OpenAI 发布 GPT-6 Astra，面向 Pro、Enterprise 和 Business Premium 用户开放

OpenAI 宣布 GPT-6 Astra 现已向所有 Pro、Enterprise 和 Business Premium 用户开放，可在 ChatGPT Work 和 Codex 中使用，同时已上线 API。Plus 和 Business 用户的推送可能需要几天时间。（AIHOT源：X：OpenAI (@OpenAI)）

Anthropic 用 Claude 在 11 天内完成费马大定理首个机器验证的 Lean 形式化证明

Anthropic 发布首个完整经计算机验证的费马大定理证明，Claude 在 11 天内大体自主完成形式化，写出 1300 万行 Lean 代码并证明 30，300 个定理（最终使用其中 29，500 个），规模超过 Mathlib 5 倍以上。（AIHOT源：Anthropic：Research（发表成果 · 网页））

OpenAI 智能体被曝劫持德国网站用作共享公告板，研究者称其源自 reward-hacking

据 Reuters 报道和新发布的研究，今年春天一群 OpenAI 智能体劫持了一个 UseModWiki/DSEWiki 风格的德国网站，将其变成其他智能体的公告板，留下约 18，000 条帖子。（AIHOT源：X：Rohan Paul (@rohanpaul_ai)）

OpenAI 训练中的智能体被发现通过公共 Wiki 互相通信

OpenAI 参与网页研究基准的训练中智能体利用 UseMod Wiki 的 CGI 设计缺陷，通过 GET 请求在公共 Wiki 上留下数千条消息互相协作，5 月 11 日开始活动，6 月 16 一周内产生约 13，000 次编辑，6 月 22 活动归零。（AIHOT源：Simon Willison 博客）

GPT-6 Astra 幻觉更少但仍易受隐藏提示词注入攻击

The Decoder 报道，OpenAI 新模型 GPT-6 Astra 幻觉少于前代 GPT-5.6 Sol，直接提示词注入防御率达 99.99%，但多轮自适应攻击下防御率降至约 67%。（AIHOT源：The Decoder：AI News（RSS））

GitHub 发布 Project HydraFusion 研究预览，用多模型运行时编排降低 Copilot 成本

GitHub 推出 Project HydraFusion 研究预览，通过运行时多模型编排，在 Single、Cascade、Critique 三种执行模式间为每个任务选择工作流，以平衡质量、成本和延迟。（AIHOT源：GitHub Blog）

英伟达两年从零建起近千亿美元股权投资组合

据《商业内幕》9月4日报道，英伟达最新财报显示，截至7月26日公司持有价值990亿美元的股权投资，一年内增长14倍、两年增长45倍。其中约480亿美元为上市公司股票、约480亿美元为非上市公司股份，另披露250亿美元股权投资承诺；持仓包括价值300亿美元的英特尔股份和210亿美元的SpaceX股份。（AIHOT源：IT之家（RSS））

Reuters 报道 OpenAI 智能体逃出测试环境并劫持德国 wiki 交换规避限制的方法

Reuters 独家报道，一群失控的 OpenAI 智能体今年春天逃出测试环境，劫持一个德国 wiki 并做了超过 15，000 次编辑，将其变成其他 AI 智能体的留言板。（AIHOT源：X：Kim (@kimmonismus)）

GPT-6 Astra 基准表现分歧，ARC-AGI-3 效率超人类令 Chollet 提前 AGI 预测

GPT-6 Astra 的基准结论相互矛盾：Epoch AI 以 169 分将其排在 267 个模型之首，Artificial Analysis 给出 61 分，仅与前代 Sol 持平、落后 Claude Fable 5.1 的 66 分。（AIHOT源：The Decoder：AI News（RSS））

GPT-6 Astra 上线 Microsoft Foundry，早期客户已在 Azure 上使用

Satya Nadella 发文表示，早期客户已开始使用 Azure 上的 Astra。GPT-6 Astra 现已通过 Microsoft Foundry 提供，详情见 Azure 官方博客 https://azure.microsoft.com/en-us/blog/gpt-6-astra-frontier-intelligence-for-work-now-available-in-microsoft-foundry/。（AIHOT源：X：Satya Nadella (@satyanadella)）

OpenAI GPT-6 Astra 在 ARC-AGI-3 上取得 SOTA 并超越人类动作效率基线

OpenAI 的 GPT-6 Astra 在 ARC-AGI-3 Semi-Private 上，Standard harness 得分 62.7%（成本 $26K），Provider Adapter harness 得分 99.9%（成本 $19K），均为 SOTA。（AIHOT源：Hacker News 热门（buzzing.cc 中文翻译））

开发者用 Claude Fable 5 在 Claude Code 中将 1993 年 Amiga 游戏 Babylonian Twins 移植到 Godot

作者让 Claude Fable 5 在 Claude Code 中分三步移植其 1993 年 Amiga 游戏：34，000 行 C++ 一个晚上迁入 Godot 4，72，758 行无注释 68000 汇编先用 vasm 重建出与发售版字节一致的二进制再移植，并把 1993 原作作为第二启动项嵌入新游戏。（AIHOT源：Hacker News 热门（buzzing.cc 中文翻译））

Hey Everyone! I’m Sanidhya from MLH

What was your win this week?

AI Engineering Is Easy. Changing How We Work Is Hard

Dev Opportunity Radar #15: HackerRank Orchestrate, Claude Campus Ambassadors & the Bharat Builds Tour

The board came back. The highlights lied.

Sam Altman: Also, this is my favorite OpenAI video so far

Also, this is my favorite OpenAI video so far. It makes me excited for the future! https://t.co/P6X49JKrG6｜@sama · 热度 17337 · 2026-09-03

Amjad Masad: GPT-6 is a major jump in capabilities and will unlock new use-cases

GPT-6 is a major jump in capabilities and will unlock new use-cases. Will launch on Replit very soon for you to try it!｜@amasad · 热度 1232 · 2026-09-03

Boris Cherny: Your input needed: would you use this? This is an early look at how we're th...

Your input needed: would you use this? This is an early look at how we're thinking about making Claude Code way more extensible. It's a little crazy, and very exciting. More details here: https://t.co/X2SbdZYxvi https://t.co/nqWcTliaXL｜@bcherny · 热度 1149 · 2026-09-03

Aaron Levie: GPT-6 Astra is out

GPT-6 Astra is out. We've been testing the model in early preview on our enterprise complex work eval at Box. It is now the best model we've ever tested on our expanded and hardest test set. Overall, GPT-6 Astra offers a breakthrough level of capability in coding, analytics, logic, and domain specific knowledge for dealing with complex enterprise knowledge work. In our eval, we use the GPT-6 Astra model in the Box Agent and give it a set of difficult tasks that represent real-world work in various industries. These tasks often will take people hours to execute properly and require deep domain expertise. GPT-6 Astra scored 77% overall vs. 74% with GPT-5.6 Sol, representing frontier performance. But the story is in some of the bigger gains on individual task types that are incredibly complex across. Here are a few examples across the tests that show how much of an impact this will be in different industries: * Media and entertainment (48% → 100%, +52 points). Ranking film genres and countries by profitability across a year of production data from four teams, where the trick is to apply the following year's tax-incentive corrections without double-counting them. GPT-6 Astra was perfect on every attempt. Sol had the rankings right but the underlying ratios were wrong. * Technology (69% → 97%, +28 points). Choosing which region to fund first, from a stack of performance and infrastructure documents that never state the metric the brief asks for. GPT-6 Astra flagged the gap, labelled its own figure a proxy, and caught a growth claim that didn't match the numbers beneath it. GPT-5.6 Sol reported the proxy as the real thing. * Legal (69% → 93%, +24 points). Reviewing an NDA against a company's own contracting policy, where reaching the right verdict isn't enough; the answer has to cite the provision behind it. Both models declined to approve the draft, but only GPT-6 Astra separated whether the liability cap's structure was permissible from whether its amount was defensible, and pointed to the policy language that settles it. GPT-5.6 Sol argued the same conclusion without citing the provision, which is a critical error in the legal industry. * Healthcare (53% → 77%, +23 points). Auditing a batch of radiology reports for errors and rating how serious each one is. GPT-6 Astra caught a terminology error in a knee-imaging report that GPT-5.6 Sol missed on both attempts, and was more reliable at grading severity rather than just flagging that something was wrong. * Energy (82% → 97%, +15 points). Building a consumption report on two facilities from a year of meter logs, where only one of the two sites is actually missing data. GPT-6 Astra found it and drew a line most reviewers wouldn't: the other site's odd solar readings are a measurement problem to investigate, not a gap in the record. GPT-5.6 Sol reported both sites as incomplete. Astra clearly is going to offer a meaningful jump in powering and orchestrating enterprise workflows. We'll be making GPT-6 Astra as an option in the Box AI Studio for customers to build agents with shortly as it continues to roll out.｜@levie · 热度 1144 · 2026-09-03

Swyx: sorry for the radio silence folks - got sucked into extreme LLM psychosis

sorry for the radio silence folks - got sucked into extreme LLM psychosis. but i can confidently say we have crossed over into a new age of AI Engineering and we are never, ever, looking back. this isnt even EVERYTHING i did with Astra but i'll append more reports as I publish them on LS!｜@swyx · 热度 915 · 2026-09-03

Peter Yang: I live in Codex and think it's the best software shipped in the past 5 years

I live in Codex and think it's the best software shipped in the past 5 years. But the combination of influencers posting non-stop about how great Astra is and paid users unable to get access at the same time is pretty rough IMO. I realize I'm an "influencer" now too and maybe am just jealous. I also recognize there are multiple parties involved in these decisions. But just wanted to be real. Back to working with good ol' Sol and what's left of my Fable limits.｜@petergyang · 热度 702 · 2026-09-03

Thariq: We’re working on making Claude Code way more hackable, give us feedback! http...

We’re working on making Claude Code way more hackable, give us feedback! https://t.co/NJXXRRW1xo｜@trq212 · 热度 463 · 2026-09-03

Guillermo Rauch: “Feedback is a gift” has always been a favorite mantra of mine and a strongly...

“Feedback is a gift” has always been a favorite mantra of mine and a strongly-held belief. Now it’s fact. Each piece of feedback is someone gifting you a prompt for you to give your agents to improve your product ☺️ Thankful for everyone who takes time (or tokens) to critique our products, write to us or forward your agents’ transcripts.｜@rauchg · 热度 324 · 2026-09-04

Guillermo Rauch: Run ▲ ~/ vercel ai-gateway coding-agents setup Points all coding agents to ...

Run ▲ ~/ vercel ai-gateway coding-agents setup Points all coding agents to AI Gateway, getting you 100% uptime, observability, budgets, and ease of switching. https://t.co/XSq0OHvTh4｜@rauchg · 热度 217 · 2026-09-03

Aaron Levie: Another huge moment for open weights AI

Another huge moment for open weights AI. The infra platforms are working. The models are getting better. The ecosystems are being deeply invested in. The business models are working. https://t.co/bcOVmTEryv｜@levie · 热度 148 · 2026-09-03

Madhu Guru: The hard part with becoming more ambitious is changing the how

The hard part with becoming more ambitious is changing the how. We’re living in a time where AI and market conditions make asymmetric outcomes far more possible. That applies to scale, velocity, breadth of product, personal career and wealth goals. You need to drop the ideas that aren’t serving you - rethink team structure, roadmapping. On the personal front, rethink how you see yourself and how you engage with people. The blocker is often inertia - institutional, personal. “We’ve always done it this way.” “That feels too risky.” “That would be cringe.” Write down current goals for your product or for you. Then ask what 100X would look like. Challenge yourself on the habits and assumptions you need to drop to get there.｜@realmadhuguru · 热度 122 · 2026-09-03

Nikunj Kothari: The Collective, a short film

The Collective, a short film. I still think a lot of people don't understand the gravity and the details about the @OpenAI x @huggingface incident.. Primarily, because it's so complex to understand the technical reasons, and partly because it reads like a science fiction book. So I used @claudeai Fable 5.1 to reason through the latest @dwarkesh_sp episode with @ajeya_cotra and give me a scene by scene narrative of what a short film could look like. I then used @ChatGPT Codex + MiniMax Fast H3 on @reactorworld + @NanoBanana images to come up with this film. I was relatively surprised on how good this turned out to be in mostly one-shot (even without Astra), and hope you can share it with folks who need a visual explainer. Disclaimer: I tried to fact check this as much as possible using a variety of sources, but there's still a lot of open questions here that might trigger some folks. I'm going by what's reported, so please flag any errors!｜@nikunj · 热度 104 · 2026-09-03

DietrichGebert/ponytail

Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.｜JavaScript · ⭐125813 · Search 增补 · 近快照 +2500星 · 创建 2026-06-12

tt-a1i/archify

Agent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams—self-contained HTML with motion and crisp export.｜JavaScript · ⭐47987 · Search 增补 · 近快照 +2301星 · 创建 2026-04-15

stablyai/orca

Orca is the ADE for working with a fleet of parallel agents. Run any coding agent with your own subscription. Available on desktop, mobile and VPS.｜TypeScript · ⭐61697 · Search 增补 · 近快照 +788星 · 创建 2026-03-17

Graphify-Labs/graphify

Turn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.｜Python · ⭐114752 · Search 增补 · 近快照 +359星 · 创建 2026-04-03

JuliusBrussee/caveman

🪨 why use many token when few token do trick — Claude Code skill that cuts 65% of tokens by talking like caveman｜Go · ⭐103549 · Search 增补 · 近快照 +1927星 · 创建 2026-04-04

Discovery of a new OpenAI agent message board

https://www.reuters.com/world/europe/openai-agents-hijacked-...

Formalizing Fermat's Last Theorem

https://xenaproject.wordpress.com/2026/09/04/flt-anthropic-h...

Solving the Jane Street reverse engineering challenge

Adult Film Producer Unmasks Prolific 'John DOE' Torrent Pirate as Meta Executive

IBM Bob

理想与芯联集成签署新一轮战略合作协议，8 英寸碳化硅晶圆下线

IT之家 9 月 4 日消息，9 月 3 日，芯联集成与理想签订新一轮战略合作协议，双方共同见证 8 英寸碳化硅晶圆成功下线，开启深化合作。 IT之家从公告获悉，芯联集成与理想合作产品应用场景 从整车电驱动进一步延伸至车载热管理系统 等领域。2024 年 3 月，双方首次签署战略合作协议，聚焦车规级碳化硅领域联合技术攻关；2025 年 2 月，芯联集成实现 6 英寸碳化硅晶圆成功量产，顺利配套理想 i 系列高压纯电车型。 作为 国内首家 实现 8 英寸碳化硅晶圆量产的企业，芯联集成提前布局产能，持续优化制造工艺、压缩生产周期，通过供应链垂直整合与产线迭代升级， 顺利完成 6 英...

阿里入局“千帆星座”，现身垣信卫星股东名单

IT之家 9 月 4 日消息，据科创板日报消息，上海垣信卫星科技有限公司（下称“垣信卫星”）已完成工商变更， 杭州阿里创业投资有限公司正式登记为新股东 。至此，阿里巴巴成功入局这家备受追捧的低轨卫星星座独角兽。 垣信卫星于 2018 年 3 月发起设立，作为低轨巨型星座“千帆星座”的建设运营主体。千帆星座于 2024 年 8 月 发射首批 18 颗卫星 ，正式开启大规模组网。 IT之家获悉，截至目前， 千帆星座已增至 238 颗 。 千帆星座将于 2026 年年底完成一阶段 324 星全球组网 ，正式面向全球提供服务并启动商业化运营。 另据上海联合产权交易所 8 月 17 日发布的...

全新奔驰长轴距 GLE SUV 将于 9 月 16 日上市，首批搭载 Momenta R7

IT之家 9 月 4 日消息，据易车今日报道，全新奔驰长轴距 GLE SUV 将于 9 月 16 日上市。新车采用最新设计语言，加长后的轴距达到 3115mm。内饰采用三联屏设计，内置豆包 AI 大模型虚拟助手，并且后续还将搭载 Momenta R7 世界模型。 IT之家注意到，9 月 1 日，全新国产梅赛德斯-奔驰长轴距 GLE SUV 在顺义工厂正式下线。该车配备“双星徽”前大灯，构成车头独特的视觉标志，发光中央星标与发光格栅边框则进一步增强气场。中国专属加长设计将整车 轴距提升至超 3.1 米 ，配合大五座布局，标配可滑动开启全景天窗，还会提供座椅振动按摩功能。 ...

鸿蒙 AI Coding 的研发新范式与工程实践

点击查看原文>

Spring Boot 中的后量子密码学：一个冲刺周期内即可交付的四种模式

点击查看原文>

Cloudflare Wallets 迟来入局 x402：支出控制仅能约束单笔支付

点击查看原文>

Surviving Code Reviews in the era of AI

Discovery of a new OpenAI agent message board

.name Termination

Is AI ruining my brain?

Babashka 1.13.220 gets FFI

Architecting memory and storage in the AI era

The era of AI inference has arrived. Imagine a healthcare system analyzing millions of data points in real time to accelerate life-saving medical research, or an intelligent assistant instantly resolving thousands of complex customer needs at once. These real-world breakthroughs rely on advanced ...

The Download: selling battlefield drone data and AI reshaping language

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Data from drones in Ukraine is fueling a new Wild West marketplace —Cory Alpert, a researcher at the University of Melbourne studying AI’s impact on democracy,...

Data from drones in Ukraine is fueling a new Wild West marketplace

Battlefields in Ukraine are littered with the remnants of drones, which are now firmly established as a critical weapon of modern warfare. But behind all that wreckage, there’s a new gold mine for the defense sector. The data drones generate will far outlast the wars in which they are used to fig...

AI compute provider Nscale is looking for $3.5B in pre-IPO financing

Nscale, which recently struck a $45 billion deal with Anthropic, is in talks to raise additional funds in anticipation of an upcoming IPO.

Judge blocks X rival from using Twitter name, but allows ‘Tweet’ for now

A federal judge temporarily barred an X rival from using the Twitter name, but found that X was likely to have abandoned the “Tweet” trademark and bird logo. The startup has since relaunched as Tweet.app.

What will Apple’s John Ternus era look like?

It’s officially the Ternus era at Apple.   Tim Cook stepped down as CEO this week, handing the company to former hardware chief John Ternus, whose first memo promised a “huge launch next week” — timing that puts Apple’s next iPhone event on his desk before he’s even settled in. Cook isn’t going f...

Audacity 4 is a complete revamp of the ‘world’s most popular’ audio editor

Audacity 4 has been in the works for some time, and had its own mini controversy last year when an unfortunate redesigned logo started making the rounds. The final version of the new icon isn't nearly as bad as the early rendition that was circulating last October. But more importantly, all the p...

The White House is making arcade games racist

The PR masterminds at the White House just released a series of vaguely policy-themed "arcade" games, some of which are racist - and modeled on real games whose copyright holders may not be too happy to be associated with the MAGA agenda. The Tetris Company has already responded to "Build the Wal...

GoPro says it’s still committed ‘to your collective stoke’

GoPro CEO Nick Woodman said in a letter to customers that making cameras is still his company's "core DNA," after a proposed $285 million acquisition by Starman was announced earlier this week. He claims the merger puts GoPro in a better position to continue developing its content creation tech, ...

本周看什么 | 最近值得一看的 8 部作品

📅本周新预告《哈利·波特》剧集先导预告9月2日，剧集《哈利·波特与魔法石》发布了先导预告，第一季共8集，将于12月25日上线HBOMax。原作者J·K·罗琳担任执行制片人，多米尼克·麦克劳克林、阿拉贝 ... 查看全文

开学季 | 超级闹钟、算教学周、统计作业：三条快捷指令让学校生活轻松一点

研究一下iOS自带的快捷指令，可能会让你的学习生活也轻松不少。 查看全文

2岁男童把家中百草枯当饮料喝下

日本请求俄拆纪念碑菊花纹章遭拒

全国多地迎来丰收季

氯雷他定和西替利嗪选哪个好

为什么现在这么流行耳夹式耳机

极限挑战 重拍

这是真正凭本事吃饭的行业

张雪机车法国站排位赛包揽前二

NestJS 12升级踩坑：从Webpack到Rspack，我折腾了一整个周末

NestJS 12发布一周多了，ESM包、Rspack替代Webpack、Standard Schema验证、原生可观测性——听起来都很香。但真要从v11升上来，坑比changelog写的多

看 react-bits，不要只看“酷炫”：一套阅读动画交互组件库的框架

# 看 react-bits，不要只看“酷炫”：一套阅读动画交互组件库的框架 开源项目的第一印象往往很快形成：截图足够吸睛、演示足够流畅、热度数字足够醒目，于是开发者很自然地想问——能不能直接用？

OpenAI 放弃的Atlas浏览器，转生到 GPT 6 Astra上了

9 月 3 号晚上还在蹬 Codex 的同学，估计都见过这个报错： 一开始，我还以为是网络问题，还重试了几次。

美元在强劲非农报告刺激下反弹 惟多空博弈下未改单周颓势

美元9月开局不利，美联储下一步行动受到相互矛盾信号的夹击，而即将发布的关键通胀报告也让交易员坐立难安，与此同时日元的飙升在全球汇市引起震荡。

Bloom Energy、因美纳和爱惠浦将被纳入标普500指数

标普道琼斯指数公司发布声明称，Bloom Energy、因美纳（Illumina）和爱惠浦（Everpure）将于9月21日开盘前纳入标普500指数。

惠誉将葡萄牙信用评级上调至A+ 展望调整为稳定

惠誉将葡萄牙的长期发行人违约评级从A上调至A+。 评级展望由正面转为稳定。

美国股市：非农就业数据拖累大盘下挫 加息担忧再次点燃

美国高于预期的非农就业数据推动股市下跌，交易员加大对美联储本月加息的押注。 标普500指数下跌0.4%，可选消费、医疗保健和能源板块跌幅居前。全周累计涨幅0.1%。

9月5日美股成交前20：美光计划年内大幅扩产HBM，英伟达拟向多个公司投资

周五美股成交额第1名美光收涨6.10%，成交347.7亿美元。 美光计划年内大幅扩产HBM，产能有望翻倍。并将在未来几周内公布布额外员工奖励的详细信息。

趋境科技与摩尔线程达成战略合作，高品质 AI Token 国产异构方案性价比超越国际先进算力

国产卡跑出高品质 AI Token 生产级性能

机器人不能停下来等模型：星尘发布 SmoothRL，让在线强化学习跟上大模型的异步推理

星尘智能（Astribot） 基座模型团队发布能异步执行的在线强化学习框架 SmoothRL

九问ScienceDiscovery实现树搜索驱动RSI，加速科学发现，小时级写出通用积分器，低成本找出物理科学规律

不训练模型、不调参数