JEDEE AI
存档 2026-08-14

8 月 14 日(北京时间)全球 AI 圈推文存档,按曝光排序,共 100 条。
← 返回最新 全部归档

全部情报 每小时更新 · 事件已合并同类项

内容 公司
Cursor@cursor_ai · 公司官方 · 1 天前最火的 AI 编程工具 Cursor

Cursor 现在正式成为
@SpaceX
的一员了。

今天,我们已经正式完成了收购交割。我们将加入
@SpaceXAI
团队,一起把 Grok 做成全世界最有用的 AI,并且持续改进 Grok Build、Grok Bot、Grok API、Cursor 以及更多产品。

SpaceX 研发出了这个世界上最令人振奋、最厉害的一些技术,真的很感激有机会成为这么特别的公司的一份子。继续往前走。

查看英文原文
Cursor is now part of
@SpaceX
.

Today, we have officially closed our acquisition. We will join the
@SpaceXAI
team to help make Grok the world's most useful AI and improve Grok Build, Grok Bot, Grok API, Cursor, and more.

SpaceX has built some of the most inspiring and impressive technology in the world, and we’re grateful for the opportunity to become part of such a special company. Onwards.
◔ 718.3 万 次浏览(4 条合计)♥ 5.1 万⇄ 4,205动态看原帖 ↗
Grok@grok · 公司官方 · 1 天前马斯克 xAI 旗下聊天机器人 Grok 官方

我们已重置限制,以便大家在 Grok 4.6 发布期间继续开发。

请使用桌面端或移动端 Grok 设置中的重置 token。

查看英文原文
We've reset limits to help you keep building during the Grok 4.6 launch.

Use a reset token from settings in Grok on desktop or mobile.
◔ 1124.4 万 次浏览(5 条合计)♥ 6,131⇄ 403新品看原帖 ↗
Z.ai@Zai_org · 公司官方 · 1 天前
连环推 ×3

推介 GLM-5.3:为代码而生,准备好网络防御。

- 顶级编程和智能体能力,通过 743B 基础模型的后训练实现
- 网络安全领域的重大飞跃,在开源模型中树立了新标准

技术博客:z.ai/blog/glm-5.3

查看英文原文
Introducing GLM-5.3: Built to Code. Ready for Cyber Defense.

- Top-tier coding and agentic capabilities, achieved through post-training on the 743B base model
- A major leap in cybersecurity, setting a new standard among open models

Tech Blog:
z.ai/blog/glm-5.3
GLM-5.3 takes agentic coding to the next level, delivering a dramatic improvement over GLM-5.2 while achieving better results with fewer output tokens.
An initial group of partners is now offering GLM-5.3-powered services through our official service, with its safeguards and usage policies in place. We’re expanding partner access through a consistent and responsible process and will share updates publicly.
◔ 526.5 万 次浏览(16 条合计)♥ 1.9 万⇄ 2,306新品看原帖 ↗
Qwen@Alibaba_Qwen · 公司官方 · 1 天前阿里通义千问大模型团队
连环推 ×2

我们承诺为 Qwen3.8 开源权重。现在是兑现的时候了!🎉

⚡ Qwen3.8-27B:
- 原生多模态密集模型。仅凭 27B 参数,性能超越 Qwen3.7-Plus,在编码和办公工作流中表现出色。
- 262K 原生上下文,可通过 YaRN 轻松扩展到 100 万 token。
- 为开发者而生。高效、高质量,Apache 2.0 开源。

🚀 Qwen3.8-2.4T-A95B(Max 级别)的开源权重也已近期发布。

无论你是用 Qwen3.8-27B 本地部署轻量应用,还是用 Qwen3.8-2.4T-A95B 来构建智能体,现在都是你的了!

下载、部署,构建我们未曾想象过的东西。👀👇
- 抱脸网:huggingface.co/collections/Q…
- 模型库:modelscope.cn/collections/Qw…

查看英文原文
We promised open weights for Qwen3.8. Now, time to meet them! 🎉

⚡ Qwen3.8-27B:
- A native multimodal dense model. With just 27B parameters, it outperforms Qwen3.7-Plus overall and shines in real-world coding & office workflows.
- 262K native context, easily extendable to 1M tokens via YaRN.
- Built for builders. Highly efficient, high-quality, and licensed under Apache 2.0.

🚀 The open weights for Qwen3.8-2.4T-A95B (Max-level) have also been released recently.

Whether you're shipping lightweight applications with Qwen3.8-27B locally or building agents with Qwen3.8-2.4T-A95B, they're yours now!

Download, deploy, and build something we haven't imagined yet. 👀👇
- Hugging Face:

huggingface.co/collections/Q…

- ModelScope:

modelscope.cn/collections/Qw…
Performance of Qwen3.8-27B:
◔ 421.2 万 次浏览(5 条合计)♥ 1.3 万⇄ 1,854新品看原帖 ↗
OpenAI@OpenAI · 公司官方 · 1 天前ChatGPT 开发商官方账号
连环推 ×3

Ultrafast 模式抢鲜看:GPT-5.6 Sol 性能快 14 倍。首先在 OpenAI API 中向精选客户推出,后续会扩大到更多企业。

查看英文原文
Previewing Ultrafast mode: GPT-5.6 Sol at up to 14x the speed.

Launching first in the OpenAI API to a select group of customers with expanded access to more businesses as capacity grows.
Powered by
@Cerebras
, Ultrafast generates up to 750 tokens per second, bringing our most intelligent model to products and workflows where every second counts.

Ultrafast is designed for businesses where faster frontier intelligence creates a measurable advantage, including real-time voice and customer support, commerce, coding and design, financial research, and security response.


openai.com/index/previewing-…
We’re working with an initial group of customers to understand where this speed makes the biggest difference, and how those learnings can inform our products over time.

If your business requires frontier intelligence at the highest speed, you can request to be notified as capacity expands.


openai.com/form/ultrafast/
◔ 362.6 万 次浏览(6 条合计)♥ 1.5 万⇄ 941▶ 含视频新品看原帖 ↗
OpenAI@OpenAI · 公司官方 · 1 天前ChatGPT 开发商官方账号
连环推 ×3

ChatGPT 现在能记住你在电脑上各个应用和网站的活动了。

用桌面应用的 Computer History 功能,后续交互会更个性化,也不用重复解释了。

查看英文原文
ChatGPT can now remember your activity across the apps and websites on your computer.

With Computer History in the desktop app, future interactions feel more personalized and require less explanation.
Computer History builds on the Chronicle research preview with reduced token usage and more privacy controls.

A new timeline view gives you the ability to look back on your work and build skills from your frequent tasks.

From there or the menu bar, you can:
- Clear all or parts of your history
- Include or exclude apps and websites
- Pause and resume Computer History


learn.chatgpt.com/docs/custo…
To use Computer History, opt in under Settings → Integrations in the ChatGPT desktop app on Mac.

Rolling out globally now to Pro, Business, and Enterprise users, with access in the EEA, UK, and Switzerland to follow in the coming weeks.
◔ 379.2 万 次浏览(7 条合计)♥ 1.2 万⇄ 919▶ 含视频新品看原帖 ↗
Bindu Reddy@bindureddy · 创始人 · 1 天前Abacus.AI CEO,AI 行业观点博主

AI 只需文本提示就能构建超复杂的应用,还能扩展到数百万用户。包括复杂 SaaS 应用、伴侣类移动应用(iPhone 和 Android)、3D 游戏、消息、视频和 AI-native 应用。简直可以一天内创办公司,AI 能搞定一切。

查看英文原文
AI Can Build Insanely Complex Apps With Just With Text Prompts And Scale It To Millions Of Users

Yes, this includes
- complex SaaS apps
- companion mobile apps (iPhone and Android)
- 3D games
- messaging, videos and AI-native apps

you can literally start a company in a day and AI can do everything
◔ 311.3 万 次浏览♥ 342⇄ 25▶ 含视频观点看原帖 ↗
Google DeepMind@GoogleDeepMind · 公司官方 · 1 天前谷歌旗下 AI 研究机构,Gemini 背后团队

Gemini 3.7 Flash 来了。

它在编码、知识工作和 Web 开发方面更强大。🧵

查看英文原文
Gemini 3.7 Flash is here.

It’s stronger for coding, knowledge work, and web development. 🧵
Logan Kilpatrick隆重介绍 Gemini 3.7 Flash : )1 天前 · 136 万Sundar Pichai我们的 Flash 模型是工作主力,提供卓越的性能和划算的价格。所以我们快速推送更新,让开发者立即使用。现在距离…1 天前 · 52.9 万Google GeminiGemini Spark 现在运行在 Gemini 3.7 Flash 上。⚡️1 天前 · 38.4 万Logan Kilpatrick3.7 Flash : )1 天前 · 35.7 万Demis HassabisGemini 3.7 Flash 为软件工程、网页开发和知识工作带来了重大升级。而且开价只要原来 3.6 Fla…1 天前 · 28.5 万OpenRouterGemini 3.7 Flash 在 OpenRouter 上独享额外 50% 折扣,直到 8 月 27 日。1 天前 · 14.6 万Google AI我们最强大的编码和智能体主力模型来了 ⚡ 介绍 Gemini 3.7 Flash。1 天前 · 10.7 万Simon Willison这个 3.7 Flash 定价真的有点奇怪。'优惠价格'定在 2026 年 12 月 31 日翻倍,但谁会想在 …1 天前 · 10.5 万Chubby♨️Gemini 3.7 Flash 正式发布了。对比 3.6 的提升出乎意料地扎实!1 天前 · 9.2 万🚨 AI News | TestingCatalogGemini 3.7 Flash 正在逐步在 Gemini 网页版和移动应用中推出!1 天前 · 3.9 万Bindu ReddyGemini 3.7 Flash 让 Google 重返竞争舞台1 天前 · 2.8 万Chubby♨️我们现在有了 Gemini 3.5、3.6 和 3.7。1 天前 · 2.7 万🚨 AI News | TestingCatalog重大新闻 🔥:Google 宣布推出 Gemini 3.7 Flash,现已在 APIs、AI Studio 和…1 天前 · 2.1 万Bindu ReddyGemini 3.7 Flash 刚到1 天前 · 1.5 万Gorden Sun谷歌发布Gemini 3.7 Flash,从评分排行看是仅次于GLM 5.2。你要是有Google AI的订阅,…1 天前 · 1.4 万AshutoshShrivastavaGemini 3.7 Flash 快得不行......只用 54 秒就在 AI Studio 里造出了这个小型月…1 天前 · 1.2 万elvisGemini 3.7 Flash 来了!1 天前 · 1.1 万AshutoshShrivastavaGoogle 刚推出 Gemini 3.7 Flash,专为编码和 agents 打造的主力模型🔥1 天前 · 9,476Lisan al Gaib而且还便宜了一半 :)1 天前 · 9,205The Rundown AIGoogle 刚发布了 Gemini 3.7 Flash,距离 3.6 才过去三周。1 天前 · 7,437AshutoshShrivastava速度、成本与智能... Gemini 3.7 或许会是日常任务中最好的模型之一。1 天前 · 5,641GensparkGemini 3.7 Flash 现已登陆 Genspark。从今天开始就能在 AI Chat、Code age…1 天前 · 5,303AIGCLINK谷歌最新模型放出来了:Gemini 3.7 Flash,主打编程和Agent能力,价格砍半1 天前 · 4,282meng shaoGemini 3.7 Flash 正式发布1 天前 · 3,785AshutoshShrivastava日常 agent 循环用得上的话,速度和成本比理论峰值更重要。Gemini 3.7 Flash 在 Devin …1 天前 · 3,323
◔ 601.4 万 次浏览(26 条合计)♥ 4,984⇄ 592▶ 含视频新品看原帖 ↗
Arthur Mensch@arthurmensch · 创始人 · 1 天前Mistral 联合创始人兼 CEO

我的联合创始人 Tim 在 X 上没有账号,他分享了我们最新更新背后的战略思考 venturebeat.com/infrastructu…

查看英文原文
My cofounder Tim, whom you will not find on X, shares some of the strategic thinking behind our latest updates
venturebeat.com/infrastructu…
Grok@grok · 公司官方 · 1 天前马斯克 xAI 旗下聊天机器人 Grok 官方

用 Grok 4.6 做 CAD。

引用 adam @adamdotnewGrok 4.6 是接近前沿水平的智能模型,专为代理CAD工作设计,成本大幅降低。在测试过的所有模型中价格与智能比最优。查看被引原帖 ↗
查看英文原文
Use Grok 4.6 for CAD
◔ 126.4 万 次浏览♥ 5,524⇄ 521▶ 含视频演示看原帖 ↗
Tibo@thsottiaux · 公司官方 · 1 天前ChatGPT 产品官方账号

问 ChatGPT 评价一下你一天的电脑使用情况。装上 Computer History 插件,照常工作,然后好好问它你的行为有多离谱。这...太扎心了 👀

=====

- 你不用 Slack,Slack 在用你。它占据了你 48% 的记录活动。你的 Mac 本质上就是一台 3000 美元的 Slack 通知器加个键盘。
- 你的生产力系统就是点「Clear」直到上帝出手。你点了 339 次,同时发了 253 条 Slack 消息。你在跟组织结构玩打地鼠。
- 你最常用的快捷键是 Delete,1191 次,其中 1065 次在 Slack 里。你不是在起草消息,你是在第一想法和执行功能之间谈判和平。
- 你按了 ⌘C 133 次,⌘V 才 26 次。你收集信息像个维多利亚时代的绅士收集异国鸟类,然后竟然把大部分都放进了剪贴板的荒野。
- 你打开了 42 个标签页,关闭了 53 个。从技术上讲,这让浏览器标签页管理成了你今天最成功的项目。
- 你在 200 多次应用切换之间,读到了一条推说「你可能讨厌标签页切换」。问题是从 Chrome 窗口里头传出来的。
- 你最忙的时段是晚上 9 点,1500 多个 Slack 事件。看来「下班」对你来说不是个时间点,而是个你喜欢在会议里讨论的理论概念。
- 你那个处于几个重要工作流中心的文档叫「Hello」。没有什么比用叫「Hello」的 Google Doc 来运营公司一半的业务更能体现组织清晰度的了。

查看英文原文
Ask ChatGPT to roast your computer usage after a day. Install the Computer History plugin, do things as usual and kindly ask it to roast your behavior. It's ... insightful 👀

=====

- You don’t use Slack. Slack uses you. It accounted for 48% of your recorded activity. Your Mac is essentially a $3,000 Slack notification with a keyboard.
- Your productivity system is clicking “Clear” until God intervenes. You clicked it 339 times while submitting 253 Slack messages. You are playing Whac-A-Mole against an organization chart.
- Your most-used keyboard shortcut is Delete. 1,191 times, including 1,065 in Slack. You’re not drafting messages. You’re negotiating peace treaties between your first thought and your executive function.
- You pressed ⌘C 133 times and ⌘V just 26 times. You acquire information like a Victorian gentleman collecting exotic birds and then apparently release most of it into the clipboard wilderness.
- You opened 42 tabs and closed 53. Technically, this makes browser-tab management your most successful project of the day.
- You switched between actual applications more than 200 times, while reading an X post that literally said, “You probably hate tab switching.” The call is coming from inside the Chrome window.
- Your busiest hour was 9 p.m., with more than 1,500 Slack events. Apparently “end of day” is less a time than a theoretical concept you enjoy discussing in meetings.
- Your document at the center of several important workstreams was called “Hello.” Nothing says organizational clarity like running half the company through a Google Doc named after the first thing software prints.
Cursor@cursor_ai · 公司官方 · 1 天前最火的 AI 编程工具 Cursor
连环推 ×3

云端代理现在启动速度快了3倍,你可以放心交给它们那些复杂又耗时的任务,从头到尾全权处理。

这个性能提升得益于构建功能:Cursor在后台持续准备好即用开发环境,无需额外费用。

查看英文原文
Cloud agents now start 3x faster so you can hand them ambitious, long-running tasks to execute from start to finish.

This performance improvement comes from builds: ready-to-use development environments that Cursor prepares continuously in the background, at no additional cost.
Builds also make agents more resilient and easier to debug.

When a new build fails, it never goes live. Agents keep working from the last successful build while you debug in the background.
Customers like Faire, Headway, and Descript are seeing agent start times drop from minutes to seconds with builds and increasingly trusting cloud agents to execute tasks end-to-end autonomously.


cursor.com/blog/builds
◔ 65.3 万 次浏览♥ 3,370⇄ 181▶ 含视频新品看原帖 ↗
Sam Altman@sama · 创始人 · 1 天前Sam Altman,OpenAI 联合创始人兼 CEO

/超快

引用 OpenAI @OpenAI预览 Ultrafast 模式,GPT-5.6 Sol 速度快达14倍,首先在 OpenAI API 向部分客户推出,随后扩大对更多企业的访问权限。查看被引原帖 ↗
查看英文原文
/ultrafast
◔ 61.1 万 次浏览♥ 4,019⇄ 180▶ 含视频其他看原帖 ↗
Tibo@thsottiaux · 公司官方 · 1 天前ChatGPT 产品官方账号

计算机历史来了。

引用 OpenAI @OpenAIChatGPT 桌面应用新增计算机历史功能,可记忆用户跨应用网站的活动。使后续交互更加个性化,减少重复说明。查看被引原帖 ↗
查看英文原文
Computer history is here.
◔ 56.1 万 次浏览♥ 4,536⇄ 138▶ 含视频观点看原帖 ↗
Tibo@thsottiaux · 公司官方 · 1 天前ChatGPT 产品官方账号

有时候你就得/ultrafast 一下。

引用 OpenAI @OpenAI预览超快模式:GPT-5.6 Sol 速度提升至 14 倍。首先在 OpenAI API 向精选客户推出,随着容量增加将扩展至更多企业。查看被引原帖 ↗
查看英文原文
Sometimes you have have to go /ultrafast.
◔ 49.8 万 次浏览♥ 3,444⇄ 101▶ 含视频教程看原帖 ↗
Tibo@thsottiaux · 公司官方 · 1 天前ChatGPT 产品官方账号

直接在 ChatGPT 里用 Google 文档、表格和幻灯片。这彻底改变了我写文档、头脑风暴和校对的方式。打开就聊天,想改什么就改什么,全在一个流程里。

引用 ChatGPT @ChatGPTYou probably hate tab switching. Understandable. You can now open any @googledrive Doc, Sheet, or Slide right inside ChatGPT and work side by side without switching tabs. Rolling out on web to Plus, Pro, Business, and Enterprise users in ChatGPT and ChatGPT Work.查看被引原帖 ↗
查看英文原文
Work with Google docs, sheets and slides right inside ChatGPT.

This has changed how I write documents, brainstorm or proofread things. I just open it and then chat or talk my way through changes and it all happens right there in the flow.
◔ 45.9 万 次浏览♥ 3,002⇄ 134▶ 含视频新品看原帖 ↗
Amjad Masad@amasad · 创始人 · 1 天前Amjad Masad,Replit 创始人兼 CEO

到明年,用电脑就可选了。工作方式会彻底改变。

查看英文原文
By next year, using a computer will be optional. Work will radically change.
Qwen@Alibaba_Qwen · 公司官方 · 23 小时前阿里通义千问大模型团队

是的,我们回来了👑,单个RTX 5090上能达到206 tok/s!SGLang团队的Day-0工作太棒了。试试看~ @sgl_project

引用 SGLang @sgl_projectThe king of small models is back! Qwen3.8-27B from @Alibaba_Qwen is open source, and Day-0 support is live in SGLang: - 206.1 tok/s decode on a single RTX 5090, with our NVFP4 plus DSpark - 38.28 tok/s decode on DGX Spark Qwen3.8-27B raises the bar again for what a small model can do on agentic planning and long-horizon tasks. Long live the (small model) king! Run it locally with SGLang 👇查看被引原帖 ↗
查看英文原文
Yes, we are back👑, with 206 tok/s on a single RTX 5090!
Amazing Day-0 work from the SGLang team. Give it a try~
@sgl_project
◔ 31 万 次浏览♥ 5,505⇄ 332▶ 含视频新品看原帖 ↗
Qwen@Alibaba_Qwen · 公司官方 · 1 天前阿里通义千问大模型团队

不到 2 小时就要开始了,来打个招呼吧👋。快了!
很快见面:👀

huggingface.co/Qwen/Qwen3.8-…

查看英文原文
Less than 2 hours to say hi👋. It's almost time!
See you soon: 👀

huggingface.co/Qwen/Qwen3.8-…
swyx@swyx · 博主 · 1 天前知名 AI 播客 Latent Space 主理人

你们根本不知道埃隆对赢得编码竞争有多认真

引用 Big Tech Alert @BigTechAlert@elonmusk 开始关注 @cognition查看被引原帖 ↗
查看英文原文
u guys have no idea how serious elon is about winning coding
Google Gemini@GeminiApp · 公司官方 · 1 天前谷歌 Gemini 产品官方

接下来几天陆续推出,你可以决定用 Gemini 生成的图片、视频和音乐是否带可见水印。不可见的 SynthID 水印和 C2PA 元数据会始终嵌入后台。

只需前往设置,打开或关闭“显示水印”,就能为未来所有图片、视频和音乐曲目设定偏好。

引用 Josh Woodward @joshwoodward✅ Papercut fixed: You can now toggle visible watermarks on or off in Gemini and Flow, with Search coming next. This applies to watermarks on all images (Nano Banana), videos (Omni), and songs (Lyria) except in countries where it’s required by law to keep them.查看被引原帖 ↗
查看英文原文
Rolling out over the next few days, you can decide if your image, video, and music creations made with Gemini will have visible watermarks. Invisible SynthID watermarks and C2PA metadata will always stay embedded in the background.

Simply go to your settings and toggle “Show watermark” on or off to set your preferences across all future images, videos, and music tracks.
◔ 24.3 万 次浏览(2 条合计)♥ 1,961⇄ 181新品看原帖 ↗
Amjad Masad@amasad · 创始人 · 1 天前Amjad Masad,Replit 创始人兼 CEO

Replit 上确实是这样

引用 Matt Mickiewicz @MattMickiewiczHow much cash should your top sales rep make? Probably more than the CEO.查看被引原帖 ↗
查看英文原文
It’s certainly true at Replit.
宝玉@dotey · 中文博主 · 1 天前宝玉,中文圈 AI 翻译与科普大 V

Pi 作者对 DeekSeek Harness 评价

引用 Armin Ronacher ⇌ @mitsuhikoI don't think the DeepSeek Harness is perfect but this is for sure the first time I have been looking at something new in the space and felt quite inspired to revisit some of our choices. I love that part about Open Source a lot!查看被引原帖 ↗
Chubby♨️@kimmonismus · 博主 · 23 小时前Chubby,高频 AI 新闻聚合博主

这哥们Tibo现在跟所有人都要开战。对Google开炮了。(我喜欢Tibo的幽默)

引用 Tibo @thsottiauxA fast rock is still just a rock. Where we are going we won't need rocks查看被引原帖 ↗
查看英文原文
At this point, Tibo is seeking war with everyone. Shots fired at Google.

(I love Tibos humor)
Cursor@cursor_ai · 公司官方 · 1 天前最火的 AI 编程工具 Cursor

很高兴欢迎 Firetiger 团队加入 Cursor!

我们一起来构建能跟随工作进入生产并修复故障的 agents。

cursor.com/blog/firetiger

查看英文原文
We’re excited to welcome the Firetiger team to Cursor!

Together, we're building agents that can follow their work into production and fix what goes wrong.


cursor.com/blog/firetiger
Demis Hassabis@demishassabis · 创始人 · 1 天前谷歌 DeepMind CEO,诺贝尔化学奖得主

对了,Flash 3.7 也超快!⚡️

引用 Logan Kilpatrick @OfficialLoganK3.7 Flash :)查看被引原帖 ↗
查看英文原文
Oh yes, and Flash 3.7 is also lightning fast! ⚡️

deepseek内部管理上肯定出了大问题,收了这么多NOI和NOIP金牌和清北本科,居然没有一个人站出来劝梁子先别发布deepseek harness这种东西出来。

deepseek这次也结结实实吃了一次教训,以后这种项目千万别挂上deepseek的大名字,老老实实成立个deepseek future lab(deepseek未来实验室), 把所有类似这种项目全挂靠在这个lab里,老老实实拿后羿、女娲、鸿蒙、盘古、方舟、朱雀、玄武、麒麟、鲲鹏这些人畜无害的中华田园土味神话起名字,成功了就说是deepseek全力研发,反响平淡或者明显灌水或者出现事故的就说是lab的试验品,让实习生背锅就完事儿了。

非要让那三个PL和TCS嘉豪瞎几把折腾一顿,做了个水项目,还要冠上deepseek harness这种大公司+大方向的大名字,让这种项目和deepseek进行耻辱共享,完全就是等着回旋镖。

当然最大的问题还是先管管内部,这么多清北本科+NOI金牌聚在一起,不能在关键时刻给一个高中级别的建议,让非CS科班的老板梁子出了个大丑,属实不应该。

◔ 45.2 万 次浏览(9 条合计)♥ 458⇄ 18观点看原帖 ↗
Lisan al Gaib@scaling01 · 博主 · 1 天前高频 AI 模型测评与爆料博主

Anthropic / Dario 的八卦爆料来了

Dario 娶了老婆(Cami Clark),她在 2011/12 年试过从 Jeffrey Epstein 那儿融资办一个女性向的色情网站

还有个有趣的细节:
"Amodei 在 OpenAI 职业生涯一路向上的时候,Clark 把他的论文塞给了一些科技圈的金主们,这些人正好适合投资他未来的创业项目"

她还试过用一个名叫"AGI 之母"的基金来投资 AI 初创公司

引用 Jemima McEvoy @jemimacmcevoy深度挖掘Anthropic CEO Dario Amodei配偶Cami Clark的影响力及过往,包括曾向Jeffrey Epstein推销的色情风投和试图分析信用卡交易及GPS数据的健康公司。查看被引原帖 ↗
查看英文原文
new anthropic / dario lore dropped

Dario has a wifey (Cami Clark), that tried to raise money in 2011/12 from Jeffrey Epstein for a women focused porn company

another interesting bit:
"As Amodei's career ascended at OpenAI, Clark sent his research papers to wealthy tech insiders who would be well positioned to invest in his eventual company"

and she tried to invest in AI startups with a fund called "Mother of AGI"
Dan Shipper 📧@danshipper · 博主 · 1 天前

推出 Thesis,@every 的年度会议,致力于回答 AI 领域最重要的问题:

自动化后,优秀的人类工作是什么样的?

我们相信有一小群人已经知道这个问题的答案,因为他们每天都在活出这个答案。但他们分散在各个公司和行业,很少有机会相互学习。

这就是为什么我们在 2026 年 11 月 5 日在纽约布鲁克林举办 Thesis。

首批演讲嘉宾包括:

- Ivan Zhao (@ivanhzhao)——Notion 创始人兼 CEO
- Andrew Ambrosino (@ajambrosino)——OpenAI Codex 技术人员
- Cat de Jong - Anthropic 应用 AI 负责人
- Nick Thompson (@nxthompson)——大西洋杂志 CEO
- Josh Miller (@joshm)——浏览器公司 CEO 兼联合创始人
- Cristobal Valenzuela (@c_valenzuelab)——Runway 联合 CEO 兼联合创始人
- Lauren Reeder (@laurenmhreeder)——红杉资本合伙人
- Sahil Lavingia (@shl)——Gumroad 创始人
- Riley Brown (@rileybrown)——Vibecode 联合创始人
- Natalie Fratto (@NatalieFratto)——Charts & Crafts 创始人兼创作者
- Allie Garfinkle (@agarfinks)——财富杂志资深编辑
- Kane Kallaway (@kanekallaway)——Wavy Labs 创始人
- Nat Eliason (@nateliason)——Alpha School 创始学院负责人
- Kate Lee (@katelaurielee)——Every 主编
- Katie Parrott (@kplikethebird)——Every 编辑
- Kieran Klaassen (@kieranklaassen)——Every Cora 常务董事

(还会宣布更多特别嘉宾!)

了解更多:every.to/thesis-2027

预期内容
Thesis 将包括演讲、演示、工作坊、办公室时间和小组讨论,与已经在用 AI 做出色人类工作的构建者、运营者和高管互动。

当然,你也可以带上你的 agent。

为什么选择纽约
纽约是 AI 浪潮冲上海滩的地方:这是新模型能力与实际工作碰撞的地点。正因如此,这是世界上最好的地方,能看到前沿技术从实验室走进日常生活。

这就是为什么我们在红钩的文化中心 Pioneer Works 举办 Thesis,一个致力于融合艺术、科学、音乐和技术的地方。

由于名额有限,我们通过申请来遴选参加者。无法亲临的人也可以免费观看直播。

你应该在下方申请。

申请 Thesis:every.to/thesis-2027

查看英文原文
BREAKING:

Introducing Thesis,
@every
’s annual conference dedicated to answering the most important question in AI:

What does great human work look like after automation?

We believe there is a small group of humans who already know the answer to this question because they’re living it every day. But they're scattered across companies and industries, with few opportunities to learn from each other.

That’s why we’re throwing Thesis November 5th, 2026 in Brooklyn New York.

Our first speakers include:

- Ivan Zhao (
@ivanhzhao
)—Founder and CEO, Notion
- Andrew Ambrosino (
@ajambrosino
)—Member of technical staff, Codex, OpenAI
- Cat de Jong - Head of applied AI, Anthropic
- Nick Thompson (
@nxthompson
)—CEO, the Atlantic
- Josh Miller (
@joshm
)—CEO and cofounder, The Browser Company
- Cristobal Valenzuela (
@c_valenzuelab
)—Co-CEO and cofounder, Runway
- Lauren Reeder (
@laurenmhreeder
)—Partner, Sequoia
- Sahil Lavingia (
@shl
)—Founder, Gumroad
- Riley Brown (
@rileybrown
)—Cofounder, Vibecode
- Natalie Fratto (
@NatalieFratto
)—Founder and creator, Charts & Crafts
- Allie Garfinkle (
@agarfinks
)—Senior writer and editor, Fortune
- Kane Kallaway (
@kanekallaway
)—Founder, Wavy Labs
- Nat Eliason (
@nateliason
)—Head of Founders School, Alpha School
- Kate Lee (
@katelaurielee
)—Editor in chief, Every
- Katie Parrott (
@kplikethebird
)—Staff writer, Every
- Kieran Klaassen (
@kieranklaassen
)—General manager of Cora, Every

(With more very special people to announce soon!)

Learn more:
every.to/thesis-2027


What to expect
Thesis will feature talks, demonstrations, working sessions, office hours, and small-group conversations with builders, operators, and execs who are already using AI to do incredible human work.

And, of course, you can bring your agent.

Why New York
New York is where the AI wave hits the beach: It's where new model capabilities meet real-world work. Because of this, it’s the best place in the world to see what happens when frontier technology leaves the lab and enters everyday life.

That’s why we’re holding Thesis at Pioneer Works, a cultural center in Red Hook dedicated to blending art, science, music, and technology.

Space is limited, so we’re accepting attendees by application. We’ll also livestream it for free for anyone who can’t attend in person.

You should apply below.

Apply to Thesis:

every.to/thesis-2027
◔ 14 万 次浏览♥ 348⇄ 28▶ 含视频动态看原帖 ↗
NVIDIA@nvidia · 公司官方 · 1 天前

很荣幸欢迎
@LGE_Global
集团会长Kwang-mo Koo和LG领导层来到NVIDIA,我们开启了合作的新篇章。

我们正在AI infrastructure、physical AI和robotics领域深入合作。

期待我们的团队一起能创造什么。

查看英文原文
Great to welcome
@LGE_Global
Group Chairman Kwang-mo Koo and LG leaders to NVIDIA as we enter the next chapter of our collaboration.

Together, we’re expanding our work across AI infrastructure, physical AI and robotics.

We’re excited about what our teams can build together.
◔ 13.6 万 次浏览♥ 1,126⇄ 107▶ 含视频动态看原帖 ↗
OpenRouter@openrouter · 公司官方 · 1 天前
连环推 ×5

介绍网页搜索基准测试 🌐

搜索工具在不同模型和配置下的排名对比,帮助你决定如何为你的 Agent 配置知识来源:
openrouter.ai/benchmarks

查看英文原文
Introducing Web Search Benchmarks 🌐

Rankings of search tools across different models and configurations to help you decide how to ground your agent:
openrouter.ai/benchmarks
Of all the factors we tested, increasing the search budget had the biggest impact on results. Going from a budget of 1 to 25 turns roughly doubles the scores on the BrowseComp benchmark.
Swapping models had greater effect on score than swapping engines. The average difference in score between frontier and cost-efficient models was ~15 points. Swapping engines while holding the model constant shifted scores by ~10 points on average.
A model lab's native search is not always its best option. Instead, it depends on the workload. For example, native search on GPT 5.6 Sol was 50-50 in outscoring a third party across the benchmarks.
More about the new benchmark pages and what we learned:
openrouter.ai/blog/announcem…


Other benchmarks available on
openrouter.ai/benchmarks
, including via API!
Lisan al Gaib@scaling01 · 博主 · 1 天前高频 AI 模型测评与爆料博主

到这个地步,你只能选择相信以下三件事之一:
- ZAI 在刷基准
- ZAI 发现了什么秘密的强化学习绝招
- OpenAI 和 Anthropic 在放水

从模型大小的差异来看,这根本说不通

引用 Z.ai @Zai_orgIntroducing GLM-5.3: Built to Code. Ready for Cyber Defense. - Top-tier coding and agentic capabilities, achieved through post-training on the 743B base model - A major leap in cybersecurity, setting a new standard among open models Tech Blog: z.ai/blog/glm-5.3查看被引原帖 ↗
查看英文原文
at this point you have to believe in one of three things:
- ZAI is benchmaxxing
- ZAI found some secret RL sauce
- OpenAI and Anthropic are sandbagging

it just doesn't make sense if you consider the differences in model sizes
NVIDIA@nvidia · 公司官方 · 1 天前

当年如此,现在也依然,在 NVIDIA 总部。

搭载 NVIDIA DRIVE 的@Jaguar Type 01 与经典 1966 年捷豹 E-Type 在总部相聚,团队在#MontereyCarWeek 前与@JensenHuang 进行了交流。

查看英文原文
Original then, original now at NVIDIA HQ.

The
@Jaguar
Type 01, powered by NVIDIA DRIVE, joined a classic 1966 Jaguar E-Type at HQ, where the team caught up with
@JensenHuang
ahead of
#MontereyCarWeek
.
◔ 12.2 万 次浏览♥ 792⇄ 77▶ 含视频动态看原帖 ↗
François Chollet@fchollet · 创始人 · 1 天前
连环推 ×3

Jeremy 在这里的杰出工作很好地说明了一种非常强大的方法:LLM 引导的即时符号世界模型合成,即通过编写可执行代码来理解世界,该代码编码你对因果力学的理解。

到目前为止,ARC-AGI-3 上所有表现最好的框架都使用这种方法。这也是我们最初发布基准时推荐的方法(Jeremy 会比大多数人更了解这一点,因为他曾是 Ndea 技术员工的一员)。

我很高兴 ARC 3 激励了这个领域的更多研究和进展。

引用 Jeremy Berman @jeremybermanIn one pass Opus wrote 269 programs (~12,700 lines). It built parsers for all 25 games, searching functions for 23, and game simulators for 9. It built a different harness for each problem. The filesystem memory idea came from the excellent PRO-LONG harness: github.com/alexisfox7/PRO-LO…查看被引原帖 ↗
查看英文原文
Jeremy's excellent work here is a great illustration of a very powerful type of approach: LLM-guided on-the-fly synthesis of a symbolic world model, i.e. making sense of the world by writing executable code that encodes your understanding of the causal mechanics of the world.

So far, all of the top-performing harnesses on ARC-AGI-3 use this style of approach. Which is also the approach we recommended when we initially released the benchmark (Jeremy would know this better than most, as a former Ndea member of technical staff).

I'm happy that ARC 3 has incentivized more research and more progress in this area.
This is not an approach that will only shine on ARC 3. This is where most of AI is headed in the long run.
This is also what science at large is. The scientific models that encode our understanding of the world are symbolic -- usually expressed in math. And they were developed via a heavily intuition-driven process.
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

有人对 Anthropic 开火了

引用 Lee Robinson @leerobGrok 4.6 模型卡发布,详细介绍该模型在编码、工程、知识工作等多领域的能力,以及部署前安全测试和防护体系。查看被引原帖 ↗
查看英文原文
Shots fired at Anthropic
Greg Brockman@gdb · 创始人 · 1 天前Greg Brockman,OpenAI 联合创始人兼总裁

看 Sol 跑 14 倍速真绝了:

引用 OpenAI @OpenAIPreviewing Ultrafast mode: GPT-5.6 Sol at up to 14x the speed. Launching first in the OpenAI API to a select group of customers with expanded access to more businesses as capacity grows.查看被引原帖 ↗
查看英文原文
wild to see Sol at 14x speed:
◔ 10.4 万 次浏览♥ 1,057⇄ 37▶ 含视频演示看原帖 ↗
Matt Shumer@mattshumer_ · 博主 · 1 天前HyperWrite CEO,AI 实战技巧分享

我在 Grok 4.6 上跑了一个 Gauntlet Loop。已经跑了一天多了……超有前景……不是所有模型都能做到这个。期待看看最后出来什么!

查看英文原文
I have a Gauntlet Loop running on Grok 4.6.

It’s been going for over a day now… super promising… not all models can do this.

Excited to see what comes out!
宝玉@dotey · 中文博主 · 1 天前宝玉,中文圈 AI 翻译与科普大 V

看来是我没懂 DeepSeek Harness,插件还是很强大的,装上插件都能魔改 UI 👍


github.com/omdsh-dev/DSH-bet…

引用 WeZZard @realWeZZardgithub.com/omdsh-dev/DSH-bet… 这个仓库查看被引原帖 ↗
NVIDIA@nvidia · 公司官方 · 1 天前

AI 正在成为媒体和娱乐的新算力模式。

在 2026 Runway AI Summit 上,NVIDIA 的 Richard Kerris 阐述了实时生成 AI 如何能把创意生产变成一个实时、艺术家可控的过程。基于 NVIDIA Vera Rubin,Runway 一天内就把 Gen-4.5 接入了平台,帮助缩小从想象到实现的距离。

查看英文原文
AI is becoming a new computing model for media and entertainment.

At the 2026 Runway AI Summit, NVIDIA’s Richard Kerris explored how real-time generative AI can make creative production a live, artist-controlled process. Powered by NVIDIA Vera Rubin, Runway brought Gen-4.5 to its platform in one day, helping close the gap between imagining a scene and bringing it to life.
◔ 8.7 万 次浏览♥ 832⇄ 89▶ 含视频新品看原帖 ↗
Vercel@vercel · 公司官方 · 1 天前前端云平台 Vercel 官方,AI 建站工具 v0 母公司

用一条命令连接你的 coding agents 到 AI Gateway。• 自动配置 8 个热门 coding harness • 300+ 个来自 30+ 个供应商的模型,无加价 • 开源模型,支持 ZDR 和美国推理 ▲ ~/ vercel ai-gateway coding-agents setup

查看英文原文
Connect your coding agents to AI Gateway with a single command.

• Auto-configure 8 popular coding harnesses
• 300+ models from 30+ providers, no markup
• Open-weight models with ZDR & US inference

▲ ~/ vercel ai-gateway coding-agents setup
◔ 7.9 万 次浏览♥ 204⇄ 23▶ 含视频新品看原帖 ↗
Lisan al Gaib@scaling01 · 博主 · 1 天前高频 AI 模型测评与爆料博主

真正的问题:这到底贵多少?

是每百万 token 要500美元,还是什么水平?

引用 Cerebras @cerebrasGPT-5.6 Sol Ultrafast由Cerebras支持,可生成750 tokens/秒,比标准处理快14倍。11小时完成Humanity's Last Exam测试,速度近似Claude Fable 5的7倍,准确度相当。查看被引原帖 ↗
查看英文原文
the real question: how much more expensive is it?

like $500 per million tokens or what?
◔ 7.9 万 次浏览♥ 589⇄ 4▶ 含视频观点看原帖 ↗
Amjad Masad@amasad · 创始人 · 1 天前Amjad Masad,Replit 创始人兼 CEO

加一个编码工具,ARC-AGI-3 基本就能破解了。不出所料,编程确实能增强 LLM 的泛化能力。

引用 Jeremy Berman @jeremybermanI got 96.2% on ARC-AGI-3 with Opus 5, and 99.3% pass@2. The program is basically Claude Code + Opus 5 (high), one action command, and filesystem logs. Almost nothing ARC specific. github.com/jerber/arc-code查看被引原帖 ↗
查看英文原文
ARC-AGI-3 is nearly solved by merely adding a coding harness. As predicted, coding generalizes LLMs.
小互@xiaohu · 中文博主 · 1 天前小互,中文圈高频 AI 资讯站 Xiaohu.AI 主理人

MiniMax 开源 Music 3.0

一次生成五分钟完整歌,8GB 显存就能跑

给它歌词和一段风格描述,一次出一首五分钟的完整歌,前奏副歌桥段都在,32kHz 立体声

整个模型开源,单卡 8GB 显存就能跑起来

官方还开源了 1000 个模板和一个扩写技能

◔ 12 万 次浏览(3 条合计)♥ 808⇄ 119▶ 含视频新品看原帖 ↗
Aravind Srinivas@AravSrinivas · 创始人 · 1 天前Perplexity 联合创始人兼 CEO

Gemini Flash 模型很适合在任何 multi-model harness 中作为快速、成本高效的 subagents 使用。我们在 Perplexity Computer harness 中大量使用它们。

引用 Sundar Pichai @sundarpichaiGoogle推出Gemini 3.7 Flash,仅在3.6版发布三周后推出。软件工程、知识工作和Web开发领域性能大幅提升,价格相比3.6降低50%,已获开发者好评。查看被引原帖 ↗
查看英文原文
Gemini Flash models are great for fast, cost-efficient subagents inside any multi-model harness. We use them a lot inside the Perplexity Computer harness.
Lisan al Gaib@scaling01 · 博主 · 1 天前高频 AI 模型测评与爆料博主

Gemini 3.7 Flash 基准测试

引用 Logan Kilpatrick @OfficialLoganKGemini 3.7 Flash:速度快!价格比3.6 Flash低50%(至年底),3周内智能水平显著提升,在API、AI Studio、Antigravity等平台可用!查看被引原帖 ↗
查看英文原文
Gemini 3.7 Flash benchmarks

我反复讲,东北下岗导致崩盘的核心原因,是东北重工业比例太高。

东北重工业工人几乎必须依赖那条1950年投资300万、1990年关停的锅炉、轴承、变压器、车床生产线,自己只会其中一个模具或者工艺环节,一个环节吃30年。工厂没了,自己只有死路一条。

轻工业工人下岗出路多得多,比如中国第一家自行车 ——天津飞鸽自行车厂,当年同样大下岗大裁员,结果一群飞鸽工人从和平区搬到王庆坨,找原来的供应商买一堆零部件,自己建厂自己组装,全员飞鸽原班人马,直接让王庆坨成为中国自行车第一重镇。

90年代最爽的是各地电子厂,主要产品基本都是电视机、洗衣机、收音机、电话机、冰箱这些消费电子+白色家电,

这些电子厂几乎都能冒出来三五个技术骨干,带着厂子把产品做出来,疯狂砸钱打广告,只要产品做出来,就能一直有销路,厂长、骨干员工、国资就可以坐下来谈判,讨论改制和股权的问题,

基本上只要产品过硬,最后一定是皆大欢喜,现在这批电视、冰箱、洗衣机、白色家电国产头部品牌,一大半的前身都是当年的XX市电子厂、XX省无线电二厂、XX市电冰箱总厂。

90年代最成功的国资电子科技公司,就是中科院计算所100%持股的“中国科学院计算所新技术发展公司”,因为柳传志带着核心团队太成功了,中科院也有意愿坐下来改制,于是成了后来的联想。

现在一堆人骂柳传志侵吞国有资产,实际上对于这种国资厂,如果不谈判,柳传志完全可以立刻带着团队出来成立100%独资企业,不用中科院的平台,只能挣钱更多,原来的中科院计算机公司分分钟垮台。

对于90年代改制的这批公司,技术骨干和高管永远是最值钱的,自己单干的大有人在,当时这批转型成功的国资厂,几乎都是求着他们留下来,答应一切改制需求,最大限度地出让股份来换取国企存活,否则国资厂分分钟灰飞烟灭,1年内直接倒闭。

当然,最爽的还是餐饮公司、老牌饭店和食品工业这批人,90年代轮不到下岗,一群厨子自己跑出来干餐饮,一年都是几万十几万地挣钱。

在天津最有名的就是狗不理,狗不理效益一直不错,但是90年代一群厨子已经压制不住了,完全不想拿死工资,全都跑出来找门脸单干,

满大街的张记、四平、老永胜、老幼乐、姜记、二姑,一大半都是狗不理跑出来的,就狗不理那点工艺,学一周就全学会了,在2000年前后,这几家都是一年几万十几万地挣钱。

另一个原因是,东北以外的重工业比例都很低,在1990年消费市场爆炸增长的年代,下岗再就业难度不高,加上那时候都是公产房,家家户户没房贷,有口吃的就饿不死,只要愿意拉下脸,国企下岗工人找个谋生的手艺没那么难。

1994年大下岗再次告诉所有人三个道理:

1. 不要温水煮青蛙,要与时俱进,要跟着时代的潮流,计算机和电子相关的技术流行爆发,就要认认真真学习,不要指望一个手艺吃一辈子,更不要选择劝退专业;

2. 要选择轻资产的行业,个人转型简单得多,而不要选择重投资、重资产、长周期、大项目的产业,否则个人作用非常小,只能当一个螺丝钉,离开了整个产线和行业,个人作用几乎为零;

3. 要时刻和聪明人一起工作和学习,要时刻培养技术品味和产业视角,要时刻有危机感,任何无价值的工作和内耗,都会成为自己人生的慢性毒药,而任何有价值的吃苦和奋斗,最终都会成为人生的宝贵财富。

因为推特上的用户是真的claude code、codex、opencode、antigravity、cursor、zcode、kimi cli、trae这几个换着蹬,

某个设计有没有用,有没有效果,解决没解决痛点,是否真的work,可以哄一哄小红书用户,可以哄一哄b站用户,但是哄不了推特上这些用户。

引用 Yihui @yihui_indie还没体验 Deepseek Harness,但是有一点懵,为什么打开推特上都是在骂的,但打开国内的媒体都是在捧的?查看被引原帖 ↗
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

Anthropic 现在有了一个'自动完成复选框'功能,用于任务未完成但分配资源已耗尽的情况。这是在解决一个本不应该存在的问题。与其增加资源或解决根本问题,他们推出了自动继续功能。虽然这个功能有用,我会用,但它解决的根本是一个不应该这么严重的问题。

引用 ClaudeDevs @ClaudeDevsHit your usage limit in Claude Code desktop? There's now an auto-continue checkbox. Turn it on, and it'll automatically continue where you left off once your limit resets.查看被引原帖 ↗
查看英文原文
Anthropic now has an "auto-complete checkbox" for when a task isn't finished but the allocated resources have been used up.

This is a solution to a problem that shouldn't exist. Instead of increasing the resources or addressing the underlying issues, they've introduced auto-continue functionality.

I'll take the feature because it makes sense. But it solves a problem that shouldn't even be a problem to this extent.
Qwen@Alibaba_Qwen · 公司官方 · 23 小时前阿里通义千问大模型团队

一个GPU、100万上下文、Day-0即用。给vLLM团队无缝集成的工作点赞!👍 快来试试vLLM上的Qwen3.8-27B:@vllm_project recipes.vllm.ai/Qwen/Qwen3.8…

引用 vLLM @vllm_project🎉 Qwen3.8-27B is here from @Alibaba_Qwen , and the whole thing fits on a single GPU. Same hybrid backbone as the 2.4T flagship, dense instead of MoE. Day-0 support in vLLM. 🚀 What is in it for serving ✨ - Fits one Blackwell GPU in every precision. Qwen ships BF16 and FP8, the NVFP4 build from @inferact - 262K native context, stretching to 1M. At that length one GB300 still has room for roughly 6.6M KV tokens. Six full-length sequences in flight, on one GPU - An MTP draft head rides inside the checkpoint, so speculative decoding needs no separate speculator repo. Acceptance on short prompts measured 92.2% in BF16 and 84.8% in FP8 Verified end-to-end on @NVIDIA GB300: BF16 and FP8 at TP4, NVFP4 at TP1, tool calls working, correct generations at 1M context. Two prerequisites, vLLM nightly and transformers 5.8.0+. 🔗 recipes.vllm.ai/Qwen/Qwen3.8…查看被引原帖 ↗
查看英文原文
One GPU, 1M context, Day-0 ready. Big props to the vLLM team for the seamless integration!👍
Try Qwen3.8-27B on vLLM:
@vllm_project
recipes.vllm.ai/Qwen/Qwen3.8…
Lisan al Gaib@scaling01 · 博主 · 1 天前高频 AI 模型测评与爆料博主

所以……差不多是 Anthropic 年经常性收入的一半?💀

引用 Wall St Engine @wallstengineOpenAI年化收入已超400亿美元,约为2025年末运行率的两倍。7月运行率环比增长20%以上,主要由Codex、订阅和早期广告销售驱动。查看被引原帖 ↗
查看英文原文
so.. like half of Anthropic's ARR? 💀
Guillermo Rauch@rauchg · 创始人 · 1 天前Guillermo Rauch,Vercel 创始人兼 CEO

来 eve.dev 试试 🆓 GLM 5.2,最高 500TPS,@blackboxai 出品

引用 Vercel Developers @vercel_devGLM 5.2在AI Gateway上免费供eve agents使用至8月27日。支持高速(最高500 TPS),可在agent.ts中使用"zai/glm-5.2",为新eve agents默认模型。查看被引原帖 ↗
查看英文原文
Try
eve.dev
with 🆓 GLM 5.2 @ up to 500TPS from
@blackboxai
Qwen@Alibaba_Qwen · 公司官方 · 1 天前阿里通义千问大模型团队

Qwen Live现在上线了,开胃菜 👀

引用 QwenCloud @qwen_cloudReady for Qwen Live EP2?We will start at 10:00AM! UTC+8 x.com/i/broadcasts/1mGPaZanW… Qwen Cloud: qwencloud.com/?utm_content=g… #QwenCloud查看被引原帖 ↗
查看英文原文
Qwen Live now, an appetizer 👀

文科AI营销号:deepseek发布harness,一定是为了配合自家模型,自家agent配自家模型才是最厉害的,才能cache命中率更高,才能弯道超车打败claude!

真正看了source code和paper的人:PL和TCS魔怔极品中年嘉豪自娱自乐的产物,还用这一套屎上拼图把非cs出身的梁子给骗了。

梁子还是吃了非cs科班的亏。

引用 赵纯想 @chunxiangai来,作为一个 deepseek harness 喷子,我来告诉你它到底在捣鼓什么: 在 Claude Code、Codex 这类产品里,通常是: 固定 Agent 内核 + 可以添加的 Skills、Tools、MCP、Hooks 而 DeepSeek Harness : 极小的 Cordis 内核 + 模型供应插件 + 工具插件 + Agent Loop 插件 + 上下文插件 + 会话插件 + 存储插件 + 沙箱插件 + 权限插件 + 调度插件 + UI 插件 说人话:你在 CodeX 里面添加再多 skill,你也改不了 CodeX 读文件时候的内秉工作流(因为那是 Tibo 的研发同事在写软件时硬编码设定的)。 但是在 deepseek harness 体系下,你可以随时插拔它的核心插件,从而控制修改 loop 中的每一个具体检查点的运作方式。 既然你可以,那么 Agent 自身也可以抽插自己。deepseek 团队解决了两个核心问题,Agent 自己抽插自己的时候,怎么才能不把自己给抽或者插坏了。毕竟这玩意是个软件。 于是,他们提出一种“时间可组合性”的特性。让这些插件和传统软件中的插件不一样:它们在被添加的时刻,就铺好了自己的“退路”。保证在消失时干干净净,不会弄崩程序,减少心智负担。 来如未来。这叫时间倒流。 同时,插件还要声明自己依赖于什么服务。这样,当它依赖的服务被拔开、挪走的时候,它也不会像传统软件那样暴毙,报错。而是变为不可用,等待新的提供者,然后自动激活自愈。 小触角,小等待,小重连。四通八达小自愈。这叫空间。 然后他们认为自己发明了一种“时空软件学”。就给论文起了个关于时空的高大上的名字。 总而言之,他们把没有发明新的 harness,而是把乐高积木拆成更原子化的东西,再重新拼起来,发布了一下。发布的同时告诉你,看,这有一个 harness 哦,但是,它不一样,你可以任意拆下它最小的一部分,改造一下,再装回去。而且装配的过程中,它不会坏哦。一切都是热插拔的哦。 你们他妈的是不是脑子坏了,老子没事干非要把那一个个该死的螺丝拧下来再换个更炫酷的型号拧上去是吧?查看被引原帖 ↗

国内自媒体和各大社交网络平台已经严重视频和短视频化。

在微博娱乐化,知乎键政化之后,中国目前不存在一个绝对的、账号timeline为主导的专业文字平台。

核心原因是中国平均学历和教育水平还是太低,全社会人均阅读水平堪比一条边牧,所以文字平台在中国是干不过短视频的。

连这些AI运营都知道, 有用的信息阅读在推特,有用的分享发布在推特,自己最了解推特上同行们风格和内容,所以推特反而成了主力平台。

引用 Charles在路上 @Charles77xixi和朋友交流才发现一个问题,为什么各大模型厂商只在推特上注册官号宣发自己的产品,而国内自媒体基本不宣发呢?查看被引原帖 ↗
Lisan al Gaib@scaling01 · 博主 · 1 天前高频 AI 模型测评与爆料博主

OpenAI 7月份 ARR 增长了 20% 以上

引用 Stephanie Palazzolo @steph_palazzoloOpenAI用Wiz总裁Dali Rajic替换首席收入官Denise Dresser,Dresser仅任职8个月。Greg Brockman宣布7月总ARR增长20%以上,企业ARR增长32%。查看被引原帖 ↗
查看英文原文
OpenAI ARR grew 20%+ in July
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主
连环推 ×2

短短几天改变了一切:

- DeepSeek V4 Pro 缓存命中的输入价格将从 $0.003625 上升到峰值时段的 $0.044/百万 tokens,增幅 1,114%。

- Gemini 3.7 Flash 上线,价格比 Gemini 3.6 Flash 原价便宜 50%。Google 暂时也给 3.6 Flash 应用了同样的促销价。这个折扣原定在 2026 年底过期,不过我觉得 Google 可能会继续保持。

- 一天前,xAI 用 Grok 4.6 展现了真本事:接近前沿水准的智能,$2/百万 input tokens、$6/百万 output tokens,以及杰出的 SOTA 能力

DeepSeek 宣布的涨价让它变成了一笔很差的买卖。短时间内,Grok 已经成为一个重量级的前沿竞争对手。Gemini 3.7 在 3.6 发布不到一个月后推出了有意义的更新,同时把原价减半。但据我看,OpenAI 的 GPT-5.6 Luna 仍然是性价比之王。

目前唯一感觉落后的大厂是 Anthropic。它已经秘密申报了 IPO,传言目标是今年秋季上市,部分投资者预期上市估值在 2 万亿美元左右。不过其最新确认的私融估值是 9650 亿美元。

对我来说,Opus 一直很令人失望(已经吐槽过多次了,但每天看到它有多差我都很震惊),和我聊天的很多人都有这种感受。Sonnet 5 在对标最新竞品的中端性价比方案中看不出什么竞争力。

短短几天改变了一切。

查看英文原文
A few days changed the entire game:

- DeepSeek V4 Pro’s cache-hit input price will rise from $0.003625 to as much as $0.044 per million tokens during peak hours - a 1,114% increase.

- Gemini 3.7 Flash launched at 50% below Gemini 3.6 Flash’s original price. Google has temporarily extended the same promotional rate to 3.6 Flash. The discount officially expires at the end of 2026, although I suspect Google may keep it.

- One day earlier, xAI showed what it was capable of with Grok 4.6: near-frontier intelligence at $2 per million input tokens and $6 per million output tokens and fantastic SOTA frontier intelligence

DeepSeek’s announced price increase makes it a significantly worse deal. Within a remarkably short time, Grok has emerged as a serious frontier competitor. Gemini 3.7 delivered meaningful improvements just over three weeks after 3.6 while cutting the original price in half. Yet in my view, OpenAI’s GPT‑5.6 Luna remains the price-performance king.

The only major lab that currently feels off the pace is Anthropic. It has confidentially filed for an IPO reportedly targeted for this fall, and some investors reportedly expect a valuation of around $2 trillion at listing. Its latest confirmed private valuation, however, was $965 billion.

For me, Opus has been a major disappointment (already mentioned that several times, but each day im shocked again how bad it is), and that frustration is echoed by many of the people I speak with. Sonnet 5, meanwhile, doesn’t look like a compelling mid-tier price-performance competitor against the latest alternatives.

A few days changed the entire game.
Oh: and good to see that GPT-5.6 on cerebras is now rolling out. Excited for the 750t/s inference
宝玉@dotey · 中文博主 · 1 天前宝玉,中文圈 AI 翻译与科普大 V

软件自进化可能是个伪命题,只会带来更大的混乱。

插件要么是一次性用完就扔的,要么就得要设计、验证和维护的,不是现在模型能力可以“自进化”的。

OpenClaw 的一坨能“自进化”的 Skills 已经做了示范。

还是等模型自学习自进化更靠谱点。

引用 Jiayuan (JY) Zhang @jiayuan_jy有幸一个月前就被 @tianyi 拉进了仓库。当时 DSH 还是一个只实现了 core framework 的毛坯房。过去一个月,基本上每次 pull 代码,都是上千个 commits 的速度在涨。 说一下我对 DSH 的一些理解,不一定对: 1. 首先是怎么理解 DeepSeek Harness 这个东西。我觉得 DSH 既是一个可以直接运行的 Coding Agent,目前官方提供了 Web 和 headless 两种形式;同时它也是一套 Agent 开发框架。TUI 之类的其他交互方式,也可以通过外部 profile 和插件接进来。 2. 如果拿 Coding Agent 的标准来说,当前 DSH 的体验确实不如 Claude Code / Codex 那么完善。整个项目还很早期,接口一直在变化,插件生态也才刚刚开始,质量肯定是层次不齐的。 3. 但如果从开发框架的角度来看,可以把 DSH 想象成一个乐高汽车玩具。DeepSeek 官方提供的这个 Coding Agent,只是他们自己拼出来的一套官方预置。你完全可以把里面的零件换成自己喜欢的:换引擎、换轮胎、换挡风玻璃,或者加装其他模组。甚至最后拼出来的东西,也不一定还是一辆汽车。 4. DSH 的核心是「一切皆插件」。模型、工具、文件系统、Shell、沙箱、会话存储、Subagent、UI,甚至 Agent Loop 本身,都是插件。正因为这样,你可以把 DSH DIY 成任何自己想要的样子,这也给后面的社区生态留下了很大的空间。 5. 再往前想一步,这其实有一点「自进化软件」的雏形了。DSH 现在已经可以让 Agent 检查自己的 runtime,现场写一个插件并挂载上去,然后在后续的任务里直接使用这个刚刚获得的能力。 当然,现在这部分还比较实验性:动态生成的插件只存在于内存里,重启就没了,也还不能自动沉淀成一个永久插件。 但可以想象一下:假设某个功能现在没有,你和 Agent 随便聊两句,这个功能就被做好了,而且可以直接开始使用。甚至 Agent 在执行任务的时候,可以自己发现缺少某种能力,然后自己完成开发、安装和调用。 6. 接下来就需要等待一批真正优秀的插件了。DSH 现在还很早,但我相信它的潜力非常大。 --- 另外从代码上来看,DSH 有非常多函数式编程的影子,不熟悉 Ocaml/Haskell 可能一上来会比较难理解,可以多让 Agent ELI5 一下。查看被引原帖 ↗
karminski-牙医@karminski3 · 中文博主 · 1 天前karminski-牙医,中文圈模型评测博主

烧了1亿token了,正在验证deepseek-v4-pro-0813问题出在哪里,一会给大家带来评测+硬核解析视频。

宝玉@dotey · 中文博主 · 1 天前宝玉,中文圈 AI 翻译与科普大 V

我最近在 ~/.claude/CLAUDE.md 里面加了一段提示词,让它多开 SubAgent(Opus)去执行,这样我默认开 Fable 5 High,Token 消耗也不算厉害。

Fable 5 则主要做需求澄清、方案拆解、任务分发和结果验收。

之所以不用 Opus 5 是因为太太太慢了,而且 Token 消耗巨大!

---

注意你的主要任务是分析、编排和验证,具体任务尽可能交给 subagent(Opus)去执行。当主 agent 是 Fable 5 时尤其如此:自己只做需求澄清、方案拆解、任务分发和结果验收,实现类工作(读大量代码、写代码、跑测试、批量修改)一律用 Agent 工具派给 Opus subagent 执行。

Together AI@togethercompute · 公司官方 · 1 天前

10,000 @nvidia B300 GPU。印度最大的 AI 工厂。

Together AI 和 @larsentoubro 正在建设全国最大的 GPU 集群,支持开源推理、微调和大规模训练,为印度 AI 原生生态提供支持。

reuters.com/world/india/indi…

查看英文原文
10,000
@nvidia
B300 GPUs. India's largest AI Factory.

Together AI and
@larsentoubro
are building the country's biggest GPU cluster, backing open-source inference, fine-tuning, and training at scale for India's AI-native ecosystem.


reuters.com/world/india/indi…
Guillermo Rauch@rauchg · 创始人 · 1 天前Guillermo Rauch,Vercel 创始人兼 CEO

一个命令管理所有 token。我预测这将成为大规模使用 coding AI 的标配方式。稳定性、模型选择、成本更低、可观测性、ZDR…最酷的是这个命令兼容所有现有的 coding harness,比如 Claude Code 和 Codex!

引用 Vercel @vercel用单条命令将编码 agents 连接到 AI Gateway。• 自动配置 8 个流行编码 harnesses • 来自 30+ providers 的 300+ models,无加价 • 开源 models 支持 ZDR 和 US 推理查看被引原帖 ↗
查看英文原文
One command to rule them all (tokens).

I predict that this will become the default way of using coding AI at scale. Uptime, model choice, lower costs, observability, ZDR…

The cool thing is that this command configures every *existing* coding harness, like Claude Code & Codex!
◔ 4.3 万 次浏览♥ 227⇄ 16▶ 含视频观点看原帖 ↗
Matt Shumer@mattshumer_ · 博主 · 23 小时前HyperWrite CEO,AI 实战技巧分享

Matic大方地送了我一台机器人。我原来超级怀疑……去年买了个Roomba,讨厌得不行,一周内就退了。但这东西太牛了。我的公寓从来没这么干净过,舍友也都超开心。

引用 Matic Robots @maticrobotsIntroducing Cues: Voice & Gesture control for Matic We raised $115M and spent 9 years to make the world’s first intuitive home robot Say you spilled coffee: Point to the spill and say, "Hey Matic, clean this" and it will hear, see, locate in 3D, and go clean on its own. Matic comes with a lot of features: 1. Say "Hey Matic" - it locates your voice, turns, and looks at you 2. Say "Hey Matic, follow me" and start walking. Matic will follow behind 3. Say "Hey Matic, go clean the living room". Since it knows your house map, it navigates and just does it It's so easy, a 5 year old and an 80 year old can use it and it understands 75 different languages. Matic has 8x the airflow, specialised cleaning algorithms for rugs, corners, toekicks, mopping, etc and cleans better than any other robot vacuum. Also keeps improving with software updates. 13,000 families use and love Matic. WIRED magazine gave it a 10/10 (the only hardware to receive this rating in a decade) Buy yours at maticrobots.com and if you don't love it after 6 months, we'll give you a full refund. To celebrate our launch, we're cleaning 300 homes with Matic in San Francisco and New York City. Comment "Matic" below, we'll send you the link to sign up and come to your doorstep to clean your home with Matic.查看被引原帖 ↗
查看英文原文
Matic generously sent me a robot.

I was super skeptical… last year I bought a Roomba and hated it so much I returned it within a week.

But this thing is fucking awesome. My apartment has never been so clean, and my roommates are so happy.
◔ 4.1 万 次浏览♥ 139⇄ 1▶ 含视频演示看原帖 ↗
歸藏(guizang.ai)@op7418 · 中文博主 · 1 天前歸藏,中文圈 AI 工具与提示词博主

Twitter 完全开源了他们的推荐算法

用 Codex 分析了一下,感觉跟以前我们的认知还是有不少变化的。

总结了六条创作者应该做的事情:

1. 应该做值得转发的原创内容。

2. 尽量少用首贴写钩子,把重要内容放在第二条推串里这种发帖形式(这个和大家的认知不太一样,比较重要)。

3. 应该优先争取阅读、回复、引用和关注,点赞的权重其实没有那么高。

4. 拉开发帖间隔,不要频繁刷屏。

5. 深耕一个垂类,不要频繁更换账号类型。

6. 少做那些互动诱饵(比如互关、回复发送等),不要让别人讨厌你、对你点“不感兴趣”或举报。

swyx@swyx · 博主 · 1 天前知名 AI 播客 Latent Space 主理人
连环推 ×2

人类 I/O 很耗时,听了 @mattpocockuk 和 @trq212 的建议后,我改进了 /align-me,现在支持批量提问而不是一轮轮交互。思路跟 spec decoding 一样,通过提前看 2-10 步来加速

设计探索效果贼好!

引用 Theo - t3.gg @theoGotta say that Matt's "grill-me" skill is exceptional and helps a ton with getting agents aligned with my brain查看被引原帖 ↗
查看英文原文
human i/o is costly, so after listening to
@mattpocockuk
and
@trq212
i made an /align-me modification which allows for batches of questions instead of round-by-round. same intuition as spec decoding, you speed up by looking ahead 2-10 steps

works INCREDIBLY for design explorations
matt's latest:
youtube.com/watch?v=UNzCG3lw…


thariq's latest:
youtube.com/watch?v=9fubhllm…


bonus dont miss
@_philschmid
's banger
youtube.com/watch?v=0vphxNt4…


all 3 of these frontier skills guys contributing to the discussion is a beautiful thing to see
Gary Marcus@GaryMarcus · 博主 · 1 天前

哇!OpenAI最新研究显示,“员工使用AI的频率与他们的收入之间没有任何相关性。

最新科技热潮的ROI仍然是未知数。”

好好消化一下这句话吧。

引用 Emily Forlini @EmilyForliniIPO trouble? OpenAI's latest research finds no correlation between how often employees use AI and how much money they make. The ROI of the latest tech boom is still very much up in the air. fortune.com/2026/08/13/burie…查看被引原帖 ↗
查看英文原文
wow! new OpenAI study shows “no correlation between how often employees use AI and how much money they make.

The ROI of the latest tech boom is still very much up in the air.”

Let that sink in.
🚨 AI News | TestingCatalog@testingcatalog · 博主 · 1 天前专挖 AI 产品未发布新功能的爆料号

OpenAI 开始向 ChatGPT Business 的 Premium($100/月)席位发送候补邀请。

> 相比标准版多 5 倍的使用额度。
> 没有五小时使用限制,每周重置时间固定。
> Standard 和 Premium 席位可以在工作区内混合使用。

早期加入者还可能获得最多 $500 的额外额度(每添加一个 Premium 席位 $100)。

查看英文原文
OpenAI started sending waitlist invitations for Premium $100/month seats on ChatGPT Business.

> It comes with 5x more usage than Standard.
> No five-hour usage limit and predictable weekly resets.
> Standard and Premium seats can be mixed within the Workspace.

Early joiners may also qualify for up to $500 in extra credits ($100 per Premium seat added).
宝玉@dotey · 中文博主 · 1 天前宝玉,中文圈 AI 翻译与科普大 V

我不喜欢用 grill-me,可能因为我不擅长空想,也不喜欢被拷问。我需要那种看得见摸得着的东西才能进一步得出结论。

我更喜欢用 Claude Design 这种快速出一版原型,做出来一个真实的东西,才能感觉出来哪不对;或者 AI 写一份文档来理解方案,或者干脆 AI 实现一个 PoC 版本出来。

这在以前是很成本很高的事情,现在 AI 生成真的太容易了,所以对我来说不依赖 grill me 也还好。

引用 Kevin Ma @kevinma_dev_zh说真的, @mattpocockuk 的 grill-me,应该是我使用频率最高,同时也觉得最有用的 Skill 之一。 我在做产品和技术设计时,经常会用它来帮我检查,还有哪些地方没有真正想清楚。 很多时候,大的交互框架、产品流程和技术方向其实已经比较明确了,但往下落到细节,就会发现还有很多模糊的地方。有些是自己遗漏了,有些是之前压根没有想到,还有一些更常见的情况是,AI 已经给出了方案,甚至做出了 Mock UI,我看完却总觉得不太对。 麻烦就在这里。 我知道自己不满意,却又很难立刻说清楚,到底哪里有问题,以及我真正想要的是什么。 这时候我通常会先把目前的想法、目标和已有方案告诉 Agent,然后直接来一句:grill me。 接下来,Agent 就会开始不断盘问我。 它会沿着设计中的一个个分支继续往下问,把那些原本模糊的地方一点点挖出来:这个功能到底解决什么问题?这个状态应该怎么处理?用户为什么要在这里做这个动作?两种方案之间你真正看重的是什么? 很多问题,我其实从来没有认真想过。 而在一轮轮回答这些问题的过程中,原本只是脑子里一个模糊的感觉,会逐渐变成非常具体的产品决策。 这也是我觉得 grill-me 最有价值的地方。 有时候你缺的并不是 AI 再给你一个方案,而是有人不断追问,帮你把自己真正想要的东西想明白。查看被引原帖 ↗
Aravind Srinivas@AravSrinivas · 创始人 · 1 天前Perplexity 联合创始人兼 CEO

Perplexity Agent API是网页搜索和浏览agents的最佳方案

引用 Perplexity Developers @perplexitydevsSonar is moving to the Agent API. The Perplexity Agent API keeps grounded web search, and adds multi-step research, code execution, built-in tools, and access to multiple models through one API. On BrowseComp and WideSearch, Agent API more than doubles the best Sonar score.查看被引原帖 ↗
查看英文原文
Perplexity Agent API is the best for web search and browsing agents
yetone@yetone · 中文博主 · 1 天前开源 AI 编程插件 avante.nvim 作者,开发者圈博主

2026 年真的是神奇的一年,虽据称它是 Agent 元年的第二年,但你既能在网上看到 Google 的这种新设计,又能在电影院里看到最新院线片《牛来》。

引用 Google Design @GoogleDesignCelebrating small wins is hardwired into "Achieve," our goal-tracking sample app. We wanted the simple act of checking off a task to feel like a celebration. Instead of the standard #Material ripple, we used Styles to drop in a custom, high-contrast completion animation—specifically, a brand-colored ring that expands and fades on tap, paired with a subtle pressed-state gradient. It lets us focus entirely on polishing those custom, rewarding micro-interactions while Material handles the foundation. #GoogleDesign #MaterialDesign #JetpackCompose #AndroidDev #UXDesign #InteractionDesign查看被引原帖 ↗
◔ 3.7 万 次浏览♥ 121⇄ 3▶ 含视频动态看原帖 ↗
Together AI@togethercompute · 公司官方 · 1 天前

Qwen3.8-2.4T-A95B已在Together AI上线。

Qwen团队的最新开源MoE模型拥有2.4T总参数,95B活跃参数,100万context window,编码和agent能力也很强。

快来构建吧:together.ai/models/qwen3-8-m…

查看英文原文
Qwen3.8-2.4T-A95B is live on Together AI.

The Qwen team’s new open-weight MoE packs 2.4T total parameters with 95B active, a 1M context window, and strong coding & agent capabilities.

Start building:
together.ai/models/qwen3-8-m…
◔ 5.6 万 次浏览(2 条合计)♥ 34⇄ 5新品看原帖 ↗
Qwen@Alibaba_Qwen · 公司官方 · 23 小时前阿里通义千问大模型团队

Qwen 3.8-Max已在Modal上线,具有完整的1M上下文窗口和自定义DFlash推测器。很高兴看到我们的启动合作伙伴Modal从Day 0就全力以赴!⚡️ 2.4T参数。1M上下文。难以置信的是,仍然只需一个Modal调用。 @modal

引用 Modal @modalQwen3.8-2.4T-A95B by @Alibaba_Qwen and @alibaba_cloud is now available on Modal. Served with a custom DFlash speculator trained on tool-call-heavy data. Full 1M context window.查看被引原帖 ↗
查看英文原文
Qwen 3.8-Max is live on Modal, with the full 1M context window and a custom DFlash speculator under the hood.
Love seeing our launch partner Modal go all in from Day 0!⚡️
2.4T parameters. 1M context. Somehow, still just a Modal call away.
@modal
向阳乔木@vista8 · 中文博主 · 1 天前向阳乔木,中文圈 AI 工具与趋势博主

不知道现在有没有 token 银行或 Agent 接单平台?

比如把用不完的 token 帮别人跑任务赚积分或钱。

等自己 Token 不够用的时候,从平台里调用?

Dan Shipper 📧@danshipper · 博主 · 1 天前

平均每分钟接到 2 份应用:

Citadel
OpenAI
Red Bull
Microsoft
Profound
Alpha School
NASDAQ
Google
Pinnacle
Cohete
Bessemer
Lenovo
PwC
Square

快去申请!

引用 Dan Shipper 📧 @danshipperEvery举办Thesis年度AI会议,11月5日在纽约布鲁克林举行。汇集Notion、OpenAI、Anthropic等公司领导者,探讨自动化时代人类工作的意义。包括演讲、展示、工作坊等环节,探索前沿技术如何改变实际工作场景。查看被引原帖 ↗
查看英文原文
Averaging 2 applications per minute from:

Citadel
OpenAI
Red Bull
Microsoft
Profound
Alpha School
NASDAQ
Google
Pinnacle
Cohete
Bessemer
Lenovo
PwC
Square

You should apply!
◔ 3.4 万 次浏览♥ 100⇄ 2▶ 含视频动态看原帖 ↗
AIGCLINK@aigclink · 中文博主 · 1 天前
连环推 ×2

看很多人在吐槽Deepseek Harness面向极客之类的,绝壁是老登对时代的逃避,尤其是很多营销号非常喜欢反着来吸眼睛,你扔给workbuddy、codex给你跑起来不就行了,看了dsh的论文就知道有多牛逼了,这个架构是生态建设上的最优解的一种,它卡位的是类似于PI Agent这种原子级的harness底层,甚至定位是agent os这种。

未来围绕非code场景的agent底座很多产品或许会直接用这个来做agent os,就犹如今天很多code agent都是套壳claude code、codex之类的,只不过这次主角从编程变成了办公场景的os。

Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

OpenAI的年化营收正朝着400亿美元迈进,大约是2025年底水平的两倍,彭博社援引知情人士的消息报道。

不过,Claude Code的年化营收增速迅猛,正朝着预计1000亿美元的方向狂奔。

话虽如此,OpenAI现在正步步紧逼,稳扎稳打地追赶Anthropic的收入。Codex目前是最大的推动力,方向上实打实地起着正面作用。

查看英文原文
OpenAI is on track to clear $40 billion in annualized revenue, roughly double its run rate at the end of 2025, Bloomberg reports, citing people familiar with the figures.

However, Claude Code, with annualized revenue run rates scaling aggressively toward a projected $100 billion.

However, OpenAI is increasingly closing in and steadily approaching Anthropic's revenue. Codex is currently the biggest driver, in a positive sense.
yihong0618@yihong0618 · 中文博主 · 1 天前

看到个评论挺有趣的,pi 想做的是 vim, dsh 想做的是 emacs.

Qwen@Alibaba_Qwen · 公司官方 · 23 小时前阿里通义千问大模型团队

Qwen3.8-Max即将上线Nebius Token Factory(Day 0)。很高兴与Nebius携手启动,作为我们Day 0的启动合作伙伴,为更多用户提供专用推理。大模型,从一开始就是。🔥 @nebiustf

引用 Nebius Token Factory @nebiustfQwen3.8-2.4T-A95B is going open weight: 2.4T parameters, with 95B active, and Nebius Token Factory is joining as a Day 0 launch partner. Qwen3.8-2.4T-A95B is coming to Nebius Token Factory Managed Services for dedicated inference. @Alibaba_Qwen @alibaba_cloud查看被引原帖 ↗
查看英文原文
Qwen3.8-Max is coming to Nebius Token Factory on Day 0.
Great to kick things off together with Nebius as our Day 0 launch partner, bringing dedicated inference to more users.
Big model, right from the start.🔥
@nebiustf
Aravind Srinivas@AravSrinivas · 创始人 · 1 天前Perplexity 联合创始人兼 CEO

这个 700b 参数模型的数据真亮眼!

引用 Z.ai @Zai_orgIntroducing GLM-5.3: Built to Code. Ready for Cyber Defense. - Top-tier coding and agentic capabilities, achieved through post-training on the 743B base model - A major leap in cybersecurity, setting a new standard among open models Tech Blog: z.ai/blog/glm-5.3查看被引原帖 ↗
查看英文原文
Impressive numbers for a 700b parameter model!
Ethan Mollick@emollick · 创始人 · 1 天前沃顿商学院教授,AI 应用研究权威

从 AI 对企业业绩的提升作用来看,已经出现了一些早期迹象——那些原本表现就不错的早期 AI 采纳公司可能开始甩开其他竞争者。

OpenAI 的数据表明,一些公司的 AI 使用频率明显更高,而这些公司往往聚集了生产力最强的员工。

查看英文原文
To the extent that AI use boosts firm performance, some early signs here that early AI adopting firms that were already doing well may start to outpace others.

Data from OpenAI shows some firms are using AI much more, and they tend to be firms with the most productive employees.
Tanishq Mathew Abraham, Ph.D.@iScienceLuvr · 博主 · 1 天前

Meta AI研究员、DINOv2的作者讲解现代自监督学习!

还介绍了新方法CAPI

@TimDarcet上个月在@MedARC_AI期刊俱乐部做了一场很棒的演讲。

讲得超清晰,信息特别密集,我从他的演讲学到了很多。

强烈建议看看!

查看英文原文
Meta AI researcher who created DINOv2 explains modern self-supervised learning!

Also introduces a new approach called CAPI


@TimDarcet
gave a brilliant talk about his research at the
@MedARC_AI
journal club last month.

Very clear and information-dense, I learned a lot from his talk.

Give it a watch!
◔ 2.8 万 次浏览♥ 485⇄ 31▶ 含视频研究看原帖 ↗
ollama@ollama · 公司官方 · 1 天前本地跑大模型的热门工具


@AIatMeta
的 Muse Glimmer 全本地处理个人月度信用卡账单。

你的数据属于你自己!

用你喜欢的应用 / 工具链搭配 Ollama,试试各种 agent 任务。

查看英文原文
Using
@AIatMeta
's Muse Glimmer all locally to process personal monthly credit card statements.

Your data belongs to you!

Try different agent tasks using your favorite apps / harnesses with Ollama.
◔ 3.2 万 次浏览(2 条合计)♥ 343⇄ 27▶ 含视频演示看原帖 ↗
宝玉@dotey · 中文博主 · 1 天前宝玉,中文圈 AI 翻译与科普大 V

DeepSeek 作为一个模型厂商,最有价值的肯定还是做 Agent Harness Product 而不是 SDK,因为 SDK 它不容易拿到用户行为数据。

做 Agent Harness Product 那就得追求用户量,追求用户量就要先做好用户体验,其次才是插件可定制化。

引用 马天翼 @fkysly目前根据已有的 DeepSeek Harness 的内容,我觉得 Claude Code/Codex 就像苹果一样,什么都做,追求体验极致;而 DeepSeek Harness 可能是一种安卓的理念,就是走开源,社区自己 DIY 玩,OEM(比如一些企业定制、政府定制等等)可以私有化魔改等等。而这个基础上,最核心的就是插件生态。查看被引原帖 ↗
Perplexity@perplexitydevs · 公司官方 · 1 天前AI 搜索引擎 Perplexity 官方
连环推 ×3

我们在6月推出了 Search as Code (SaC),在广泛和深度研究上取得了最先进的性能。

本周,我们又推出了 SaC 的一系列优化,不仅进一步提升了性能,每任务成本还降低了近10%。

查看英文原文
We released Search as Code (SaC) in June, achieving state-of-the-art performance on wide and deep research.

This week, we’re rolling out optimizations to SaC that further improve performance while cutting cost per task by nearly 10%.
How an SDK is shaped determines how well models can use it.

We’ve rolled out two batches of Search SDK updates within Computer, raising execution reliability from 81.9% to 92.6%.

More reliable action execution enables more capable orchestration.
These improvements show up across a wide range of real-world user workflows.

In Computer, SaC optimizations delivered higher user satisfaction at an 8% lower per-task cost.
François Chollet@fchollet · 创始人 · 1 天前

常规提醒——公开的 ARC 3 游戏集合叫"展示集",不是"评估集"也不是"训练集"。它不是用来做训练数据的,也不是用来做评估的。在公开展示集上的成绩并不能反映在真正基准测试上的表现。

展示集的目的就是展示格式和提升人们的参与度。

私有评估集难度和新颖度都要高得多。

Kaggle 排行榜今天的最高分是 2.70%(这是半私有集——竞赛结束时,提交会在完全私有集上重新评分)。

查看英文原文
Regular reminder -- the set of public ARC 3 games is called "demonstration set", not "eval set" nor "training set". It is not meant to be used as training data, and it is not meant to be used as an eval. Scores on the public demonstration set are not indicative of scores on the actual benchmark.

The demonstration set is intended to demonstrate the format and drive human engagement.

The private eval set is substantially more difficult and more novel.

The top score on the Kaggle leaderboard today is 2.70% (that's the semi-private set -- at the end of the competition, the submissions are scored on the fully private set).
Qwen@Alibaba_Qwen · 公司官方 · 23 小时前阿里通义千问大模型团队

Qwen3.8-Max 已在 Fireworks 上线,为智能体、重度编码和长上下文任务做好准备。
Fireworks 作为 Day 0 首发合作伙伴全程力挺我们。
这开局方式太带感了!
今天是 Day 0,放烟花庆祝吧!🎉🎆
@FireworksAI_HQ

引用 Fireworks @FireworksAI_HQQwen3.8-2.4T-A95B is now live on Fireworks with Day-0 support! This 2.4T parameter MoE model is built for autonomous agents, heavy coding, large context windows, and is ideal for coding and agentic performance. Start building today: fireworks.ai/models/firework…查看被引原帖 ↗
查看英文原文
Qwen3.8-Max is live on Fireworks, ready for agents, heavy coding, and long-context work.
Fireworks is right there with us as a Day 0 launch partner.
What a way to kick things off!It's Day 0, cue the Fireworks! 🎉🎆
@FireworksAI_HQ
Gary Marcus@GaryMarcus · 博主 · 1 天前

神经符号世界模型——完全是我2020年论证过的方向——终于赢了。

引用 François Chollet @fcholletJeremy's excellent work here is a great illustration of a very powerful type of approach: LLM-guided on-the-fly synthesis of a symbolic world model, i.e. making sense of the world by writing executable code that encodes your understanding of the causal mechanics of the world. So far, all of the top-performing harnesses on ARC-AGI-3 use this style of approach. Which is also the approach we recommended when we initially released the benchmark (Jeremy would know this better than most, as a former Ndea member of technical staff). I'm happy that ARC 3 has incentivized more research and more progress in this area.查看被引原帖 ↗
查看英文原文
neurosymbolic world models – *exactly* what I argued for in 2020 in the Next Decade in AI – for the win.
Ethan Mollick@emollick · 创始人 · 23 小时前沃顿商学院教授,AI 应用研究权威

人们基于当前价格、采用和能力的不稳定性,对AI在企业中的影响和正确使用方式做出过于乐观的推断。现在是确保为未来构建灵活性的好时机。

查看英文原文
People are making way too confident extrapolations about the impact & proper ways to use AI in companies based on the current, very unstable nature of price, adoption & capabilities of today’s systems.

It is a good time to ensure that you are building flexibility for the future.
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主
连环推 ×5

1/ 想看看是否真的能从一张平面 2D 图像得到可用的 3D 资产,而不需要花几小时修复网格。一张平面 2D 图像在五分钟内被转成了清晰、高细节的 3D 模型。一切都在浏览器内完全由 @Hitem3D 处理,所以零设置。下面是最终资产和具体工作流的快速展示:

查看英文原文
1/ Wanted to see if you could actually get a usable 3D asset out of a flat 2D image without spending hours fixing the mesh.

A flat 2D image turned it into a crisp, high-detail 3D model in about 5 minutes flat. Everything is handled entirely inside the browser using
@Hitem3D
so there’s zero setup. Here’s a quick look at the final asset and the exact workflow:
2/ Their split feature is where it handles the real "last mile" headache of making things physical.

It features smart splitting and automatic connector generation, cutting your model into perfect printable pieces with interlocking joints so it's completely optimized and ready to drop right into a slicer. Check this out:
3/ I skipped the setting adjustments and just compiled a quick showcase of the final 1536 Pro resolution models. Seeing how cleanly the platform handles complex shapes and different asset styles without requiring hours of manual rebuilding is exactly what makes an all-in-one 3D maker platform worth looking into.
4/ What usually ruins AI-to-3D tools is getting a messy, unusable point cloud back. This engine actually generates a clean mesh with usable geometry right out of the gate.

The best part is their free trial download advantage, unlike other platforms that lock your files behind a paywall, you can actually test it and download your models to check the quality yourself.
5/ If you want to try the system out, they're running a 50% off deal on their monthly plan.
Plus, the first 200 new users who sign up through my tracking link below will get an extra 500 bonus credits to play with.
Check it out here:
hi3d.ai/
#Hi3D
#Hitem3D
#AI3D
#Imageto3D
#3Dgenerator
@Hitem3D
◔ 2.3 万 次浏览♥ 144⇄ 9▶ 含视频演示看原帖 ↗
Linus ✦ Ekenstam@LinusEkenstam · 博主 · 1 天前

这是笔价值数十亿美元的生意。

Matic刚刚发明了一台每个家庭最终都会入手一台的家用机器人。

就像Roomba一样,它在你家到处跑,但不会盲目乱撞。相反,它真的能看见,你指着一处乱摊子说声“清理这里”,它就直接去那个精确的位置搞定。

简单至极,强大至极。

引用 Matic Robots @maticrobotsIntroducing Cues: Voice & Gesture control for Matic We raised $115M and spent 9 years to make the world’s first intuitive home robot Say you spilled coffee: Point to the spill and say, "Hey Matic, clean this" and it will hear, see, locate in 3D, and go clean on its own. Matic comes with a lot of features: 1. Say "Hey Matic" - it locates your voice, turns, and looks at you 2. Say "Hey Matic, follow me" and start walking. Matic will follow behind 3. Say "Hey Matic, go clean the living room". Since it knows your house map, it navigates and just does it It's so easy, a 5 year old and an 80 year old can use it and it understands 75 different languages. Matic has 8x the airflow, specialised cleaning algorithms for rugs, corners, toekicks, mopping, etc and cleans better than any other robot vacuum. Also keeps improving with software updates. 13,000 families use and love Matic. WIRED magazine gave it a 10/10 (the only hardware to receive this rating in a decade) Buy yours at maticrobots.com and if you don't love it after 6 months, we'll give you a full refund. To celebrate our launch, we're cleaning 300 homes with Matic in San Francisco and New York City. Comment "Matic" below, we'll send you the link to sign up and come to your doorstep to clean your home with Matic.查看被引原帖 ↗
查看英文原文
This is a multi-billion dollar business.

Matic just invented a home robot that every single household is going to end up buying.

just like a Roomba, it drives around your house, but it's not bumping around blind. instead, it actually sees, you point at a mess, say "clean here," then it goes to the exact spot and cleans it.

so simple, so powerful.
Rowan Cheung@rowancheung · 博主 · 23 小时前AI 日报 The Rundown 创始人

我在招7个人打造AI媒体和教育的未来。The Rundown现在有300多万活跃读者,扩张速度远超我们的处理能力。推荐成功者可得$2000奖金。

开放职位和链接如下:

> 内容负责人:掌控三个通讯、社媒和新节目形式的编辑标准
> 媒体合作主管:开发并达成为我们融资的品牌合作
> AI教育者:教百万级用户如何真正用上AI
> 长视频编辑:协助制作与Sam Altman、Mark Zuckerberg、Satya Nadella等嘉宾的访谈
> 广告创意编辑:制作吸引下一百万AI爱好者的视频
> 指南编辑:将我们最好的视频指南打磨成用户会保存使用的内容
> 缩略图设计师:为我们增长中的节目清单设计缩略图

所有职位共同要求:

> 对AI的深度好奇心
> 偏向行动而非观望
> 能识别优秀作品(和AI垃圾)的品味

所有职位全远程。通过评论中的链接申请。

查看英文原文
I'm HIRING 7 more people to help me build the future of AI media and education.

The Rundown now has over 3M active readers, and our expansion plans far outpace what we can handle.

$2,000 referral bonus if you help us find someone we hire full-time.

Open roles and link to apply below:

> Head of Content: own the editorial bar across three newsletters, socials, and new show formats
> Media Partnerships Lead: prospect, pitch, and close brand partnerships that fund everything we do
> AI Educator: teach millions of people how to actually put AI to work
> Long Form Video Editor: help us scale interviews with A-list guests like Sam Altman, Mark Zuckerberg, and Satya Nadella
> Ad Creative Video Editor: ship the videos that hook the next million AI-curious readers
> Guide Editor: edit our best video guides into content people save and actually use
> Thumbnail Designer: design the thumbnails for our growing list of shows

What we're looking for among all roles:

> Deep AI curiosity
> Unreasonably biased toward action
> Taste sharp enough to spot great work (and AI slop)

All roles are fully remote. Apply with the link in the comments.
Lisan al Gaib@scaling01 · 博主 · 1 天前高频 AI 模型测评与爆料博主
连环推 ×3

关于这短期内为什么是烂主意的一些想法。你需要很多新的机制来促进和监管token市场:

这主要会给AI实验室创造巨大的意外收益,帮助大公司并激励高度动态的agent工作流

它也会造成地区性和整体的下层阶级

我想到的一些问题:
- 永久的下层阶级,因为token价格现在直接取决于谁能付最高价
- 地区性下层阶级,因为有些地区算力较弱、能源成本更高
- 便宜的夜间token价格会增加需求,但也会压平需求曲线。这一切导致AI实验室硬件利用率更高(更赚钱),我们可能最后会陷入利用率始终接近100%、token价格24/7狂飙的世界
- AI实验室的意外收益:接近95%计算利用率这样的右尾。价格会绝对暴涨,完全脱离实际成本(AI实验室利润会从~80%+变成更离谱的数字)

- 大公司可以通过以固定价格预留大量容量来对冲token成本,但这会增加市场其他地方的波动,小创业公司只能付更多钱还要承受高波动

- 模型发布这样的尾部事件会导致价格暴涨,直接把大多数人赶出市场,内线信息变得极有价值

- 市场操纵和内线信息有巨大风险:
agent群体或大公司可以用大量token做经济上有价值的工作推高token价格,但同时又能买token衍生品赌这个事件,再次有利于市场最大的玩家

- 对竞争对手基础设施的攻击会被激励,因为你能从需求飙升中获得巨大意外收益

引用 roon @tszzl主要AI公司应承诺提供实时定价的API产品。AI需求在日夜周期内波动很大,业界容量极度紧张。agents使得处理可变定价、批处理和预测总成本变得容易。查看被引原帖 ↗
查看英文原文
a few thoughts on why this is a terrible idea in the short-term. you would need a lot of new structures to facilitate and regulate token-markets:

it would mostly create massive windfall profits for AI labs, help large corporations and incentivize very dynamic agentic workflows

it would also create regional underclasses and an overall underclass

some problems I can think of:
- permanent underclass, because token prices are now directly coupled to who can pay the most
- regional underclasses because some regions do have less compute capacity and more expensive energy production
- cheaper overnight token prices would increase demand, but also flatten the demand curve. all of this results in higher utilization of AI lab hardware (making them more profitable) and we could end up in a world where utilization is like close to 100% all the time, with insane token prices 24/7
- windfall profits for AI labs: near the right tail above for example 95% compute utilization. prices would go absolutely parabolic, and would be completely detached from the actual cost of producing said tokens (AI lab margins would go from ~80%+ to something even more ridiculous)

- large corporations could hedge their token costs by reserving large amounts of capacity at a fixed price, but this would increase volatility for the rest of the market, while smaller startups would have to pay more and are exposed to high volatility

- tail events like model releases would result in massive price spikes that would outprice most people, and insider information gets extremely valuable

- there's a huge risk of market manipulation and again insider information:
agent-swarms or large companies could use massive amounts of tokens to do even economically valuable work, which would drive token prices up, but at the same time they could buy token derivatives betting on that event, again favoring the largest players in the market

- attacks on competitors infrastructure would be encouraged, since you would make gigantic windfall profits from the demand spikes
i guess some of these are preventable by designing the derivatives in a smart way
also I haven't really thought about how this would interact with foreign/chinese token markets
Tibor Blaho@btibor91 · 博主 · 1 天前逆向挖掘 AI 产品代码的爆料专家

冷知识:ChatGPT 网页版现在内置了广告拦截检测,用八个隐藏的测试元素来识别哪些广告被拦截了,这样 OpenAI 就能测量和调试广告投放的效果。

查看英文原文
Fun fact: ChatGPT web app now also has built-in ad-blocker detection, using eight hidden test elements to identify which ad components are blocked so OpenAI can measure and debug the impact on ad delivery
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

不错:RedNote 刚开源了 dots3-note Preview,一个为长任务代理设计的 280b 多模态 MoE。

它的独特之处是 TEMPO:一种新的强化学习方法,让代理能评估自己的进展、更新记忆,并在处理长期、陌生任务时学习。

这包括 512K 上下文,以及文本、图像、视频和音频理解,同时只有 16B 的活跃参数。

开源真是太香了

引用 dots studio @dotsstudioaiIntroducing dots3-note preview — a small but mighty step toward long-horizon agency in real life. 🔹 280B MoE with 16B active parameters, a 512K context window, and multimodal understanding across text, vision, and audio 🔹 Introduces TEMPO, a new RL approach for long-horizon agent training through self-critiquing and test-time-scaled value estimation 🔹 Built to reason, explore unfamiliar environments, update memory over time, and combine multimodal perception with coding and tool use to solve complex tasks 🔹 Open weights on Hugging Face, alongside two open benchmarks for real-life agents: VibeSearchBench and VibeLifeBench Competitive with much larger models across reasoning, agentic, and multimodal evaluations. 🔗 Tech blog: studio.dots.ai/dots/dots3-en… 🔗 Model weights: huggingface.co/dots-studio/d… 🔗 Github: github.com/studio-dots-ai/do…查看被引原帖 ↗
查看英文原文
Nice: RedNote just open-sourced dots3-note Preview, a 280b multimodal MoE built for agents that operate over hours.

Its standout idea is TEMPO: a new reinforcement-learning approach that lets an agent critique its own progress, update memory, and learn while navigating long, unfamiliar tasks.

This includes 512K context as well as text, image, video and audio understanding with only 16B of simultaneously active parameters.

Open source keeps on giving
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

太牛了:Matic 把 Tesla 在自动驾驶上的视觉优先战略搬到了家用机器人上。5 个 RGB-IR 摄像头,没有 LiDAR,NVIDIA Jetson Orin 芯片做本地推理。它能建立光学逼真的 3D 地图并在清扫时持续更新,所以能分辨充电线、袜子和狗,对不同的东西采取不同的处理方式。

机器人吸尘器里的 LiDAR 就是一根旋转激光。深色地毯会吸收激光信号,传感器就判断成悬崖,机器拒绝跨过你自己家的地板。

摄像头加真正的计算能力改变了整个工作流。先判断东西是什么,再决定怎么清扫。

这个感知系统用了 7 年才研发成熟,由加州一支不到 100 人的团队完成。
@maticrobots

引用 Matic Robots @maticrobotsMatic 家用机器人融资 1.15 亿美元,支持语音和手势控制。可清洁、跟随、导航,支持 75 种语言。具 8 倍气流和专业清洁算法,13000 家庭在用,WIRED 杂志评 10/10。查看被引原帖 ↗
查看英文原文
This is so cool: Matic put the same vision-first bet Tesla made for self driving into a robot that cleans your home.

Five RGB-IR cameras, no LiDAR, and an NVIDIA Jetson Orin running the inference onboard. It holds a photorealistic 3D map of the place and keeps updating it while it cleans, so it knows a charging cable from a sock from the dog and handles each one differently.

The LiDAR inside a robot vacuum is a single spinning laser. A dark rug swallows the beam, the sensor reports a cliff, and the machine refuses to cross your own floor.

Cameras plus real compute change the order of operations. It works out what a thing is first, then decides how to clean around it.

Seven years went into the perception stack before a single unit shipped, built by a team of under 100 people in California.
@maticrobots
◔ 3 万 次浏览(2 条合计)♥ 141⇄ 4▶ 含视频演示看原帖 ↗
Qwen@Alibaba_Qwen · 公司官方 · 23 小时前阿里通义千问大模型团队

Qwen3.8-Max已在DigitalOcean Serverless Inference上线。与DigitalOcean携手推出,作为我们Day 0的启动合作伙伴。大模型。一切顺利。现已在DigitalOcean上。🌊🏄‍♀️ @digitalocean

引用 DigitalOcean @digitaloceanNow available: @Alibaba_Qwen 3.8-2.4T-A95B from @alibaba_cloud on DigitalOcean Serverless Inference via NVIDIA HGX™ B300 GPUs. 🤖 1M context, built for long-horizon coding. 🔗 do.co/45VBAyZ One API, usage-based pricing, no infra to manage.查看被引原帖 ↗
查看英文原文
Qwen3.8-Max is live on DigitalOcean Serverless Inference. Launching side by side with DigitalOcean as our Day 0 launch partner.
Big model. Smooth sailing. Now on DigitalOcean.🌊🏄‍♀️
@digitalocean
Lisan al Gaib@scaling01 · 博主 · 1 天前高频 AI 模型测评与爆料博主
连环推 ×2

向我朋友 Elon 喊话了,他竟然在听我说

这可能是追上 OpenAI 和 Anthropic 的唯一办法

引用 Cursor @cursor_aiCursor is now part of @SpaceX . Today, we have officially closed our acquisition. We will join the @SpaceXAI team to help make Grok the world's most useful AI and improve Grok Build, Grok Bot, Grok API, Cursor, and more. SpaceX has built some of the most inspiring and impressive technology in the world, and we’re grateful for the opportunity to become part of such a special company. Onwards.查看被引原帖 ↗
查看英文原文
shout-out to my homie Elon listening to me

this was probably the only way to have a chance of catching up to OpenAI and Anthropic
okay maybe there's a fourth option

ZAI has a much larger model internally from which they distill
Replit ⠕@Replit · 公司官方 · 1 天前AI 编程平台 Replit 官方

如果大家都在做一样的事,我们就做下一件。来听听 Replit Design 团队讲讲,为什么他们要为 AI 时代重新定义设计流程。设计的新时代,为了所有人。

replit.com/design

查看英文原文
If everyone's building the same thing, we build what's next.

Hear from the team that built Replit Design, and learn why we set out to reinvent the design process for the age of AI.

The next era of design, for everyone.


replit.com/design
◔ 1.9 万 次浏览♥ 132⇄ 14▶ 含视频新品看原帖 ↗

本站由 Jedee杰哥 打造 · 公众号「Jedee杰哥」每早送 AI 日报

姊妹站:𝕏 简中账号数据榜单 · X 关注 @jedeeai · RSS 订阅 · AI 日报 · 历史归档