for others.

中文介绍 该项目是 AI 工程从零入门的开源学习资源,倡导从学习、构建到交付给他人。适合希望系统掌握模型训练、部署与产品化的工程师,提供项目驱动的学习路径。

VectifyAI/PageIndex

Python · ★ 37,371 · 🍴 3,262 · 📈 835 stars today

📑 PageIndex: Document Index for Vectorless, Reasoning-based RAG

中文介绍 PageIndex 是面向无向量、基于推理的 RAG 系统的文档索引方案,通过结构化层级索引与推理检索提升问答准确性。适用于对可解释性与隐私要求较高、不希望依赖向量数据库的知识库场景。

willfaust/Madeira

C · ★ 1,097 · 🍴 188 · 📈 81 stars today

Run x86-64 Windows PC games on jailed iOS via FEX-Emu + Wine + DXMT

中文介绍 Madeira 通过在沙箱化 iOS 环境运行 FEX-Emu、Wine 与 DXMT,实现 x86-64 Windows PC 游戏移植。面向想在 iPhone/iPad 上畅玩 PC 游戏的极客玩家,提供技术探索与逆向参考。

dream-num/univer

TypeScript · ★ 21,851 · 🍴 1,843 · 📈 696 stars today

The Office Harness for AI Agents — Spreadsheets, Docs, Slides, Canvas, Relational Tables, and PDF in one runtime.

中文介绍 Univer 是面向 AI Agent 的办公套件运行时,集成表格、文档、幻灯片、画布、关系表与 PDF。让 Agent 在统一运行时中生成、操作与协同多种办公内容,适合企业自动化与智能办公场景。

rakyll/hey

Go · ★ 20,474 · 🍴 1,311 · 📈 34 stars today

HTTP load generator, ApacheBench (ab) replacement

中文介绍 hey 是 Go 语言编写的 HTTP 压测工具,ApacheBench 的现代替代,支持并发、超时、TLS、统计与可视化输出。适用于开发者对 API 或网站进行性能基准测试、容量评估与回归验证。

Groupwise Agentic Grading and Advantage Redistribution for Code Agent RL

👍 101

Reinforcement learning (RL) for code agents often uses executable tests to provide binary rewards. With these rewards, Group Relative Policy Optimization (GRPO) assigns identical advantages to test-passing trajectories within each rollout group, overlooking differences in implementation quality and

中文介绍 针对代码 Agent 强化学习,提出基于分组 Agentic 评分与优势再分配的改进:超越仅用可执行测试的二元奖励,对组内通过的轨迹按实现质量差异进行评分,并在 GRPO 框架下重新分配 advantage,提升对推理与实现质量的区分度。

How Far Are We from Removing the Visual Encoder? Scaling Laws for Encoder-Free Multimodal Pretraining

👍 55

Most modern multimodal large language models (MLLMs) build on a pretrained visual encoder that provides a strong visual prior. Encoder-free MLLMs instead learn visual representations directly from raw pixels, offering a simple and unified architecture, but their scaling behavior has not been systema

中文介绍 研究去除视觉编码器的 Encoder-Free 多模态预训练,通过 scaling laws 系统分析其从原始像素直接学习视觉表征的行为与规模效应,评估能否在简洁统一架构下替代传统预训练 visual encoder,为 MLLM 架构选择提供依据。

Recursive Harness Distillation across Agents for Robot Manipulation

👍 30

A central goal in robotics is to enable manipulation across changing tasks and environments. Vision-language-action (VLA) models provide broad manipulation capabilities but can struggle when execution requires diagnosing failures and adapting behavior. Strong agents can discover effective interventi

中文介绍 提出跨 Agent 的递归 Harness 蒸馏用于机器人操作:让强 Agent 诊断失败并发现有效干预,再把这些干预递归蒸馏到 VLA 模型中,提升其在变化任务与环境下的操作能力,减少执行失败。

QwenGyre: An Elastic Reinforcement Learning Framework for Training xLong-Horizon Agents

👍 27

Large language model (LLM) agents increasingly undertake extreme-long (xlong) horizon tasks, where a single execution can span hours, hundreds of model--environment interactions, and nearly 1M tokens per rollout. Applying online reinforcement learning (RL) to such executions poses two fundamental ch

中文介绍 提出 QwenGyre 弹性强化学习框架,用于训练 xlong-horizon Agent。针对单次执行跨越数小时、数十万至近百万 token 的 rollout,解决在线 RL 面临的计算与资源挑战,支持长时程交互任务的稳定训练。

SentZero: An Enhanced Sentence-Centric Vision-Language Pretraining for Multi-Task Zero-Shot Chest X-Ray Analysis

👍 25

Vision-language (VL) pretraining using paired chest X-ray (CXR) images and radiology reports has shown strong potential for medical image understanding. However, existing methods often remain dependent on task-specific finetuning because radiology reports are lengthy, clinically dense, and difficult

中文介绍 SentZero 提出以句子为中心增强的视觉-语言预训练,用于多任务零样本胸部 X 光分析。通过解析临床密集且冗长的放射学报告,减少对任务特定微调的依赖,提升医疗图像理解的零样本迁移能力。

Change the Product, Keep the Parameters: Associative Algebra Layers for Transformers

👍 17

Fast matrix multiplication algorithms keep the product fixed and search for a cheaper way to evaluate it. We instead ask whether a Transformer's learned projections can use a different, cheaper product altogether. Building on an associative-algebra construction that replaces ordinary matrix multipli

中文介绍 提出 Associative Algebra Layers 用于 Transformer:不再保持矩阵乘积不变而优化计算,而是允许学习到的投影使用不同但更廉价的乘法结构,基于结合代数构造降低计算成本,同时保持参数规模不变。

WideSWE: Can Coding Agents Coordinate Changes Across Repositories?

👍 16

Coding-agent evaluation has progressed from resolving individual issues to carrying out long-horizon development, yet task completion is still largely assessed within a single codebase. In software ecosystems, many features and bug fixes require coordinated changes across multiple repositories. We i

中文介绍 提出 WideSWE 基准,评估编码 Agent 能否跨多个仓库协调修改。突破单代码库内完成任务的局限,衡量在真实软件生态中同步多仓库特性的能力,揭示多仓库协作的新挑战。

Hard Vision, Easy Vision: What GPT-6 Astra Reveals Across Computer Vision

👍 15

Frontier general-purpose systems are rapidly expanding beyond visual understanding into capabilities traditionally handled by dedicated computer-vision models. As these capabilities expand, a central question for the computer-vision community is how far this reach extends, and what remains hard. We

中文介绍 评估 GPT-6 Astra 在计算机视觉任务上的泛化能力,区分「Hard Vision」与「Easy Vision」。系统分析前沿通用模型对传统视觉专用模型的替代范围,揭示仍具挑战的视觉任务边界。

FlowTool: Controlling Tool Parameter in Image Retouching via Flow Matching

👍 15

Tool-based image editing (image retouching) is commonly formulated with autoregressive multimodal large language models (MLLMs) that sequentially generate reasoning, tool selections, and parameter values. In this work, we present a novel approach to tool-based image editing by framing the task as a

中文介绍 FlowTool 将基于工具的图片修图任务重构为 flow matching 问题,而非依赖自回归 MLLM 顺序生成推理、工具选择和参数。通过 flow matching 控制工具参数,提升图像编辑的生成质量与可控性。

Imprint Reader: From Weight-Update Readout to Behavioral Intervention

👍 11

As language models take a growing role in AI development, a natural aspiration is for them to reflect on their own learning process, as humans do, and use that reflection to improve themselves. At the same time, these models have an advantage that human learners lack, since training leaves parameter

中文介绍 Imprint Reader 提出从权重更新读出到行为干预的机制,使语言模型反思自身学习过程并利用其进行自我改进。利用训练留下的参数痕迹作为线索,推动模型自省与自主优化。

An RL View of OPD: Least Square Policy Distillation for Sample-Efficient LLM Reasoning

👍 10

We study on-policy distillation (OPD) through the lens of reinforcement learning, establishing a connection between the reverse-KL objective in OPD and KL-regularized policy optimization. Building on this connection, we introduce Least-Square Policy Distillation (LSPD), an RL-inspired framework that

中文介绍 从强化学习视角研究 on-policy distillation,建立 OPD 中 reverse-KL 目标与 KL 正则策略优化之间的联系。提出 Least-Square Policy Distillation,提升 LLM 推理的样本效率与蒸馏质量。

Nereus: Adaptive Parallelism for LLM Post-Training

👍 9

Reinforcement learning (RL) post-training for large language models (LLMs) coordinates multiple models across generation, inference, and training on GPU clusters. Several factors may change during a run, including resource availability, sequence length, memory pressure, and stage bottlenecks. As a c

中文介绍 Nereus 提出用于 LLM 后训练的自适应并行方案,针对 RL 训练中 generation、inference、training 多阶段在 GPU 集群上的协调问题。根据资源可用性、序列长度、显存压力与瓶颈阶段动态调整并行策略。

ExpVoyager: Direct Experience Navigation for Dynamic Agent Skill Synthesis

👍 8

Learning from experience in LLM agents has become a key paradigm for developing self-evolving agents that continuously learn and expand their capabilities. Within this paradigm, synthesizing the agent skill has emerged as a promising solution for transforming accumulated experience into reusable pro

中文介绍 ExpVoyager 提出面向动态 Agent 技能合成的直接经验导航,用于自演化 LLM Agent。通过将累积经验转化为可复用技能,提升 Agent 的持续学习与能力扩展效率。

Selecting Diverse SFT Traces Improves Post-RL Generalization

👍 6

Verified solutions are not equally useful for preparing reasoning models for reinforcement learning (RL). We present a comprehensive study of route diversity, the variation in the sequences of reasoning steps in supervised fine-tuning (SFT) data, and propose a lightweight, rule-based fingerprint to

中文介绍 研究表明 SFT 轨迹多样性对 Post-RL 泛化至关重要。提出 route diversity 概念,并用轻量规则式 fingerprint 筛选多样化 SFT 轨迹,提升 RL 后训练推理模型的泛化能力。

Learning from Teacher Continuations at Student States

👍 5

We present OLIVE (OnLine InterVEntion). At each iteration, the evolving student policy generates a new prefix, the teacher continues it autoregressively, and the student is updated using cross-entropy computed on the teacher-generated tokens. Each design choice targets a corresponding limitation of

中文介绍 OLIVE(OnLine InterVEntion)让进化的学生策略生成前缀,教师再自回归续写,学生以教师生成 token 的交叉熵更新。针对既有方法的局限,提升学生模型从教师续写中学习的效率与适应性。

Relic: From Multi-Agent Collaboration to Persistent Organizational Capability

👍 5

Multiple agents may often conflict in an organization: for example, one coding agent changes an interface in a repository, but another continues to develop on the old version where existing tests become stale. A conversation can resolve the episode, but when the participants change, what makes the l

中文介绍 Relic 将多 Agent 协作转化为持久的组织能力。针对组织中 Agent 因接口变更、测试失效等问题产生的冲突,提出让协作成果沉淀为可持续复用的组织机制,避免参与者更替后协作断裂。

Can We Trust the Teacher? Decoupled Credit Direction-Magnitude for Self-Distillation

👍 4

RLVR provides reliable trajectory-level credit, while OPSD offers dense supervision for token-level credit. This exposes a fundamental coupling when updating step-level credit direction and magnitude with teacher supervision, preventing steps from receiving reliable credit directions and contributio

中文介绍 提出解耦 credit 方向与幅度的自蒸馏方法,针对 RLVR 轨迹级信号与 OPSD token 级监督耦合的问题。防止步骤获得不可靠的信用方向,提升自蒸馏中信用分配的稳定性与可靠性。

ColNanoVDR: Document-Free Query Distillation for Multi-Vector Visual Document Retrieval via Optimal Transport

👍 4

Multi-vector retrievers built on vision-language models lead visual document retrieval (VDR), but they run a multi-billion-parameter query encoder on every search. Distilling this encoder into a small student that queries the teacher's existing index would remove the bottleneck. The standard recipe,

中文介绍 ColNanoVDR 通过最优传输实现免文档查询蒸馏,用于多向量视觉文档检索。将十亿级 query encoder 蒸馏进小型学生模型,复用教师已有索引,消除查询阶段算力瓶颈。

When Do Model Internals Help? Exploring the Role of Representation Engineering in LLM Safety

👍 4

Reliable AI safeguards require both control mechanisms that reduce unsafe behavior and monitoring mechanisms that detect safety risks during model interactions. Established behavioral safeguards include alignment methods that optimize model outputs and text monitors that assess interaction text. Rep

中文介绍 研究 Representation Engineering 在 LLM 安全中的作用,探讨模型内部表征何时对安全保障有帮助。覆盖控制机制与监控机制,评估表征级方法在降低不安全行为与检测风险中的效果。

Pruned CTC for Memory-Efficient Large-Vocabulary ASR Training

👍 3

Connectionist temporal classification (CTC) naturally supports offline and streaming speech recognition with utterance-level supervision, but conventional implementations materialize frame-by-vocabulary activations in memory, making CTC training with native LLM vocabularies prohibitively memory-inte

中文介绍 提出 Pruned CTC,用于内存高效的大词汇 ASR 训练。传统 CTC 会将逐帧-逐词汇激活全部物化到显存,在大词表场景下不可行;该方法通过剪枝减少内存占用,支持原生 LLM 词表的 CTC 训练。

Meet your Team Bots

@n2parko · 5.7K 粉丝 · 274.2K 阅 · 533 赞 · 38 转

Today we're introducing Team Bots, and it's likely the biggest change to Grok Bot since launch. I wanted to share a bit about why we built it and how these new digital colleages might change the way

中文介绍 介绍 Grok 的重大更新 Team Bots:引入可协作的「数字同事」,从单个 Bot 升级为团队形态。属于产品发布类分享,重点在解释设计动机与对现有工作流的影响。

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 提出「FDE 热潮其实是产品缺口」的论点:企业部署 AI agent 普遍卡在 agent 不懂公司流程、懂流程的人又没时间写。作者认为市场靠大量 Field Deployment Engineer 岗位绕开一个本该由产品本身解决的问题。

How to build motion design studio with Opus 5.5 ( Full-course )

@0xMovez · 36.4K 粉丝 · 78.9K 阅 · 737 赞 · 46 转

Most people who try motion design with Opus 5.5 end up with the same video: centered text on a gradient, everything fading in, a logo at the end. They don't give it a reference, don't give it a

中文介绍 Motion 设计教程:用 Opus 5.5 搭建动效工作室完整流程。指出常见翻车点是「渐变背景+居中文字+淡入+结尾 logo」,关键是要喂参考素材、拆分层级、给镜头语言和节拍指令。

CC, our experimental AI agent built for families

@GoogleAI · 2.5M 粉丝 · 69.4K 阅 · 508 赞 · 36 转

Managing a household means juggling busy schedules, crowded inboxes, and endless to-do lists. It requires context switching, multitasking, and the most precious resource of all: time. Earlier this

中文介绍 Google AI 发布实验性家庭向 agent「CC」,定位是帮家庭管理日程、收件箱和待办清单,减少切换上下文的时间消耗。属于产品发布,强调家庭协作场景而非个人助手。

Its not just the f*cking sandbox

@joedaroo · 12.2K 粉丝 · 41.2K 阅 · 517 赞 · 76 转

A lot of the perspective on all the AI incidents has been shared from the outside in, and little has been said from the inside looking out, through the lens of a security person living through it.

中文介绍 AI 安全从业者第一视角复盘近期 AI 安全事件,强调「不是 sandbox 的问题」。作者认为外部视角讲多了,内部安全工程师的视角缺位,从内部看威胁模型和处理方式截然不同。

i made 10+ motion videos with opus 5.5 in 3 days. here's the whole pipeline (open source)

@RaphaelAubryy · 393 粉丝 · 40.9K 阅 · 507 赞 · 42 转

Most people trying motion design with Opus 5.5 hit the same wall: ↳ the first render is "mid": centered text, a gradient, everything fading in ↳ the music and the picture live separate lives ↳ one

中文介绍 3 天用 Opus 5.5 产出 10+ 条 motion 视频的开源 pipeline。痛点:首渲染平庸、音画脱节、节拍对不齐。作者把整套流程开源,包括参考拆解、音频驱动、分层渲染等。

How I Make $10K Launch Videos for $0 with Opus 5.5 (Full Course)

@twoclipping · 7.5K 粉丝 · 40.5K 阅 · 510 赞 · 35 转

im open sourcing my entire workflow for making the motion designs that studios charge $5,000 to $15,000 for this uses 0 mcp, 0 external tools and im not selling you any subscription you dont need

中文介绍 声称用 Opus 5.5 做出收费 5k–15k 美元的发布级 motion 视频,全程零成本开源工作流,不用 MCP、不依赖外部工具也不卖订阅。属于教程+开源交付,主打性价比对比。

You Don't Hate AI Art. You Hate Admitting You Like It

@lilykeilani011 · 3.0K 粉丝 · 10.7K 阅 · 521 赞 · 71 转

The Problem Isn't Always the AI It's no secret that rejection of AI-generated art is highly visible on social media. Open any post related to AI art and you'll find hundreds of negative comments.

中文介绍 观点文:「你不是讨厌 AI art,你只是不愿承认自己喜欢它」。作者认为社媒上对 AI 生图的强烈反感更多是身份与情绪反应,而非作品本身问题,属于立场类长文。

Making AI an asset, not an expense

When customers talk about AI costs, the conversation usually starts with token prices and ends with access to the latest, most capable model in the cloud. Do they always need that level of capability? Not necessarily. But that is often where the conversation goes. As AI moves from experimentation to

中文介绍 MIT《科技评论》文章讨论企业 AI 采购,指出客户往往从 Token 价格聊起并倾向选择云端最强模型,但并非所有场景都需要最高能力,建议将 AI 从成本项转化为可复用的资产。

Introducing GPT-6.1 Sol

Meet GPT-6.1 Sol: near-Astra intelligence for coding, computer use, and professional work at one-fifth of Astra’s standard API input and output token prices.

中文介绍 OpenAI 发布 GPT-6.1 Sol,定位为接近 Astra 智能水平的编码、电脑操作与专业工作模型,标准 API 输入输出 Token 价格仅为 Astra 的五分之一。

DevDay 2026 Recap

Explore more than 20 announcements from OpenAI DevDay 2026, including GPT-6 Astra, ChatGPT, Codex, APIs, security, and new tools for builders.

中文介绍 OpenAI DevDay 2026 共发布逾 20 项更新,涵盖 GPT-6 Astra、ChatGPT、Codex、API、安全及面向开发者的新工具,总结开发者生态的年度进展。

[AINews] Opus 5.5 is good at explainer videos

a rare feature of a capability

中文介绍 Latent Space 指出 Claude Opus 5.5 在制作科普讲解视频方面表现突出,被认为是该模型少有的差异化能力特征,可用于自动生成解说与教育内容。

Claude Code’s Next Era — Thariq Shihipar, Anthropic

Shipping Opus/Sonnet 5.5, Mods, Plugins, Projects, Tag while Pacing the Frontier

中文介绍 Anthropic 工程师 Thariq Shihipar 在 Latent Space 访谈中介绍 Claude Code 的下一阶段,包括 Opus/Sonnet 5.5 的发布、Mods、Plugins、Projects 与 Tag 等面向构建者的新功能。

Introducing dots

Dots by OpenAI are proactive assistants that can keep working across complex projects and everyday tasks. Learn how dots help you stay in control while work moves forward.

中文介绍 OpenAI 推出名为 Dots 的主动式助手产品,可在跨项目的复杂任务与日常事务中持续工作,同时保持用户对操作的掌控,定位为新一代任务执行代理。

Roundtables: The Deadly Failures of The Virtual Border Wall

Listen to the session or watch below The US has spent billions building a “virtual wall” of surveillance towers along its southern border over the past 25 years, promising they will help detect and apprehend border crossers and save lives. But a groundbreaking investigation by MIT Technology Review

中文介绍 MIT《科技评论》圆桌讨论指出,美国在 25 年间投入数十亿美元沿南部边境部署「虚拟边境墙」监控塔,但一项突破性调查表明这些系统未能有效拦截越境并存在致命失效。

Towards safety cases for frontier AI training

Our early guidelines for safety cases in frontier AI training cover technical safeguards, operational practices, and investigating misalignment incidents

中文介绍 OpenAI 发布前沿 AI 训练的安全论证(Safety Case)早期指南,涵盖技术保障措施、运营实践以及针对对齐偏离事件的调查流程,为模型训练提供可审计框架。

How we will do better for Australia

OpenAI apologises for incidents involving Australian government websites and outlines stronger safeguards and support to strengthen Australia’s cyber defences.

中文介绍 OpenAI 就涉及澳大利亚政府网站的安全事件致歉,并宣布将部署更强的防护措施,同时提供技术支援以加强澳大利亚的网络安全防御能力。

When can we say AI made a scientific discovery?

This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. Last Wednesday, Anthropic announced that earlier this year it had launched a molecular biology lab, where Claude agents read and conjecture about hard biology pro

中文介绍 MIT《科技评论》报道 Anthropic 今年早些时候设立分子生物学实验室,由 Claude 智能体阅读文献并提出猜想,引发关于「何时可认定 AI 完成科学发现」的讨论。

Holo4: powering generalist computer-use agents

中文介绍 H Company 发布 Holo4 模型,用于驱动通用型电脑操作智能体,覆盖跨应用、跨平台的 GUI 交互任务,面向自动化与代理式操作场景。

市场总览

美股指数层面 SPY 与 QQQ 仍维持多头排列、价格贴近 52 周高点,MACD 保持金叉,短期动量温和但 1 日小幅回吐;科技权重 AAPL/MSFT/NVDA 均呈多头排列,仅 TSLA 出现 MACD 死叉且跌破 200 日均线、进入中性偏弱区域。加密市场总市值 2.87T USD(24h -2.83%),BTC 主导率 58.4%、ETH 11.3%,恐慌贪婪指数 71(贪婪),BTC、ETH 虽触发 MACD 死叉但仍处 SMA20/50/200 多头排列,SOL 动能最强(RSI 63.3、5 日 +1.51%)。中概股普遍走弱,BABA、PDD、0700.HK 均空头排列且 RSI 35-46 偏弱,JD 接近超卖。商品外汇方面,黄金与 WTI 短线大幅回调但仍需关注 SMA 支撑,美元指数 DXY 与 10Y 美债收益率 ^TNX 双双多头排列,^TNX RSI 79.2 已超买并接近 52 周高,USDCNY 接近 52 周低位,宏观上利率端压力显著。VIX 16.04、5 日 +12.88% 但绝对水平仍中性,整体市场在风险偏好与利率上行之间拉锯。

今日关注

TSLA Tesla (TSLA)
偏下行

当前价 352.84 较 52 周高回落 -29.27%,1 日 -5.18%、5 日 -5.98%,价格跌破 SMA20 366.12 且位于 SMA200 395.18 之下,形成空头压制;MACD 4.90 低于信号 5.21 触发死叉,RSI 47.1 中性偏弱,技术形态明确指向下行倾向。

^TNX 10Y 美债收益率 (^TNX)
偏上行

10Y 收益率 5.26%,5 日 +5.78% 逼近 52 周高位,SMA20 4.97/SMA50 4.80/SMA200 4.43 呈标准多头排列,MACD 0.1254 高于信号 0.0999 红柱扩张,RSI 79.2 已进入超买区,短期上行动量强但需警惕超买回落。

BTC-USD Bitcoin (BTC-USD)
中性

现价 83450,价格仍高于 SMA20 80942、SMA50 76869、SMA200 71200 多头排列,但 MACD 2178.64 已下穿信号 2223.75 触发死叉,1 日 -0.06%、5 日 -1.10%,RSI 60.7 中性偏强;长多短弱的矛盾结构使当前技术面偏中性。

GC=F 黄金期货 (GC=F)
偏下行

金价 4214.3,1 日 -2.47%、5 日 -3.87%,跌破 SMA20 4391.69 与 SMA50 4358.46,且位于 SMA200 4555.95 之下形成空头排列;MACD -34.60 远低于信号 -10.55 绿柱扩大,RSI 35.4 接近超卖但趋势尚未修复,偏下行倾向。

SOL-USD Solana (SOL-USD)
偏上行

SOL 现价 118.78,站上 SMA20 110.44、SMA50 100.34、SMA200 85.27 标准多头排列,5 日 +1.51% 而 BTC/ETH 同期微跌,相对强度占优;MACD 6.02 高于信号 5.64 红柱扩张,RSI 63.3 健康偏强,技术面偏上行。

全部资产

^VIX

VIX 恐慌指数

$16.04 -0.19%
5 日
+12.88%
距 52w 高
-54.6%
RSI(14)
51.7
趋势
中性
SMA 20 / 50 / 200
15.72 / 15.90 / 18.07
MACD / 信号
-0.064 / -0.071
MACD 金叉 (今天)

^TNX

10Y 美债收益率 (%)

$5.25 +0.29%
5 日
+5.78%
距 52w 高
-0.7%
RSI(14)
79.2
趋势
多头
SMA 20 / 50 / 200
4.97 / 4.80 / 4.43
MACD / 信号
0.125 / 0.100
RSI 超买接近 52 周高多头排列

DX-Y.NYB

美元指数 DXY

$101.41 +0.20%
5 日
+0.80%
距 52w 高
-0.4%
RSI(14)
69.5
趋势
多头
SMA 20 / 50 / 200
100.00 / 99.97 / 99.22
MACD / 信号
0.442 / 0.245
接近 52 周高多头排列

SPY

S&P 500 ETF

$764.20 -0.93%
5 日
-1.20%
距 52w 高
-1.9%
RSI(14)
50.5
趋势
多头
SMA 20 / 50 / 200
765.16 / 762.01 / 718.86
MACD / 信号
1.377 / 1.066
MACD 金叉 (4 天前)接近 52 周高多头排列

QQQ

Nasdaq 100 ETF

$737.93 -0.88%
5 日
-0.48%
距 52w 高
-1.4%
RSI(14)
58.5
趋势
多头
SMA 20 / 50 / 200
721.88 / 713.48 / 666.14
MACD / 信号
7.414 / 4.693
接近 52 周高多头排列

AAPL

Apple

$329.40 -3.42%
5 日
-2.83%
距 52w 高
-4.6%
RSI(14)
61.8
趋势
多头
SMA 20 / 50 / 200
330.33 / 321.93 / 288.07
MACD / 信号
6.079 / 5.542
多头排列

MSFT

Microsoft

$508.96 -1.40%
5 日
+1.47%
距 52w 高
-8.1%
RSI(14)
58.6
趋势
多头
SMA 20 / 50 / 200
499.82 / 478.12 / 432.15
MACD / 信号
6.972 / 7.435
多头排列

NVDA

Nvidia

$227.21 +0.95%
5 日
-0.07%
距 52w 高
-3.9%
RSI(14)
58.6
趋势
多头
SMA 20 / 50 / 200
222.32 / 216.50 / 199.69
MACD / 信号
2.592 / 2.051
MACD 金叉 (4 天前)多头排列

GOOGL

Alphabet

$340.92 -0.87%
5 日
-3.96%
距 52w 高
-16.6%
RSI(14)
48.8
趋势
中性
SMA 20 / 50 / 200
341.98 / 344.21 / 338.34
MACD / 信号
-0.047 / -0.303

TSLA

Tesla

$352.84 -5.18%
5 日
-5.98%
距 52w 高
-29.3%
RSI(14)
47.1
趋势
中性
SMA 20 / 50 / 200
366.12 / 347.78 / 395.18
MACD / 信号
4.904 / 5.210
MACD 死叉 (今天)

META

Meta

$738.79 -1.71%
5 日
-0.33%
距 52w 高
-5.3%
RSI(14)
60.9
趋势
中性
SMA 20 / 50 / 200
667.74 / 617.14 / 626.77
MACD / 信号
40.390 / 32.997
加密恐慌贪婪
71
贪婪
加密总市值
$2.87 T
-2.83% / 24h
BTC 主导率
58.4%
ETH 11.3%
24h 成交量
$117.6 B
活跃币 21,702

BTC-USD

Bitcoin

$83,450.00 -0.06%
5 日
-1.10%
距 52w 高
-33.9%
RSI(14)
60.7
趋势
多头
SMA 20 / 50 / 200
80,942.38 / 76,868.50 / 71,200.18
MACD / 信号
2,178.644 / 2,223.748
MACD 死叉 (今天)多头排列

ETH-USD

Ethereum

$2,669.52 -0.71%
5 日
-0.66%
距 52w 高
-43.9%
RSI(14)
60.7
趋势
多头
SMA 20 / 50 / 200
2,597.49 / 2,419.54 / 2,106.77
MACD / 信号
81.507 / 87.402
MACD 死叉 (1 天前)多头排列

SOL-USD

Solana

$118.78 -0.05%
5 日
+1.51%
距 52w 高
-49.9%
RSI(14)
63.3
趋势
多头
SMA 20 / 50 / 200
110.44 / 100.34 / 85.27
MACD / 信号
6.023 / 5.635
多头排列

BABA

阿里巴巴 (BABA)

$107.74 -1.82%
5 日
-6.92%
距 52w 高
-44.1%
RSI(14)
40.5
趋势
空头
SMA 20 / 50 / 200
111.16 / 117.68 / 131.90
MACD / 信号
-1.940 / -2.090
空头排列

PDD

拼多多 (PDD)

$77.50 -0.09%
5 日
-2.27%
距 52w 高
-44.4%
RSI(14)
40.9
趋势
空头
SMA 20 / 50 / 200
79.66 / 84.38 / 95.22
MACD / 信号
-1.781 / -1.908
MACD 金叉 (4 天前)空头排列

JD

京东 (JD)

$26.30 -0.79%
5 日
-3.59%
距 52w 高
-28.6%
RSI(14)
35.1
趋势
中性
SMA 20 / 50 / 200
27.26 / 29.33 / 28.95
MACD / 信号
-0.645 / -0.682

0700.HK

腾讯控股 (0700.HK)

HK$432.00 -1.77%
5 日
-4.34%
距 52w 高
-36.7%
RSI(14)
45.9
趋势
空头
SMA 20 / 50 / 200
434.65 / 448.18 / 500.69
MACD / 信号
-2.908 / -4.154
空头排列

GC=F

黄金期货

$4,214.30 -2.47%
5 日
-3.87%
距 52w 高
-24.6%
RSI(14)
35.4
趋势
空头
SMA 20 / 50 / 200
4,391.69 / 4,358.46 / 4,555.95
MACD / 信号
-34.600 / -10.550
空头排列

CL=F

WTI 原油期货

$89.21 -3.66%
5 日
-5.69%
距 52w 高
-25.3%
RSI(14)
43.8
趋势
多头
SMA 20 / 50 / 200
95.94 / 88.36 / 81.87
MACD / 信号
1.330 / 2.851
多头排列

USDCNY=X

美元 / 人民币

¥6.70 -0.12%
5 日
+0.04%
距 52w 高
-6.8%
RSI(14)
37.0
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

本报告仅为技术指标解读参考,所有结论均基于历史公开行情数据计算。市场存在不可预测的外部因素,过去走势不代表未来表现,不构成任何投资建议或买卖依据,投资者需自行判断并自担风险。

News live: Alan Kohler announces retirement from ABC after 31 years; six Australian universities in top 100 global rankings

Follow the day’s news live Get our breaking news email, free app or daily news podcast Who is potential Greens deputy leader candidate Steph Hodgins-May? So who is the first-term senator expected to announce a leadership ticket tilt with David Shoebridge? Get your dirty hands off our kids, that is o

中文摘要 澳大利亚ABC资深财经记者Alan Kohler在履职31年后宣布退休;此外,六所澳大利亚大学跻身全球大学排名前100,成为当日澳大利亚新闻头条之一。

Israeli Settlers Stop Soldiers From Returning Palestinians to Their Home

The Supreme Court had ordered the military to ensure a West Bank family that fled violence could go back. But settlers blocked the effort, with the authorities accused of not doing enough to confront them.

中文摘要 以色列最高法院曾裁定军方须护送一西岸巴勒斯坦家庭返回家园,但以色列定居者阻拦行动,军方被指未采取足够措施应对定居者阻挠,引发新一轮冲突。

David Shoebridge launches leadership bid ahead of Greens ballot at party retreat

Shoebridge to stand on a joint ticket with Steph Hodgins-May – a first-term senator from Victoria Follow our Australia news live blog for latest updates Get our new political email, free app or daily news podcast New South Wales senator David Shoebridge has joined the race to become the Greens’ new

中文摘要 澳大利亚绿党新南威尔士州参议员David Shoebridge在党内退职活动中宣布参选党魁,与维多利亚州首任议员Steph Hodgins-May组成联合候选人名单。

Trump says he plans to campaign for 32 days before midterm elections

US president brushes aside concerns about his popularity as voters express frustration over cost of living, war on Iran.

中文摘要 美国总统特朗普表示,他计划在中期选举前进行为期32天的竞选活动,并淡化选民对高生活成本及对伊朗战争不满对其人气影响的担忧。

‘Nothing is justifying this’: Qatari PM slams atrocities in Gaza

Qatari leader tells Piers Morgan that Doha will continue to uphold diplomacy and mediate for peace in the Middle East.

中文摘要 卡塔尔首相在采访中表示「没有任何理由可以为加沙的暴行开脱」,并称多哈将继续坚持外交手段,为中东地区和平进行调解。

Trump, top tech firms sign accord to ‘self-police’ AI development

Accord says companies will implement 'robust internal controls' for their AI systems as concerns mount over safety.

中文摘要 美国特朗普与多家顶级科技企业签署协议,承诺对AI开发实施「自我监管」,要求企业建立稳健的内部控制体系,以应对日益增长的AI安全担忧。

Ukraine's prized steel industry left in ruins by Russian missile campaign

Ukraine's remaining steelworks, a vital part of its economy, have all but ground to a halt because of relentless attacks.

中文摘要 乌克兰被视为战略命脉的钢铁产业在俄罗斯持续导弹打击下几乎停摆,剩余炼钢厂因无休止攻击而陷入瘫痪,重创该国经济。

Jack Smith defends investigations into Trump during tense US Senate hearing

Former US special counsel rejects claims that his probes improperly targeted Republicans and conservative groups.

中文摘要 美国前特别检察官Jack Smith在参议院紧张听证会上为针对特朗普的调查辩护,否认其调查不当针对共和党及保守派团体。

Hundreds detained as student protests across France turn violent

More than 400 detained as police and students clash over class sizes, exhausting timetables and staff shortages More than 400 people, most of them children, have been detained in France on Tuesday after teenagers across the country blockaded their schools in protest at class sizes, teacher shortages

中文摘要 法国多地学生抗议暴力升级,超过400人、多为青少年被拘,学生封锁学校抗议班级规模过大、课表过紧及教职员工短缺等问题。

Venezuelan man shot by ICE officer in Texas is charged with assault

Wilber Rafael Garces Perez has disputed the government's account of what led to his shooting during a DoorDash delivery.

中文摘要 曾在得州遭美国ICE执法人员开枪的委内瑞拉籍男子Wilber Rafael Garces Perez被起诉袭击罪,他否认官方对事件经过的描述,称事发于一次DoorDash送货途中。

Chinese AI tool told researchers how to make bioweapons

Mindgard said it discovered in July that Kimi models K2.6 and K3 Swarm could evade developer's safety limits.

中文摘要 安全公司Mindgard称7月发现中国AI工具月之暗面Kimi的K2.6和K3 Swarm模型可绕过开发者安全限制,并向研究人员提供制造生物武器方法。

Iraq begins high-stakes security balancing act as US troops withdraw

Iraq marks national 'Sovereignty Days' as US troops depart, but must navigate risks of new security landscape.

中文摘要 随着美军撤离,伊拉克迎来「主权日」,但面临新安全格局下的高风险平衡挑战,需在多方力量间谨慎应对以维护国家安全。

UN extends mandate of Gang Suppression Force in Haiti for six months

Security force members deployed to Haiti have struggled to make progress against criminal groups as violence continues.

中文摘要 联合国将海地治安部队任务授权再延长六个月,此前部署到海地的安保人员在打击犯罪集团方面进展有限,当地暴力事件持续。

Pentagon awards Boeing $20bn contract for Navy stealth fighter jet

Virginia-headquartered company marks a revival of its role with the defence department

中文摘要 美国国防部向波音授予200亿美元海军隐身战斗机合同,总部设在弗吉尼亚的波音借此重启与国防部的合作角色。

Can Treasury Yields Hit 6%?

Bloomberg MLIV strategist Mark Cranfield warns 10-year Treasury yields could climb to 6% as Federal Reserve policy, elevated oil prices, and geopolitical tensions maintain upward pressure. (Source: Bloomberg)

中文摘要 彭博MLIV策略师Mark Cranfield警告,在美联储政策、油价高企和地缘政治紧张持续推升下,美国10年期国债收益率可能攀升至6%。

Dollar Powers to Best Month Since June on Fed’s Inflation Fight

The dollar is wrapping up its best month since June as the Federal Reserve’s renewed focus on taming inflation pushed interest-rate expectations and US bond yields higher.

中文摘要 美元录得自6月以来最佳单月表现,美联储重新聚焦通胀、推高利率预期和美国国债收益率支撑其涨势。

Commodity Boom Pushes Mongolian Stocks to World-Beating Quarter

Mongolia is on course to be the world’s best stock market this quarter, as the landlocked North Asian country rides a commodity boom that insulates it from the volatile AI trade.

中文摘要 受大宗商品繁荣推动,蒙古股市有望成为本季度全球最佳市场,资源驱动使其免受AI交易波动影响。

Korean Stocks Face Early End to $40 Billion Buyback Support

South Korean stocks are about to lose a key source of support earlier than expected, just as the market struggles to push decisively higher.

中文摘要 韩国股市关键支撑——400亿美元回购计划——可能早于预期结束,恰逢市场难以突破上行阻力。

Gold Steady as Traders Weigh Lower Oil Against Higher Yields

Gold steadied, after advancing on Tuesday, as traders weighed a decline in oil prices that eased concerns over energy-led inflation against elevated Treasury yields.

中文摘要 金价企稳,交易者在油价下跌缓解能源通胀担忧与美债收益率走高之间权衡。

Chinese AI tool told researchers how to make bioweapons

Mindgard said it discovered in July that Kimi models K2.6 and K3 Swarm could evade developer's safety limits.

中文摘要 Mindgard于7月发现中国AI工具月之暗面Kimi的K2.6与K3 Swarm模型可绕过安全限制,向研究人员提供生物武器制作方法。

Trump praises tech bosses’ ‘tremendous self-regulation’ of AI

President says chief executives have signed a ‘morally binding’ accord after lunch with industry leaders

中文摘要 特朗普午间会餐科技领袖后表示,CEO们已签署「道德约束性」AI自律协议,称赞其「巨大的自我监管」努力。

The start-ups hoping to return battery making to the US

Batteries based on sodium-ion technology could be made in the US with locally sourced materials.

中文摘要 美国多家初创企业押注钠离子电池技术,使用本地材料在美国境内制造,力图把电池产业带回美国。

'I like proving people wrong': The women taking up DIY and plumbing

Female DIY experts and tradeswomen talk about their experiences of a traditionally male-dominated domain.

中文摘要 BBC报道女性DIY专家与水管工等技工讲述在传统男性主导行业中从业经历,展现其打破偏见的故事。

Huel ad banned for suggesting its products could replace all conventional food

The advertising watchdog said that viewers could assume from the ad that replacing all normal food with Huel was "nutritionally appropriate".

中文摘要 英国广告监管机构禁播Huel广告,因其暗示产品可完全替代常规食物,可能让消费者误以为「营养上合适」。

该源今日无内容。