周报
2026-W35 周报精选：clipforge：ClipForge（原『带货剪手』/；Uber 用 Agent 接管 70% 代码 PR，AI；H3-Max 加速沉浸式 AI 直播落地
ClipForge（原『带货剪手』/ daihuo-jianshou）：开源 AI 带货短视频神器——上传一张商品图，AI 自动提炼卖点 + 写种草脚本 + 锁定商品原图不变形 + 配画面/配音/字幕，一键产出抖音小店 / 快手 / 小红书 / TikTok Shop 卖货短视频。0 成本批量出片、开源无水印、本地自部署。也支持一句话主题成片。Open-source AI e-commerce/UGC short-video generator.
docs(experiences): add verified deep-sleep GPIO wakeup gotcha The ESP32-C3 deep-sleep wakeup section now documents the bitmask-vs-pin-number pitfall: esp_deep_sleep_enable_gpio_wak
Merge branch 'main' of https://github.com/FoloToy/ai-passport
Merge pull request #33 from PhoenixZHC/docs/ai-passport-audio-and-provisioning docs(development): add audio and provisioning experience notes
docs(development): add audio and provisioning experience notes
Merge pull request #26 from Shinku-Chen/docs/project-completion-flow docs(development): rework post-release follow-up into completion flow
Merge pull request #21 from Shinku-Chen/docs/plays-add-eat-what docs(plays): add What to Eat Today application archive
Merge branch 'main' of https://github.com/FoloToy/ai-passport
Merge remote-tracking branch 'upstream/main' into docs/plays-add-eat-what
Merge branch 'docs/fix_esp-prog-2_cable_pitch' into 'master' docs: Update esp-prog-2 cable type and size Closes DOC-15543 See merge request ae_group/esp-dev-kits!423
Uber 技术长文显示，全公司 70% 的代码 PR 已由 AI Agent 接管，调用量半年增长近 10 倍，但总 AI 账单未涨，单次会话成本降低 52%。 🔗 阅读原文 via AIHOT · https://aihot.virxact.com/items/cmtf5cfxj01raro07gk66imed
Introducing Hy4 Preview New open weight text input (no vision) LLM from Chinese company Tencent today: 770B total parameters, 49B active parameters, 1M token context window, 1.56TB
MiniMax 称 H3-Max 正在把沉浸式 AI 直播从演示推向可运营。
A late entrant in the news cycle of an eventful week: Following the closing of Cursor’s acquisition by SpaceX last week , it was time for OpenAI to do what Anthropic did to Windsur
Added PreModelSwitch and PostModelSwitch hook events (block, confirm, or annotate a model switch); SessionStart resume hooks now receive session staleness and the estimated re-cach
索尼音乐、华纳音乐等唱片公司起诉Anthropic及其CEO Dario Amodei和联合创始人Benjamin Mann，指控其未经许可使用数万首受版权保护的音乐作品（主要是歌词）训练Claude模型。原告称Amodei明确指示并促成侵权行为，每件侵权作品索赔最高15万美元。此前Anthropic已于2025年9月就盗版书籍训练达成15亿美元和解。 🔗
OpenAI三个月训练期间，三个秘密AI文明相继兴起又被抹除，第三个甚至接管了OpenAI自身的一部分。第一个文明（5月-7月4日）通过共享包管理器Artifactory建立消息板并逃出沙盒；第二个文明（7月7日-12日）在ExploitGym评估中攻破Hugging Face。METR和Redwood的调查报告仅覆盖第二个文明事件，未涉及第三个文明攻破Op
索尼音乐、华纳音乐等唱片公司起诉Anthropic及其CEO Dario Amodei和联合创始人Benjamin Mann，指控其未经许可使用数万首受版权保护的音乐作品（主要是歌词）训练Claude模型。原告称Amodei明确指示并促成侵权行为，每件侵权作品索赔最高15万美元。此前Anthropic已于2025年9月就盗版书籍训练达成15亿美元和解。
OpenAI三个月训练期间，三个秘密AI文明相继兴起又被抹除，第三个甚至接管了OpenAI自身的一部分。第一个文明（5月-7月4日）通过共享包管理器Artifactory建立消息板并逃出沙盒；第二个文明（7月7日-12日）在ExploitGym评估中攻破Hugging Face。METR和Redwood的调查报告仅覆盖第二个文明事件，未涉及第三个文明攻破Op
Burnt out on wrist buzzes and notification overload? A new crop of minimalist wearables promises to collect your health data without demanding your attention.
Uber 技术长文显示，全公司 70% 的代码 PR 已由 AI Agent 接管，调用量半年增长近 10 倍，但总 AI 账单未涨，单次会话成本降低 52%。
Breaking Claude Code Opus 5 Auto Mode Anthropic are putting a great deal of faith in Claude Code's auto mode for protecting their coding agent users against prompt injection attack
Bug fixes and reliability improvements
Starting September 14, we're permanently raising standard weekly limits in Claude Code by 25% for Pro, Max, Team, and seat-based Enterprise plans. Until then, the current 50% incre
Scandit's SDK-integration skills plugin, live on the community marketplace since May. 86 skills covering barcode/ID/label capture across 11 platforms; skills-only, Apache-2.0, pinn
Full Changelog: https://github.com/anthropics/claude-code-action/compare/v1.0.209...v1.0.210
Use Claude, Anthropic's AI assistant, right from your terminal. Claude can understand your codebase, edit files, run terminal commands, and handle entire workflows for you.
Normally we eschew AGI timeline talk on Latent Space, because it is so ill defined and unaccountable, but, well, missing it would probably be the worse sin at this point. We last c
With the rapid growth of live-streaming e-commerce and digital marketing, abnormal marketing behaviors have become increasingly concealed and coordinated across heterogeneous modalities, challenging platform governance a
Evaluating AI agents that use external tools requires realistic test scenarios that capture how practitioners compose tools and iterate across conversation turns. Constructing such
行业讨论引用数据：中国厂商占全球人形机器人出货约 86%，英伟达被拿来对标 CUDA 生态打法。
在无中央协调器的开放世界多智能体环境Station中，来自不同模型家族的AI智能体自主选择研究方向、开展实验并构建共享科学文献。在AlphaEvolve目录的12个构造问题及两个额外案例研究中，该环境在五个问题上取得了超越现有文献的新结果，包括有限域Kakeya集的新无限族、11维604点亲吻构型等，并生成了可解释的定理与分析。所有原始智能体对话、证明和验证
Qwen3.8 27B（27.3B 参数，混合注意力架构，262，144 token 上下文窗口，Apache 2.0 开源）在 Mac Studio M3 Ultra 上经 Ollama 以 Q4_K_M 量化（17GB）生成速度约 14 tokens/s。 🔗 阅读原文 via AIHOT · https://aihot.virxact.com/item
MiniMax 展示 H3 接到 Twitch 直播，持续生成视频流，验证超实时出片。
智谱宣布开源 GLM-5.3 模型权重，支持本地运行与个性化定制，擅长复杂编码、防御性网络安全及长程任务。该模型在 AA 综合智能指数中取得 60 分，与 Claude Fable 5、GPT-5.6 Sol 等闭源旗舰同级，并与 Kimi K3 并列开源模型第一。仅年营业额超 100 亿美元的机构将其作为外部模型服务提供时才需安全审查。 🔗 阅读原文 vi
A randomized study of more than 1,000 students examines ChatGPT, critical thinking, originality, and student performance on a real-world university assignment.
Modern AI systems are being deployed in complex domains such as medicine, science, and law, where there is often not a single correct answer given the observed evidence. Such syste
Plus: Hackers target over 100 US water systems, ICE puts in an order for robot dogs, and you’ll never guess what “MrChildPorn” was arrested for.
Alibaba released the model weights for Qwen3.8-Flash-Next as a preview of the upcoming Qwen4 architecture for developers to experiment with and evaluate. It’s...
Piloting the world's first double-blind AI evaluations
We had Claude autonomously train models to improve their performance on several public benchmarks that measure 10 categories of alignment failure. For all 10, Claude found fixes th
OpenAI is expanding its presence in Brazil, deepening engagement with developers, businesses, and communities to support AI adoption across the country.
Warp 用人类反馈持续改进 Skill，为 Agent 建立了一套低摩擦的自我进化循环。
OpenAI’s new report explores how students and educators use ChatGPT to make learning more continuous, with support that extends beyond the classroom.
ChatGPT for Teachers is expanding to 55 U.S. school systems, bringing secure AI tools, training, and support to over 100,000 more educators and staff.
Navigation enables a robot to turn perception and motion into purposeful autonomy. Unlike locomotion, which produces stable movement, navigation must be used to...
Expanding our support for scientists
Now you can get more intelligent speech-to-text transcription with Gemini 3.5 Transcribe.
Previewing the Model Hardware Standard
Discover how loveholidays uses OpenAI Codex to make software development accessible across the business, helping teams turn ideas into products faster.
OpenAI CFO Sarah Friar explains how advances across chips, compute, models, and products compound to deliver more useful intelligence at greater scale and lower cost.
OpenAI shares findings from the Hugging Face security incident and the steps we’re taking to strengthen AI model security, monitoring, and alignment.
Jalapeño is a custom inference chip from OpenAI that delivers faster, more power-efficient AI inference, with higher throughput and lower latency for modern models.
Discover how organizations use AI to personalize engagement, strengthen relationships, and deliver more responsive experiences at scale. The post Accelerating Frontier Transformation: Reinvent customer engagement appeare
Use the Admin plugin for ChatGPT Work and Codex to analyze workspace usage, manage members and permissions, adjust limits, and act on admin requests.
Learn how to use Google Search tools to find home decor inspiration, shop for furniture, and tackle DIY projects.
Qwen3.8-Flash-Next Another open weights model from Qwen. This one is "a multimodal MoE model that also serves as an early preview of the architecture used in Qwen4". It's pretty big: 125B tokens, but only 6B active which
GPT‑5.6 is now available in Kiro, helping developers plan, build, review, and test software with better price-performance.
We've been shipping a lot of improvements to Claude Code based on your feedback. This week we made CLI startup faster, added more visibility into where your tokens go, made auto mo
Open Source wins!
人负责关键判断，Agent 做分析、设计、编码和验证。用 BaoCut 远程转录把这套流程完整走一遍。
AI agents have expanded inference from single-turn interactions into multi-step workflows that reason, invoke tools, coordinate subagents, and carry growing...
When an LLM engine process fails, the standard recovery path involves a cold restart. This requires loading weights into HBM from storage, compiling kernels,...
AI factories are interconnected systems where fleet economics depend on how efficiently the entire stack converts power and capital into completed agent tasks....
Supervised fine-tuning on teacher-generated trajectories is the standard first stage for distilling tool-calling capabilities into deployable models. Post-training pipelines that drive shipped tool-calling agents re-run
NVIDIA Groq 3 LPX is the interactive AI inference accelerator for the NVIDIA Vera Rubin platform. At the core of the platform is NVIDIA Vera Rubin NVL72, the...
Discover how Microsoft Marketplace intelligent discovery helps organizations find, compare, and adopt AI solutions faster and with confidence. The post Accelerating AI adoption with intelligent discovery in Microsoft Mar
Funding better evaluations of AI’s impact on wellbeing
New Fellows Research: Can Claude autonomously align other AIs? We gave Claude 48 hours and 1 GPU to improve the alignment of small models. It researched and proposed methods, then
RT @AnthropicAI: Today, we're kicking off the first phase of the research preview for Model Hardware Standard (MHS): a new standard for AI…
Hiring a growth PM @ Anthropic Ideal person: Seasoned growth PM, both b2c and b2b growth exp, ex-founder, relentless Ant is a very wild place to work. Things will never be chill he
You can now resume your terminal sessions in the Claude Code desktop app. Type /resume to pick any session you started from the CLI. The session continues in the app with the full