JEDEE AI
存档 2026-08-12

8 月 12 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
Tibo@thsottiaux · 公司官方 · 1 天前ChatGPT 产品官方账号

我之前说过每 100 万新增活跃用户就 reset Codex,直到 1000 万。结果我们远远突破了这个目标,从 1000 万之后就没声音了。明天给你个小惊喜。

查看英文原文
I previously promised a reset for every 1M in additional active users for Codex, until 10M.

We blew past that and have been silent since 10M. Little surprise for you tomorrow.
Sundar Pichai@sundarpichai · 创始人 · 1 天前谷歌 CEO

每月超过 10 亿人使用 @Geminiapp 来迸发新想法、完成工作。这是我们增长最快的产品,也是达到 10 亿用户里程碑的第 14 个产品。
感谢 @JoshWoodward 和整个 Gemini 团队,也感谢所有与我们一同踏上这段旅程的人 - 更多精彩还在后头!

查看英文原文
1B+ people are now using
@Geminiapp
every month to spark new ideas and get things done. It’s our fastest growing product ever, and our 14th to hit the 1B-user mark.

Kudos to
@JoshWoodward
& the entire Gemini team, and thank you to everyone who has been on this journey with us - much more to come!
◔ 408.2 万 次浏览(4 条合计)♥ 7,616⇄ 648▶ 含视频动态看原帖 ↗
OpenAI@OpenAI · 公司官方 · 1 天前ChatGPT 开发商官方账号

现已推出预览版:ChatGPT Linux 桌面应用。在支持的 Linux 系统上,在你工作和开发的地方使用 ChatGPT、ChatGPT Work 和 Codex,配合你的项目和浏览器工作流。

查看英文原文
Now in preview: The ChatGPT desktop app for Linux.

Use ChatGPT, ChatGPT Work, and Codex where you already work and build, with your projects and browser workflows on supported Linux systems.
Tibo@thsottiaux · 公司官方 · 1 天前ChatGPT 产品官方账号
连环推 ×2

你为啥切换到 Codex 了?喜欢它哪些地方,咱们还能咋改进?别说重置啊。

查看英文原文
Why did you switch to Codex?

What do you like about it and what could we improve? Don’t say reset.
Also don’t say Linux, we just shipped that.
Grok@grok · 公司官方 · 1 天前马斯克 xAI 旗下聊天机器人 Grok 官方

Grok Bot 最初是个内部原型,已迅速改变了我们的工作方式。今天我们公开发布测试版。

引用 Grok Bot @botIntroducing Grok Bot, now in early beta. Bots are AI teammates that do real work for you. They sign in to your tools, use them just like you do, and come back with finished work.查看被引原帖 ↗
查看英文原文
Grok Bot started as an internal prototype and has rapidly changed the way we work.

Today, we're releasing a beta to the public.
◔ 212.7 万 次浏览(13 条合计)♥ 3,850⇄ 294▶ 含视频新品看原帖 ↗
Tibo@thsottiaux · 公司官方 · 1 天前ChatGPT 产品官方账号

终于成了...Codex & ChatGPT 桌面版现在支持 Linux 了。

感谢大家的耐心等待,要是你等不及已经下单 MacBook 就把订单取了吧。真的就这么好用。

👀

引用 OpenAI @OpenAINow in preview: The ChatGPT desktop app for Linux. Use ChatGPT, ChatGPT Work, and Codex where you already work and build, with your projects and browser workflows on supported Linux systems.查看被引原帖 ↗
查看英文原文
We did it, finally... Codex & ChatGPT desktop, now on Linux.

Thanks for waiting and you can cancel that MacBook order if you got impatient. It’s that good.

👀
◔ 119.9 万 次浏览(5 条合计)♥ 1.2 万⇄ 580新品看原帖 ↗
Guillermo Rauch@rauchg · 创始人 · 1 天前Guillermo Rauch,Vercel 创始人兼 CEO

我们本来说好要飞行汽车的,结果等来的却是无限超级智能给所有人。

查看英文原文
We were promised flying cars and all we got is infinite superintelligence for everyone
OpenAI Developers@OpenAIDevs · 公司官方 · 1 天前OpenAI 开发者平台官方
连环推 ×3

现在您可以将来自其他代理的工作与 ChatGPT Work 和 Codex 保持同步。导入项目、聊天、技能和插件,查看导入历史记录,并在设置中选择自动更新。现已在 ChatGPT 桌面应用中可用。

查看英文原文
You can now keep your work from other agents in sync with ChatGPT Work and Codex.

Import projects, chats, skills, and plugins, review your import history, and opt in to automatic updates in Settings.

Now available in the ChatGPT desktop app.
More on bringing your existing workflow into Codex:
Here's how you can set up automatic syncs:


learn.chatgpt.com/docs/impor…
◔ 89.2 万 次浏览♥ 3,893⇄ 297▶ 含视频新品看原帖 ↗
el.cine@EHuanglu · 博主 · 1 天前

完了..现在没人能看出这是AI生成的了

引用 el.cine @EHuangluAI video is now undetectable查看被引原帖 ↗
查看英文原文
its over.. no one can tell this is AI now
◔ 60.7 万 次浏览♥ 267⇄ 16▶ 含视频观点看原帖 ↗
Tibo@thsottiaux · 公司官方 · 1 天前ChatGPT 产品官方账号

导入你的世界。Codex。运行。

引用 OpenAI Developers @OpenAIDevsYou can now keep your work from other agents in sync with ChatGPT Work and Codex. Import projects, chats, skills, and plugins, review your import history, and opt in to automatic updates in Settings. Now available in the ChatGPT desktop app.查看被引原帖 ↗
查看英文原文
Import your world. Codex. Run.
◔ 38 万 次浏览♥ 2,832⇄ 84▶ 含视频演示看原帖 ↗
@levelsio@levelsio · 博主 · 1 天前独立开发者标杆,AI 产品连续创业者

一样

Claude 现在越来越爱说教了,天天直接拒绝我

要是 Grok 写代码够给力的话,我很乐意直接跳过去

我的网站后端已经完全跑在
@xAI
上了

引用 Alex @ HispanicNomad | Freedom Based Lifestyle @hispanicnomad用户考虑取消Claude订阅。认为最近模型过度谨慎,频繁拒绝创建和编写请求内容,非Fable模型基本不可用。正考虑转向Manus或中文模型。查看被引原帖 ↗
查看英文原文
Same

Claude is getting extremely preachy every day flat out refusing me regularly

I'm pretty happy to jump ship to Grok if it gets good enough for coding

My sites already run fully on
@xAI
back end
Lex Fridman@lexfridman · 博主 · 1 天前

这是我与 Khabib Nurmagomedov (@TeamKhabib) 的对话,他是有史以来最伟大的格斗手之一。

与他训练、进行这次深入的交谈,以及在过去几个月了解他,都是真正的荣幸。

我们完全用俄语进行了这次对话,翻译配音成英文。YouTube 上提供两种语言的音轨和字幕。在 X 上,我分别发布了英文和俄文版本的视频。翻译和配音工作量巨大(这对我很重要),由人类和 AI 协作完成。感谢 @ElevenLabs 团队的帮助。

祝 Islam Makhachev、Khabib 及其团队周末 UFC 330 的比赛好运!

如我在视频中所说,我真的很感谢你们所有人,感谢你们这些年的支持。爱你们 ❤️

时间戳:
0:00 - 介绍
1:35 - 为什么达吉斯坦产生如此多伟大的格斗手
11:41 - 达吉斯坦历史战士文化
21:02 - 在达吉斯坦山区的艰苦训练
24:53 - 名利权的陷阱
35:34 - Khabib 统治力的秘诀
46:47 - 纪律与努力
59:15 - 达吉斯坦街头搏斗
1:12:56 - Khabib 的压制力
1:18:40 - Khabib vs Conor McGregor
1:31:25 - Khabib 最艰难的比赛
1:39:12 - 有史以来最伟大的 MMA 格斗手
1:56:57 - 执教
2:02:30 - Daniel Cormier
2:06:48 - Islam Makhachev
2:16:46 - Dana White
2:20:16 - Joe Rogan
2:22:31 - Ali Abdelaziz
2:25:08 - 足球
2:32:23 - Messi vs Ronaldo
2:50:23 - Dubai
2:54:14 - 信仰
3:01:57 - 对年轻人的建议

查看英文原文
Here's my conversation with Khabib Nurmagomedov (
@TeamKhabib
), one of the greatest fighters of all time.

It was truly an honor to get to train with him, to have this deep conversation, and to get to know him over the past few months.

We did this conversation entirely in Russian. It's translated and dubbed into English. Both language audio tracks and subtitles are available on YouTube. Here on X, I'm posting the English & Russian language versions as seperate videos. The translation & dubbing was a huge amount of work (which I deeply care about) with humans & AI collaborating. Thanks to the amazing team at
@ElevenLabs
for their help.

Good luck to Islam Makhachev, Khabib & team this weekend at UFC 330!

And as I've said in the video, I'm truly grateful for all of you, for your support all these years. I love you all ❤️

Timestamps:
0:00 - Introduction
1:35 - Why Dagestan produces so many great fighters
11:41 - History of Dagestan warrior culture
21:02 - Hard training in the mountains of Dagestan
24:53 - The trap of fame, money, and power
35:34 - The key to Khabib's dominance
46:47 - Discipline and hard work
59:15 - Street fights in Dagestan
1:12:56 - Khabib's top pressure
1:18:40 - Khabib vs Conor McGregor
1:31:25 - Khabib's toughest fight
1:39:12 - Greatest MMA fighters of all time
1:56:57 - Coaching
2:02:30 - Daniel Cormier
2:06:48 - Islam Makhachev
2:16:46 - Dana White
2:20:16 - Joe Rogan
2:22:31 - Ali Abdelaziz
2:25:08 - Football
2:32:23 - Messi vs Ronaldo
2:50:23 - Dubai
2:54:14 - God
3:01:57 - Advice for young people
◔ 32.1 万 次浏览♥ 2,992⇄ 208▶ 含视频动态看原帖 ↗
Sundar Pichai@sundarpichai · 创始人 · 1 天前谷歌 CEO

新的Pixel 11系列发布,专为Gemini Intelligence设计,拥有众多实用新功能。

我特别兴奋的几个:Rambler,一个为自然说话方式优化的新语音输入。Magic Capture,轻松拍照。还有HiLight,当手机屏幕朝下时会微妙发光,让你知道生活中重要的人在给你打电话。

引用 Rick Osterloh @rosterlohThe best of Gemini is right at the heart of the new Pixel 11 phones and the new Pixel Watch 5. We call it Gemini Intelligence. It’s no longer just an operating system that helps you use your phone, it’s an intelligence system that gets things done for you.查看被引原帖 ↗
查看英文原文
The new Pixel 11 line up is here, designed for Gemini Intelligence and with lots of new helpful features.

A few I’m excited about: Rambler, a new voice input built for how you actually talk. Magic Capture for effortless photos. And HiLight, a subtle glow when your phone is face down that lets you know when important people in your life are trying to call you.
Grok@grok · 公司官方 · 1 天前马斯克 xAI 旗下聊天机器人 Grok 官方

Imagine Image 2.0 现已推出到我们的 API:精确编辑、信息图、广告、游戏素材、UI/UX 原型、故事板等。

在我们的 Playground 试试看:
console.x.ai/team/default/im…

查看英文原文
Imagine Image 2.0 now available on our API: precise editing, infographics, ads, game assets, UI/UX mockups, storyboards, and more.

Try it on our playground:
console.x.ai/team/default/im…
◔ 29.3 万 次浏览(2 条合计)♥ 754⇄ 87▶ 含视频新品看原帖 ↗
Google Gemini@GeminiApp · 公司官方 · 23 小时前谷歌 Gemini 产品官方
连环推 ×16

现在你可以在 Gemini 里接入更多应用来完成更多事。在设置里把你喜欢的服务连接到 Gemini,就能在一个地方轻松规划、创建和完成你的待办事项。这是 14 个新增的 Gemini 集成应用。🧵

查看英文原文
You can now connect to even more apps in Gemini to get more done.

By linking your favorite services to Gemini in your settings, you can easily plan, create, and tackle your to-do list all in one place.

Here are 14 new apps to use with Gemini. 🧵
Plan your next dinner date by booking restaurant reservations with
@opentable
.

Just ask Gemini to "Book an outdoor restaurant for two people tomorrow night using OpenTable.”
Find events and tickets with
@Ticketmaster
.

Just ask Gemini to “Find live jazz tickets for this Friday in New York on Ticketmaster.”
Build your custom website with the help of
@wix
.

Just ask Gemini to “Create a website about my photography business on Wix.”
Play, discover, and enjoy your favorite music and podcasts with
@iHeartRadio
.

Just ask Gemini to “Stream top interview podcasts on iHeartRadio.”
Discover and book experiences with
@fever_gl
.

Just ask Gemini to “Find immersive art exhibitions in Miami this weekend on Fever.”
Listen to your favorite songs and playlists for any mood with
@pandoramusic
.

Try asking Gemini to “Play my high-energy climbing playlist on Pandora.”
Find, explore and book tours and other activities with
@GetYourGuide
.

Just ask Gemini to “Find a sunset canyon hike on Get Your Guide.”
Find local contractors and handyworkers with
@angi_home
.

Just ask Gemini to “Find local lawn treatment services on Angi.”
Manage your meetings and get transcription help with
@otter_ai
.

Just ask Gemini to “Summarize client wedding vision and vendor planning calls in
Otter.ai
.”
Rent a car in Brazil with
@localizabr
.

Just ask Gemini to “Find car rentals in Rio, Brazil with Localiza.”
Compare and hire local service professionals with
@thumbtack
.

Just ask Gemini to “Find top-rated local painters on Thumbtack.”
Schedule a doctor appointment with
@zocdoc
.

Just ask Gemini, “I need some new glasses. Find an appointment on Zocdoc within walking distance from work.”
Keep track of your meeting notes with
@meetgranola
.

Just ask Gemini to “Summarize the glaze and clay preferences from today’s workshop in Granola.”
Optimize your workflows with
@Zoho
.

Just tell Gemini to “Show my active deals from Zoho Bigin.”
This new slate of MCP connections is coming to Gemini over the next few weeks.

To learn more about how you can connect your favorite services directly to Gemini to help you plan trips, manage projects, and tackle your daily to-do list, visit our blog:
goo.gle/3TTbF8s
Google DeepMind@GoogleDeepMind · 公司官方 · 1 天前谷歌旗下 AI 研究机构,Gemini 背后团队
连环推 ×4

SL2T 是我们的突破性手语转文字模型,为 @Android 上的聋人和听力障碍用户推出新功能。

首先在 Pixel 11 上推出美国手语转英文功能,用户可以直接用手语输入 Gboard 和 Live Transcribe,无需打字。

查看英文原文
SL2T is our breakthrough sign language-to-text model powering new features for Deaf and hard of hearing users on
@Android
.

Starting with American Sign Language-to-English on Pixel 11, people can sign directly into Gboard and Live Transcribe instead of typing.
For years, building AI for sign language has been slow due to complex technical challenges and misconceptions about how the languages work.

We addressed this by training a model to translate simultaneous movements of the hands, body, and face directly into text, using huge amounts of data.
SL2T is state-of-the-art on academic benchmarks, plus it's optimized for real-world use like one-handed signing while holding a phone.

To protect privacy, it tracks body poses on-device, while our servers translate them into text.
We built SL2T with the Deaf community - guided by Deaf Googlers and our AI Sign Language Advisory Committee.

Bringing ASL input to phones is just the beginning and we’re working to expand this technology to more sign languages and applications.

Find out more →
goo.gle/3RM3kmx
◔ 22.1 万 次浏览(2 条合计)♥ 1,560⇄ 212▶ 含视频新品看原帖 ↗
Chubby♨️@kimmonismus · 博主 · 23 小时前Chubby,高频 AI 新闻聚合博主

DeepSeek v4 Pro 正式版发布了!Grok 4.6 API 也被曝光了!

太兴奋了!看到最近 DeepSeek v4 Flash 那波疯狂的更新,我预期这个也会有好几次迭代!

看起来今天是个大发布日啊!

引用 leo 🐾 @synthwavedd🚨 DeepSeek V4 Pro (GA) is now rolling out on the DeepSeek API and on DeepSeek Chat查看被引原帖 ↗
查看英文原文
DeepSeek v4 Pro GA is rolling out! Grok 4.6 API spotted as well!

Really excited! After the most recent insane update to DeepSeek v4 Flash i expect sevel updates on this one as well!

big release day it seems!
◔ 37.5 万 次浏览(3 条合计)♥ 477⇄ 20新品看原帖 ↗
Google Gemini@GeminiApp · 公司官方 · 1 天前谷歌 Gemini 产品官方

最后机会!在今晚 PT 时间 11:59pm (8/11) 前可以免费用 Gemini Omni 创建最多 10 个视频。

社区的一些灵感可以帮你开始 👇

查看英文原文
Last chance! Reminder that you can create up to 10 videos with Gemini Omni at no cost through 11:59pm PT tonight (8/11).

Some inspiration from the community to get you started 👇
Kling AI@Kling_ai · 公司官方 · 1 天前快手旗下可灵 AI 视频官方

有个地方没人知道它存在。由 mr.cobra_ai 创建。

查看英文原文
There’s a place no one knows exists.
Created by mr.cobra_ai
◔ 13.4 万 次浏览♥ 144⇄ 8▶ 含视频其他看原帖 ↗
🚨 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 🔥:美国大学生现在可以免费领一年的 ChatGPT Plus。

OpenAI 正在筹备"返校"活动,把 ChatGPT 定位为考试复习、学习等的必备工具!

> 从课程学习到期末项目,ChatGPT 帮你向前迈进。

> 这项优惠面向 220+ 所美国参与校园的在校学生。领取免费一年并验证学生身份,看看你的学校是否在列表里。

> 通过我们的安全验证合作伙伴 SheerID 领取免费一年并完成学生身份验证。验证后,你可以查看优惠详情并选择是否激活。

查看英文原文
OPENAI 🔥: College students in the US will be able to claim a year of ChatGPT Plus for free.

OpenAI is preparing a new "back-to-school" campaign, positioning ChatGPT as a tool for exam preparation, studying, and more!

> From study sessions to final projects, ChatGPT helps you move forward.

> This offer is available to currently enrolled students at 220+ participating U.S. campuses. Claim your free year and verify your student status to see if your school is included.

> Claim your free year and verify your student status with our secure verification partner, SheerID. After verification, you can review the offer and choose whether to activate it.
◔ 13.3 万 次浏览♥ 1,132⇄ 47▶ 含视频动态看原帖 ↗
Chubby♨️@kimmonismus · 博主 · 23 小时前Chubby,高频 AI 新闻聚合博主

来了:DeepSeek 4 GA 的基准测试已经开始流传。
升级还是不错的,可惜拿来和 Opus 4.8 对比而不是 Opus 5。
结果普遍不错,与开源 SOTA(Kimi k3 和 GLM-5.2)势均力敌,接近 5.6 Sol 和 Fable 5 的水平。

引用 Chubby♨️ @kimmonismusDeepSeek v4 Pro GA is rolling out! Grok 4.6 API spotted as well! Really excited! After the most recent insane update to DeepSeek v4 Flash i expect sevel updates on this one as well! big release day it seems!查看被引原帖 ↗
查看英文原文
Here we go: DeepSeek 4 GA Benchmarks are circulating already.

A very solid upgrade, but it's a bit of a shame they're comparing it to Opus 4.8 instead of Opus 5.

The results are consistently good, putting it on par with open-source SOTA (Kimi k3 and GLM-5.2) and approaching the performance of 5.6 Sol and Fable 5.
◔ 22 万 次浏览(2 条合计)♥ 976⇄ 56研究看原帖 ↗
Chubby♨️@kimmonismus · 博主 · 23 小时前Chubby,高频 AI 新闻聚合博主

Grok 4.6 发布了,疯狂的进步!xAI 确实在出绝活!

根据 xAI 的基准,Grok 4.6 在 AA Intelligence Index 上与 GPT-5.6 Sol 并肩(都是 61),在 CursorBench、FrontierCode 和 AA-Briefcase 上还领先。

但在 DeepSWE 和 Terminal-Bench 上还是落后于 GPT-5.6 Sol。

xAI 说这个模型进行了更长时间的补充训练、重新生成了 SFT 轨迹,以及在代码、网页开发、CAD 和内核优化方面的代理强化学习。公司还表示在更长的任务中看到了更多的'自我测试和验证'。

Grok 4.6 今天就可以通过 Cursor、Grok Build、API 和多个基础设施合作伙伴使用。

这些是公司报告的数据,但相比 Grok 4.5,在几乎所有公开评估中都有显著提升。


@xai

@cursor_ai
点赞!看着真的没得说!!

引用 Chubby♨️ @kimmonismusDeepSeek v4 Pro GA is rolling out! Grok 4.6 API spotted as well! Really excited! After the most recent insane update to DeepSeek v4 Flash i expect sevel updates on this one as well! big release day it seems!查看被引原帖 ↗
查看英文原文
Grok 4.6 released, absolutely insane crazy jump! xAI was cooking!

According to xAI’s benchmarks, Grok 4.6 matches GPT-5.6 Sol on the AA Intelligence Index at 61 and leads it on CursorBench, FrontierCode and AA-Briefcase.

But it still trails GPT-5.6 Sol on DeepSWE and Terminal-Bench.

xAI says the model received a longer supplemental training run, regenerated SFT trajectories and agentic reinforcement learning across coding, web development, CAD and kernel optimization. The company also reports seeing more "self-testing and verification" during longer tasks.

Grok 4.6 is available today through Cursor, Grok Build, the API and several infrastructure partners.

These are company-reported results, but the improvement over Grok 4.5 is substantial across nearly every published evaluation.

Kudos,
@xai
and
@cursor_ai
This looks so good!!
◔ 40 万 次浏览(10 条合计)♥ 866⇄ 55新品看原帖 ↗
宝玉@dotey · 中文博主 · 1 天前宝玉,中文圈 AI 翻译与科普大 V

我是能不用worktree就不用worktree,要么不同任务用branch分支,要么就直接同一个分支多任务,一般不太用担心冲突,首先只要任务类型不一样冲突可能性较小。

另外现在 Harness 比如说 Claude Code、Codex 都很聪明,遇到冲突会等另一个执行完。

使用 worktree 主要是在做耗时较长的 PoC 场景,只是验证一下,同时要做其他任务,完了就删除目录。

引用 宝玉 @doteyworktree 会创建很多目录,导致磁盘空间占用比较厉害查看被引原帖 ↗
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

2025年AI预测博客'AI 2027'的预测几乎全部应验了。24条预测中有19条已经成为现实,其他大部分也都走在准确的轨道上。

他们预测到现在和明年会发生:

到2026年底,越来越强大、越来越便宜的AI代理开始替代初级软件工程师,重塑整个经济。

到2027年,超人类编码代理会自动化AI研究本身,引发智能爆炸,可能在年底前导致人工超级智能出现——以及严重的对齐危机。

在我看来,我们完全符合这个时间表。疯了。

查看英文原文
Almost all the predictions from the 2025 prediction blog "AI 2027" have come true.

19 out of 24 predictions have materialized, and we are well on track for the majority of them to prove accurate.

They predicted for this and next Frage:

By late 2026, increasingly capable and affordable AI agents begin replacing junior software engineers and reshaping the economy.

In 2027, superhuman coding agents automate AI research itself, triggering an intelligence explosion that could produce artificial superintelligence- and a severe alignment crisis - by year’s end.

Sounds to me we are very much aligned with that timeline. Crazy.
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

笑死,中国老哥可能知道50种越狱这些模型的办法吧

引用 Can Bölük @_can1357guys you do know you can just disable thinking, and instead give it a "deep_think" tool, and it will call it with internal CoT reasoning format right? gl fixing that查看被引原帖 ↗
查看英文原文
LMAO

china bros probably know of 50 different ways to jailbreak the models
everyone wants more efficient reasoning

so it's a bit hard to avoid neuralese
clem 🤗@ClementDelangue · 创始人 · 23 小时前HuggingFace 联合创始人兼 CEO

走起!


huggingface.co/Qwen/Qwen3.8-…

查看英文原文
Let's go!


huggingface.co/Qwen/Qwen3.8-…
◔ 13.4 万 次浏览(3 条合计)♥ 1,585⇄ 110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

我们对 Vercel 构建系统背后的数据库进行了迁移,这是我们最关键的服务之一,当时还在每分钟处理 6000 次部署。

迁移怎么做的、我们的想法,以及学到的东西都在这里 ↓

vercel.com/blog/how-we-migra…

查看英文原文
We migrated the database behind every build at Vercel, one of our most critical services, while it handled 6,000 deployments a minute.

The story of how it happened, our philosophy, and what we learned ↓

vercel.com/blog/how-we-migra…
Google AI@GoogleAI · 公司官方 · 1 天前谷歌 AI 官方账号

明天是日全食 🌑☀️ 无论你是在格陵兰、冰岛和西班牙的全食区,还是在线观看部分地区,我们都给你准备好了。

用 @GoogleAIStudio 开发的交互式 3D 可视化工具可以实时追踪月影扫过全球的过程,还能从雷克雅未克或瓦伦西亚这些观测站模拟当地的天空景象,快进或快退来看关键的日食时刻:goo.gle/4g4baQt

查看英文原文
Tomorrow is the total solar eclipse 🌑☀️ Whether you’re in the path of totality across Greenland, Iceland, and Spain, or watching the partial phases online, we’ve got you covered.

The interactive 3D visualizer built in
@GoogleAIStudio
lets you follow the Moon's shadow across the globe in real-time, simulate local sky views from stations like Reykjavík or Valencia, and fast-forward or rewind to catch key eclipse milestones:
goo.gle/4g4baQt
◔ 9.5 万 次浏览♥ 621⇄ 85▶ 含视频演示看原帖 ↗

中国科技行业下一轮最大的泡沫破裂,应该就是具身智能+VLA+世界模型,会有几个公司上市后股价跌到零,剩下一批公司应该跟AI四小龙时代的一批调参大师一样突然关停或者暴雷。

虽然具身智能+VLA+世界模型的泡沫破裂,远远比不上2000年的美国dot com bubble互联网泡沫危机,但肯定比AI四小龙要猛烈得多。

方式也跟AI四小龙时代一样,一批大厂senior level到VP分别下场做demo,手搓个垃圾pick and place跑不通的灵巧手,或者全员跳舞,或者弄一堆metaverse元宇宙级别的物理世界仿真4399小游戏,烧完就关门。

这一轮会比以前任何一轮都要惨烈,核心原因就是宇树科技IPO全面开绿灯,马上要冲击上市,这一两年国内PEVC投资人已经全部杀红了眼,随便一个垃圾phd、垃圾教授、垃圾demo都能投进去几个亿,等着一个个也IPO开绿灯,让A股股民接盘。

表面上是科技大泡沫、技术大骗局,实际上本质还是金融监管的巨大漏洞,让宇树科技抢跑上市,导致后患无穷。

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

现在这事儿简直疯了;本来是个段子,现在竟然真的成了生产力工具。

Codex的里程碑这块儿,达到1000万用户的时候之前都是周周重置。

现在老哥给我们留了个大招:估计又是一次重置或者'囤积'重置。

OpenAI在社区运营上投入巨大。感谢啊!

引用 Tibo @thsottiauxI previously promised a reset for every 1M in additional active users for Codex, until 10M. We blew past that and have been silent since 10M. Little surprise for you tomorrow.查看被引原帖 ↗
查看英文原文
It’s just crazy at this point; what started as a running gag is turning into a productivity boost.

Regarding the milestones reached with Codex - up to the 10-million-user mark - there were weekly resets.

Now he has a surprise in store for us: presumably another reset or a "banked" reset.

OpenAI is investing significantly in community engagement. Thanks for that!
Lisan al Gaib@scaling01 · 博主 · 1 天前高频 AI 模型测评与爆料博主

他又做到了,EpochAI 的又一个开放问题被解决了

引用 levent @__alpoge__+++-+-+-+-+--+--+++-++----++--+----+-----+++---++-+---+++++++-+--+-+-+--++-+--+--+----+-+-+-+-++-++---+--++++--++-++++-+++++--+---++-+---+++++++++--+-+-+--++-+--+--+-----+++-+++-++-+-++++-++-+--+--+------++--+-++-+-+---+--+-+-+-+--++---+++++-----+++++++---+---+--+-+--+-+--+-++-++-++++++-+-++-++-+-+---+--+-+-+-+--++---+++++-----+++----+-+-++---+-++----+---+++++--+---+--++-+++++++-+--+---+----++-+++----++---+++---++++-+-+--+++-+--++++-+++-----++-+++-+++-++++++++-+--+---+----++-+++----++---+++---+++---+-+--++----+--+-+++-+--++------+++--++++++-+--++--+-++---+--++---+---++--+-+----+++-+-++--++++-++-+---+-++--++++++---++++++++-+--++--+-+++--+--++---+---++--+-+-+--+++--+---++++-+-++-+---+-+-+-----+--++-+----+-+-+----+-----++-+++++-+++--+--++++++-++--++---++-+++----+-+--+--++-+-+-+++++-++--+-+++-+-+-+----+-----++-+-+++-+++--+--++++++-++-+++---++++---+++++-++--+--+++-+++--+--+-++-+++--+++-+++--+-+--++++-----+-+-+--+--+---+-++---+++----+++-----+--+++++---+---++-++-+--+-++--+++-+++--+-+--++++-----+-+-+--+--+---+-+++-------+-+-++--++++---+--+-++-+--+---+-++-+-+++++-++--+----++++----+++---++--++-+---+-+--+++++++-+-+--++----+++-++-+--+-++-+++-+--+++-++++-++--+----++++----+++---++--++-+---+-+-------------+-+++-+++---++--++--+-+++--+-+-++---+----++++--+-+-++++-++---+--+----++-++-++++++++++++++-+---+---+++--++--++-+---++-+-+-+-+-+----++++--+-+-++++-++---+--+----++-++-++----++-++++--+++++--+--+--+--+++-----++--+-+-+++-+++-+++-+-+++-+++-+++++-++++---+++---+--+--++++--+++++--+---+++-+--+-----+-+++++++------+++-+--++-+++--++--+++-+-+++-++-+++-+----++-------+--+-+-++-+-+--+--+-+--+-+-+-++-+-+---+++++-+-++-++----++--++++-+--++-+---+-++--+-+----+------+--+-+-+--+---+++++-+++-++++---+-++-++--+-++-+---+++-+-+++-+-+--+++-+-----++-+++--+--++++-++-+-++++--+--++---+++++---+---++++-++----+--+-+-++++++-+++-++++--+---+--++--+-+-++-+++-+++---++---++----++------+--+--++++-+++-++++--+++++-+-+-++++-+-+-++--+++++-----+++-+++++-+---+--+++++-+---++------++-+++-+++-++---++-+--+--+-++++--+-+++-+++-+---++++查看被引原帖 ↗
查看英文原文
he did it again

another one of EpochAI's open problems is gone
Matt Shumer@mattshumer_ · 博主 · 1 天前HyperWrite CEO,AI 实战技巧分享
连环推 ×2

这个可控视频生成的方法真有意思,而且很简单:

- 用代码先把视频的“骨架”搭出来,把所有的动作、物体等以低精度嵌入进去
- 再把生成的“草稿”丢给视频模型去“变真实”

引用 GMI Cloud @gmi_cloudyou don't need Seedance 2.5, here's cheaper and more efficient workflow we made this entire game shooting scene in one shot for $1.97 with DeepSeek V4 Flash 0731 + MiniMax H3 we use DeepSeek to code the raw Three.js scene first: exact camera movement, character motion, timing, everything. keep tweaking it there for basically pennies. then feed the final footage into H3 and let it cook the realism 48 mins, $1.97 total, one H3 generation way cheaper than rerolling a video model 10 times查看被引原帖 ↗
查看英文原文
Such an interesting and simple approach to controlled video generation:

- use code to create the “skeleton” of the video, baking in all actions, objects, etc. at low fidelity
- pass the resulting “draft” through a video model to “make it real”
This might be a useful short-term approach while we wait for the bitter lesson to work its magic on long-form video generation.
◔ 8.3 万 次浏览♥ 510⇄ 19▶ 含视频教程看原帖 ↗

这个道理我讲了将近10年了,绝大多数人根本不听我的任何警告,油盐不进。

所以我现在彻底反过来了,我支持任何985以上的robotics机器人嘉豪,一定要到美国EECS top 10大学扎扎实实读个robotics PhD,扎扎实实做五年机器人,或者去个头部工业机器人干三年,

否则他们永远不知道我到底在跟他们讲什么。

引用 枫原万葉 @46ge5国内具身智能/机器人卷得要死,加班非常严重,大把人月休1天2天(包括智元等头部企业 性质类似于早期的自动驾驶,因为算法垃圾,导致无止境的落地调优,case 分析修补,还有硬件链路问题,听着都累(我也不认为当前模式能跑通查看被引原帖 ↗
NVIDIA@nvidia · 公司官方 · 1 天前
连环推 ×2

我们荣幸地宣布Glassdoor将我们的创始人兼CEO @JensonHuang 列为2026年最佳CEO榜单第一名。🎉

这是一项非常有意义的认可,基于使NVIDIA成为今日之NVIDIA的人们的反馈。

查看英文原文
We’re honored that Glassdoor has named our founder and CEO,
@JensenHuang
, No. 1 on its 2026 Best CEOs list. 🎉

A very meaningful recognition, grounded in feedback from the people who make NVIDIA what it is.
Read the story:
nvda.ws/4gt2Ta0
@levelsio@levelsio · 博主 · 1 天前独立开发者标杆,AI 产品连续创业者

Skyr 属于最少加工的食品。

我建议用 NOVA 量表,这是判断你的食物是否干净的好办法,本质上就是判断该不该吃。拍张照片,问 AI 你的食物是 NOVA 哪个等级就行。

我现在基本不吃 NOVA 3 和 4 组的食物了(偶尔作弊日除外😏)

有意思的是,几十年前还是 NOVA 1(未加工)的食物,现在不少都变成 NOVA 4(超加工)了。原因就是通货膨胀。

未加工食物便宜又乏味,不好营销,没什么利润。一块牛排就是一块牛排,哪儿买都差不多价钱。但要是你把它包装成「烤肉大师甜烟熏美国安格斯牛肉」,配上精美的标签、广告宣传,加上一堆盐水、色素、添加剂来增加重量,就能卖更高的价钱了。

卖超加工食品利润高得多,比卖未加工食品强太多了(那东西乏味还没利润)。

感谢资本主义啊!

我父母在荷兰吃过的某种肉,原来是未加工的,现在卖的是完全超加工的——防腐剂、乳化剂、色素、添加剂什么都有。他们压根没注意到,因为看起来还差不多。但这东西让他们不舒服。

还有酸奶。我长大的时候酸奶就是酸奶,发酵牛奶,NOVA 1(未加工)。现代酸奶不是这样了。我从前常买的高端品牌丹尼斯的 OIKOS 酸奶,原来就两种成分:牛奶和发酵菌。现在成分是一些牛奶、点奶油、大量来路不明的奶粉,还是用发酵菌来省钱,结果价格反而涨了!这还是原味的呢。调味酸奶更别说了,全是人工甜味剂、乳化剂、各种添加剂。

所以你真的没法再用眼睛判断食物是不是干净了。大部分食物都被改成了加工或超加工的,为了省钱或者为了营销卖出更高的价格。

拿不准的话,试试 NOVA 量表吧。

引用 Big Jim @_jmwebb“Not processed like X. But made through a process 😊查看被引原帖 ↗
查看英文原文
Skyr is minimally processed

I recommend using the NOVA scale, it's a great way to figure out if your food is clean or not, and essentially if you should be eating or not

Just take a pic and ask AI what NOVA scale your food is

I generally flat out avoid eating from groups 3 and 4 nowadays (except cheat days once in a while ;))

What's interesting is that a lot of the food that used to be NOVA 1 (unprocessed) a few decades ago has now become NOVA 4 (ultra processed) due to inflation

The reason is that unprocessed food is cheap and boring and hard to market, so you can't make much profit on it. A steak is a steak and it costs roughly the same everywhere. But if you sell it as

Grill Masters Sweet & Smoky BBQ Angus Beef

You can add a beautifully designed label, market it in ads, and all the stuff you add it (salt water, coloring, additives) to increase the weight, so you can charge much more

Selling ultra processed food is a much better business to be in than selling unprocessed food (boring and no profit margin)

Thank capitalism for this!

An example is a type of meat my parents used to eat in the Netherlands, it was unprocessed then, and is now sold as ultra processed full of preservatives, emulsifiers, coloring and additives, and my parents have no idea because it looks roughly the same. It makes them sick though

Also yogurt, I grew up and it was just you know, yogurt, fermented milk, classified as NOVA 1 unprocessed. But modern yogurt isn't that anymore. A premium brand I used to buy called OIKOS by Danone, used to have 2 ingredients: milk and fermentation bacteria. Now it's some milk, a bit of cream, a lot of milk powder (with dubious origins) and fermentation bacteria to save money but the price went up! And that's the plain one. The flavored yogurts are full of artificial sweeteners, emulsifiers, additives etc.

So you can't really trust your eyes anymore to judge if food is clean or not, because most of it has been turned into processed or ultra processed to save money and/or market to you to sell for a higher price

When in doubt, try the NOVA scale
Mistral AI@MistralAI · 公司官方 · 1 天前法国明星大模型公司
连环推 ×6

Mistral 正在为欧洲汇集掌控 AI 未来所需的推理基础设施、开源模型和长期承诺,并为全球定方向。🧵:mistral.ai/news/regional-inf…

查看英文原文
☁️Mistral is bringing together the inference infrastructure, open models, and long-term commitments Europe needs to control its AI future, and setting a roadmap for the world. 🧵:
mistral.ai/news/regional-inf…
⚡️A coalition that secures long-term AI capacity: We’re aggregating long-term compute demand in Europe to determine what capacity is built, where it’s located, and whom it serves. Through these multi-year commitments, European Compute Units, participants can use the capacity across the range of products available on Mistral Compute as their needs evolve.
🌎Regional Endpoints and Priority Tier: Customers choose where inference runs - Europe or the U.S. - so requests stay in-region, and local capacity holds when demand peaks, aligning with data residency, regulatory, and latency requirements.
🎯Sovereign intelligence through model choice: We're expanding our platform to third-party open models, starting with
Z.ai
’s GLM-5.2, so enterprises match workloads to the right model and keep the intelligence they build.
⚙️That is the framework Mistral is building toward: one in which enterprises, governments, and startups can use the best AI available, shape it around their own knowledge, and retain the value it creates.
💡The world needs an open-source platform, and that’s exactly what we’re building to give our customers more choice and the flexibility to choose the right model for the right task. As part of this, we’ll continue to innovate in frontier, efficient open models across modalities.
◔ 7.9 万 次浏览(2 条合计)♥ 921⇄ 98观点看原帖 ↗
ollama@ollama · 公司官方 · 1 天前本地跑大模型的热门工具

NVIDIA Nemotron 3.5 Lightning在Ollama上线了!30B模型,专为常驻agents设计,完全本地化。

Claude Code
ollama launch claude --model nemotron-3.5-lightning

Hermes Agent
ollama launch hermes --model nemotron-3.5-lightning

OpenClaw
ollama launch openclaw --model nemotron-3.5-lightning

- 30B mixture of experts模型,3B活跃参数
- 1M token上下文
- 为持续运行的agents而构建:编码、工具调用、多轮对话
- 吞吐量是其他同类开源模型的4倍,任务完成时间快30%

模型页面
👇👇👇

引用 NVIDIA AI @NVIDIAAIIntroducing NVIDIA Nemotron 3.5 Lightning⚡ An open 30B MoE model with 3B active parameters, built for always-on agents to complete high-volume, specialized tasks faster. It delivers up to 4x the output speed of similar-sized models.查看被引原帖 ↗
查看英文原文
NVIDIA Nemotron 3.5 Lighting is available on Ollama!

It's a 30B model made for always-on agents. All local.

Claude Code
ollama launch claude --model nemotron-3.5-lightning

Hermes Agent
ollama launch hermes --model nemotron-3.5-lightning

OpenClaw
ollama launch openclaw --model nemotron-3.5-lightning

- 30B mixture of experts model with 3B active
- 1M token context
- built for agents that stay running: coding, tool calling, multi-turn
- 4x higher throughput and 30% lower task completion time compared to other leading open models of similar size

Model page
👇👇👇
◔ 15.9 万 次浏览(4 条合计)♥ 781⇄ 75新品看原帖 ↗
Guillermo Rauch@rauchg · 创始人 · 1 天前Guillermo Rauch,Vercel 创始人兼 CEO

留意这个页面

vercel.com/ai-gateway/leader…

查看英文原文
Keep an eye on this view

vercel.com/ai-gateway/leader…
Firecrawl@firecrawl · 公司官方 · 1 天前
连环推 ×2

Firecrawl 现在是 @opencode 中web搜索的无密钥选项。

将我们设为您的搜索提供商,您的 AI agents 将获得最新搜索结果,精度顶尖(在 SimpleQA 上达 94.7%),无需 API 密钥或注册。

只需更新 opencode2,选择 Firecrawl,立即开始搜索!

查看英文原文
Firecrawl is now a keyless option for web search in
@opencode
🔥

Set us as your search provider & your AI agents get live results with state-of-the-art accuracy (94.7% on SimpleQA), no API key or signup at all.

Just update opencode2, select Firecrawl, & start searching today!
Check out the docs for OpenCode2 here:
opencode.ai/v2/docs


Learn more about Firecrawl Keyless:
firecrawl.dev/blog/firecrawl…
◔ 7.2 万 次浏览♥ 799⇄ 64▶ 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

说实话,我对任何用 pangram 来'曝光'别人这事儿的看法都是负面的

这总是假设最坏的情况 —— 就是纯粹的 AI 垃圾,发布者什么都没贡献

这就像发明了个'计算机程序检测器',然后用它抹黑某项工作,理由是人家没用脑子手工算所有东西

引用 Susan Zhang @suchenzangoh hell yeah american meituan let's goooo查看被引原帖 ↗
查看英文原文
I'm honestly negatively updating on anyone using pangram to "out" people

it always assumes the worst case that it's just slop, and the person who published it contributed nothing to it

it's like inventing a "computer program detector" and then trying to discredit work, becaused they didn't calculate everything with their brains
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

恭喜你发现了真实世界,在 AI 领域,群体是周期性失忆的。

引用 马天翼 @fkysly有点穿越,Pi 不是 OpenClaw 早期的时候就顺带火了么?4大基本 tool use,怎么现在又开始火了查看被引原帖 ↗
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主
连环推 ×3

天哪我恨 Opus。太懒了。真的让我很生气。所有东西都得重新检查,还得不停地说:做对了!

查看英文原文
god i hate opus. its so freaking lazy. it makes me really angry. Everything has to be double-checked, and you have to keep saying: do it right!
I thought Sonnet 5 was a bad release. My mistake. Opus 5 is the worst release ever.
how can a company with an upcoming IPO and a estimated valuation of $1.2t screw up so badly
Lisan al Gaib@scaling01 · 博主 · 1 天前高频 AI 模型测评与爆料博主

又在用内部模型了

祝AI实验室外的所有科学家好运 :)

引用 levent @__alpoge__+++-+-+-+-+--+--+++-++----++--+----+-----+++---++-+---+++++++-+--+-+-+--++-+--+--+----+-+-+-+-++-++---+--++++--++-++++-+++++--+---++-+---+++++++++--+-+-+--++-+--+--+-----+++-+++-++-+-++++-++-+--+--+------++--+-++-+-+---+--+-+-+-+--++---+++++-----+++++++---+---+--+-+--+-+--+-++-++-++++++-+-++-++-+-+---+--+-+-+-+--++---+++++-----+++----+-+-++---+-++----+---+++++--+---+--++-+++++++-+--+---+----++-+++----++---+++---++++-+-+--+++-+--++++-+++-----++-+++-+++-++++++++-+--+---+----++-+++----++---+++---+++---+-+--++----+--+-+++-+--++------+++--++++++-+--++--+-++---+--++---+---++--+-+----+++-+-++--++++-++-+---+-++--++++++---++++++++-+--++--+-+++--+--++---+---++--+-+-+--+++--+---++++-+-++-+---+-+-+-----+--++-+----+-+-+----+-----++-+++++-+++--+--++++++-++--++---++-+++----+-+--+--++-+-+-+++++-++--+-+++-+-+-+----+-----++-+-+++-+++--+--++++++-++-+++---++++---+++++-++--+--+++-+++--+--+-++-+++--+++-+++--+-+--++++-----+-+-+--+--+---+-++---+++----+++-----+--+++++---+---++-++-+--+-++--+++-+++--+-+--++++-----+-+-+--+--+---+-+++-------+-+-++--++++---+--+-++-+--+---+-++-+-+++++-++--+----++++----+++---++--++-+---+-+--+++++++-+-+--++----+++-++-+--+-++-+++-+--+++-++++-++--+----++++----+++---++--++-+---+-+-------------+-+++-+++---++--++--+-+++--+-+-++---+----++++--+-+-++++-++---+--+----++-++-++++++++++++++-+---+---+++--++--++-+---++-+-+-+-+-+----++++--+-+-++++-++---+--+----++-++-++----++-++++--+++++--+--+--+--+++-----++--+-+-+++-+++-+++-+-+++-+++-+++++-++++---+++---+--+--++++--+++++--+---+++-+--+-----+-+++++++------+++-+--++-+++--++--+++-+-+++-++-+++-+----++-------+--+-+-++-+-+--+--+-+--+-+-+-++-+-+---+++++-+-++-++----++--++++-+--++-+---+-++--+-+----+------+--+-+-+--+---+++++-+++-++++---+-++-++--+-++-+---+++-+-+++-+-+--+++-+-----++-+++--+--++++-++-+-++++--+--++---+++++---+---++++-++----+--+-+-++++++-+++-++++--+---+--++--+-+-++-+++-+++---++---++----++------+--+--++++-+++-++++--+++++-+-+-++++-+-+-++--+++++-----+++-+++++-+---+--+++++-+---++------++-+++-+++-++---++-+--+--+-++++--+-+++-+++-+---++++查看被引原帖 ↗
查看英文原文
again using an internal model

good luck to all scientists outside of AI labs :)
Lisan al Gaib@scaling01 · 博主 · 23 小时前高频 AI 模型测评与爆料博主

今天是美好的一天:

DeepSeek-V4-Pro
Grok 4.6
Qwen3.8-2.4T

还有可能 OpenAI 也会发布点东西

查看英文原文
today is a good day:

DeepSeek-V4-Pro
Grok 4.6
Qwen3.8-2.4T

and maybe something from OpenAI
Qwen@Alibaba_Qwen · 公司官方 · 1 天前阿里通义千问大模型团队

从排名第 22 升至第 4!Qwen3.8-Max 进步明显。感谢突出推荐~✨

引用 Vals AI @ValsAIQwen 3.8 Max nearly doubled its score on Legal Research Bench in under three months, climbing from #22 to #4.查看被引原帖 ↗
查看英文原文
From #22 to #4 on Legal Research Bench! Solid progress for Qwen3.8-Max. Thanks for highlighting~✨
swyx@swyx · 博主 · 1 天前知名 AI 播客 Latent Space 主理人

这已经是今年最重要的论文之一了。

latent.space/p/ainews-how-to…

方法论讲得不太清楚,所以这里是我的进一步总结

引用 Alexander Panfilov @kotekjedi_mlWe can finally talk about it: We found a way to extract hidden reasoning of frontier models using a vulnerability in the APIs of every frontier AI company. We verified that our reasoning token count matches billed API thinking tokens 1:1 for most of the prompts we queried.查看被引原帖 ↗
查看英文原文
this is already one of the most important papers of this year.


latent.space/p/ainews-how-to…


the methodology doesnt seem clearly explained so here are some notes with a further distillation
Qwen@Alibaba_Qwen · 公司官方 · 1 天前阿里通义千问大模型团队

快去 @openart_ai 上试试 Qwen-Image-3.0!🎨👀

引用 OpenArt @openart_aiQwen Image 3.0 is now on OpenArt ✨ The most Real Qwen image model yet. Native text across 12 languages, precise 10px type, and full interfaces like web pages, games, and livestreams rendered with real world knowledge. Available now on OpenArt: openart.ai/suite/create-imag…查看被引原帖 ↗
查看英文原文
Try Qwen-Image-3.0 on
@openart_ai
! 🎨👀
向阳乔木@vista8 · 中文博主 · 23 小时前向阳乔木,中文圈 AI 工具与趋势博主

没有开思考模式,看看DeepSeek V4 Pro生成鹈鹕测试的效果。

提示词:“鹈鹕骑自行车svg,撞上大象,循环播放”

引用 向阳乔木 @vista8DeepSeek-V4-Pro-0813 来了,喜大普奔! 但是今天还是8月12日啊,为啥用0813命名啊,哈哈。 百万tokens输入(缓存命中)0.025元 (vs flash 0.02元) 百万tokens输入(缓存未命中)3元 (vs flash 1元) 百万tokens输出 6元 (vs flash 2元) 并发限 500,Flash 是 2500 注意:价格是当下价格,最近会上涨。查看被引原帖 ↗
Matt Shumer@mattshumer_ · 博主 · 1 天前HyperWrite CEO,AI 实战技巧分享

OpenClaw 大众版现已推出。

引用 Matt Shumer @mattshumer_I’ve been testing Grok Bot for a couple of weeks, and I’ve been surprised by how much I’ve loved it. Honestly, this feels like it could be the thing that gets millions of normal people using agents for the first time. As many of you know, I haven’t always had the best experience with Cursor’s products, but this one feels different. The best way I can describe it is an agent for everything, not just code. The interface feels like iMessage, and you create bots that each have a job and actually get better over time as they learn how you work. There are a lot of products attempting this, but the little details are what makes Grok Bot special. For example, I set up a researcher bot and a writer bot, then made a Chief of Staff bot and asked it to get the other two working together on a project. I checked in fully expecting that to fall apart, because there was no way it just would work out of the box. It worked out of the box. Had a few more experiences like this too. The team clearly really cares about nailing the experience. My only real complaint (which, if they nail it, will end up being a huge win) was the model router, which wasn’t great when I tested it. You don’t choose a model for your Grok Bot. It’s all done automatically on the backend. Incredible for regular users when it’s done well, but frustrating for power users when done poorly. I’m told they’ve made it much better since I tested. It also clearly has sub-agents/workflows in the harness, as it runs Gauntlet Loops natively, and it handled everything I threw at it, from real work to 3D game builds. Very much worth your time to check out!查看被引原帖 ↗
查看英文原文
OpenClaw for the masses is here.
🚨 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 🔥:现在用户可以在 Codex、ChatGPT Work 和其他 AI agents 之间保持项目、聊天、技能和插件的同步。

这很不错,特别是如果你经常测试不同实验室的不同订阅的话。

引用 OpenAI Developers @OpenAIDevs以下是设置自动同步的方法:learn.chatgpt.com/docs/impor…查看被引原帖 ↗
查看英文原文
OPENAI 🔥: Users can now keep projects, chats, skills, and plugins in sync between Codex, ChatGPT Work, and other AI agents.

That's neat, especially if you regularly test different subscriptions from different labs.
◔ 6.3 万 次浏览♥ 561⇄ 32▶ 含视频新品看原帖 ↗
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

又开始打哑谜了。这是 OpenAI 的 Andrew Ambrosiono 今天发的,估计是在暗示即将推出的 Astra 模型家族。

为什么这有必要说?因为 Astra 因为安全和网络安全问题撤回了,这可能把发布时间推得很远。而且确实有充分理由这么做。

这条帖子可以两种方式理解:

1) Astra 的问题和隐患比预期更快得到解决了(原本计划这周前发布撤回),现在他们在为用户做准备。

2) 他们就是想让这个模型保持热度。

为什么重要?Astra 是新的模型家族,配合新预训练,最终应该能打破 Fable 的主导地位。OpenAI 通过 Codex 取得了重大进步,5.6 Sol 版本也是不错的模型。但 Fable 5 在很多关键指标上还是领先的。鉴于即将发布新模型,可以合理推测 Fable 5.5 或其他什么版本肯定会在 IPO 前推出。Anthropic 的 IPO 延到了 9 月/10 月,所以新的 Fable 也不远了。

所以,要最终超过 Anthropic,很多就取决于 Astra,我敢说 OpenAI 对 Astra 的发布是押重注的。从这个角度看,这个打哑谜的帖子不仅是个笑话。这是个信号。Astra 很关键。必须得继续被讨论。这是个大突破。

查看英文原文
We're back to vague posting. This was posted today by OpenAI's Andrew Ambrosiono and is presumably a reference to the upcoming "Astra" model family.

Why is this even worth mentioning? Because Astra's withdrawal, due to security concerns and cybersecurity issues, has likely pushed the release far into the future. And there were indeed good reasons for this.

This post can therefore be interpreted in two ways:

1) A solution to the concerns and problems surrounding Astra was found more quickly than expected (a release before the withdrawal was planned for this week), and now they are preparing users for it.

2) They want to at least keep the model in the public eye.

Why is this relevant? Astra is the new model family that, along with new pre-training, is finally supposed to put an end to Fable's dominance. OpenAI has taken a huge step forward with Codex and presented a very good model in version 5.6 Sol. Nevertheless, Fable 5 still leads in numerous key benchmarks. And considering the upcoming release, it's safe to assume that a Fable 5.5, or whatever it may be called, will certainly arrive before the IPO. Anthropic's IPO was postponed until September/October, so a new fable isn't far off.

Therefore, to finally pull ahead of Anthropic, a lot depends on Astra, and I dare say that OpenAI is placing a lot of emphasis on Astra's release. In that sense, the "vague post" is more than just a joke. It's a sign. Astra is significant. It (must) remain in the conversation. It's a major breakthrough.
◔ 6.6 万 次浏览(2 条合计)♥ 541⇄ 19观点看原帖 ↗
Guillermo Rauch@rauchg · 创始人 · 1 天前Guillermo Rauch,Vercel 创始人兼 CEO

@aisdk 的增长简直离谱。每月 8050 万次下载,增速远超所有 AI Labs SDK。关键是它还完全开源、与厂商无关。

引用 AI SDK @aisdkAI SDK has crossed 20M downloads per week. We’re grateful to every developer who helped us reach this milestone.查看被引原帖 ↗
查看英文原文
The growth of
@aisdk
has been astonishing. ~80.5M downloads every 30 days, and growing faster than all AI labs SDKs. Most importantly: open and provider-agnostic.

有四种对话播客,值得我花时间1.5~2倍速听完,

一种是主持人或者嘉宾控场能力爆棚,娱乐效果拉满,包袱和哏一个接一个,剪辑效果再创作一流,比如papi酱、大物是也,都是走当年康熙的风格;

一种是嘉宾是大明星,我从小就喜欢,愿意听TA讲自己的故事,比如HOPICO采访陈奕迅、方大同、王力宏、苏打绿 ,都是非常精彩的碰撞,有的我反复听了好几遍,

一种是主持人很牛,嘉宾也很牛,双方都是真正经历过大风大浪,或者是某个领域重大变革的亲历者/见证者,分析某个行业时可以随意跳入跳出,深入浅出讲一大堆硬核知识和观点分享,这种要碰运气,我个人比较喜欢咸鱼堡的某几期医疗行业的采访,

第四种是王阿姨和斑马酱的Yi马当先,可惜停更了。

大概就这些。

这些AI领域的播客最大的问题是,嘉宾表达欲望不强,主持人业务和技术水平更是差一大截,观看效果实在是不行。

引用 老董叔 @zhuodong666这几年大量ai领域的播客听下来,很明显不是任何一个ai研究者都适合上去聊两句的。真正好的效果,被采访者往往需要在大模型某个特定方向上有很深的积累,同时对整个大模型整体方向也有全面的认识,这时候才能以点带面,越聊越深入,这才是一场播客的合格水平。查看被引原帖 ↗
Matt Shumer@mattshumer_ · 博主 · 23 小时前HyperWrite CEO,AI 实战技巧分享
连环推 ×2

AI 视频最大的问题就是控制能力。你很难做出完全符合要求的效果。这个技术就是来解决这个的。在 3JS 里画草图,用图像生成来生成风格,再和视频生成融合。怎么做:somethingbig.ai/controllable…

查看英文原文
AI video's biggest problem is control. You rarely get quite what you asked for.

This technique fixes it.

Draft in 3JS, generate style with image gen, and fuse with video gen.

How to do it:
somethingbig.ai/controllable…
This technique is based on the idea
@gmi_cloud
shared, but with improvements that increase stability and allow for style control.

Their approach was great for motion control, but I often ran into issues where style barely changed from the original. This fixes that!
◔ 5.3 万 次浏览♥ 198⇄ 12▶ 含视频教程看原帖 ↗

这也是采访者的问题,采访者大部分完全没听懂。

我点开过几期,我明显感觉那个播客的主持人手里有个大纲,列了七八个问题,然后嘉宾直接一条条往下讲,讲得模糊的也直接划过去了,

关键是主持人,不仅没听懂,而且就跟个大傻子似的,直勾勾盯着采访者,接不住话茬子,也不知道哪里该切入具体聊下去。

引用 老董叔 @zhuodong666现在大量的播客采访听起来味同嚼蜡,请上来的ai研究员要么只能在自己研究领域说上两句,但是扩展到整个大模型系统上就云山雾罩不知所云,要么满嘴片汤话都是都整个ai大行业的泛泛之谈,但是没有对自己多年积累的方向有任何深刻认识,要么表达不通顺说话漂移严重,查看被引原帖 ↗

terminal UI从来没死过,这东西哪怕在2023年以前,也一直大量存在于各种cli工具中,大量语言和框架已经把基建做得差不多了;

terminal UI也从来没活过,短期一堆LLM Agent糊了个UI,在大众强烈需求的情绪反弹下,依然会回到electron大号聊天机器人的技术栈。

TUI这种葱姜蒜级别的下脚料技术, 在今天依然被反复炒作,反复回锅,

充分证明了哈珀发明了合成氨这件事,导致太多人吃饱了撑的没事干了。

《立党基础教育》最核心的精髓,一个是给6岁孩子孩子买mac mini+大显示器,一个是订阅coding plan(claude/OpenAI/智谱/kimi/阿里/grok),

最核心是建立“党哥家庭图书馆”,卧室装满这些书,可以陪伴孩子从6岁到18岁的数学物理计算机教材,让孩子随机翻、随机看、躺着看、拉屎看,挖掘天赋潜能。

一定要听党哥的,党哥就住天津图书大厦旁边,这些书党哥没给童年的自己买齐,党哥想扇自己100个大嘴巴子,你们一定不要给自己家聪明的孩子留遗憾,先把书买齐了,孩子只看10%,每本书只看一章,都是有用的。

1. 小学、初中、高中全年级的数学、物理、信息学基础官方教材;

2. 小学、初中、高中全年级的数学、物理、信息学学科竞赛教材和题集;

上面这些书一定要买齐,供孩子随时按照兴趣翻阅,吃饭时翻阅,拉屎时翻阅,无聊时翻阅,

3. 大学本科基础数学教材《微积分》《线性代数》《概率统计》《复变函数》,如果孩子看得津津有味并且觉得太简单,立刻按照北大数学系培养方案继续买实分析、高代、抽代、组合数学等等,

喜欢物理直接买四大力学教材,或者直接买《费曼物理学讲义》,按照孩子需求买齐买全,挖掘孩子在数学和物理上的最大天赋;

4. 一定要买齐计算机教材,从6岁可以陪伴到60岁的几本书,都是党哥自己掏真金白银在天津或者Texas亲自买过的纸质书,哈佛MIT斯坦福也用这些书当做本科生或者graduate level(硕士博士阶段课程)教材,记住,一定要买齐,摆孩子卧室里,让孩子随机翻阅,随机看,拉屎看,吃饭看,睡觉看,周末看,看一章节就算稳赚。

记住,这些书尽量买齐,在中国就买影印版中文翻译版,一本书50块钱,买齐了并不贵,在英语国家可以买二手英文版,这些书可以陪伴绝大多数人从6岁到60岁。

a. 《Python Crash Course: A Hands-on, Project-based Introduction to Programming》(Python编程:从入门到实践),孩子的第一本编程书,让孩子一边拿着书,一边对着电脑敲敲敲,敲敲敲,敲敲敲,

python入门了以后,可以买《Thinking in Java》《C++ Primer》,挑着学Java和C++这两本书;

b. 计算机三套经典,党哥都买了纸质版,党哥都没看完。

- 计算机基础教材CSAPP《Computer Systems A Programmers Perspective》(CSAPP)

- 算法和数据结构圣经《introduction to algorithms》(算法导论)

- 可以陪伴到50岁、几代人读的圣经大套装《The Art of Computer Programming》(TAOCP)

记住,党哥没看完是因为岁数太大了(算法导论看得最多,其他两本进度很少),你的孩子只有6~18岁,如果激发了巨大的兴趣,你的孩子是完完全全有精力看完的。

如果18岁之前能扎扎实实看完这三本书,可以说比Google和Meta里面90%的员工要强。

c. 全球公认的计算机其余主要专业教材

操作系统《Operating Systems Concepts》(操作系统概念,恐龙书)

编译原理《Compilers: Principles, Techniques, and Tools》(编译原理,龙书)
《Modern Compiler Implementation in C》(现代编译原理,虎书)
《Advanced Compiler Design and Implementation》(编译器设计与实现,鲸书)

数据库《Database System Concepts》(数据库概念)

计算机网络《Computer Networks》Andrew S. Tanenbaum

计算机图形学《Fundamentals of Computer Graphics》(图形学虎书)

机器学习和深度学习《Pattern Recognition and Machine Learning》(PRML)和《Deep Learning》 by MIT Press

党哥不是拉书单,党哥给你推荐图书是非常谨慎小心的,已经砍到不能再砍,精简到不能再精简了,精简到再砍掉任何一本就会出现大问题了,

记住,这些计算机科学computer science书籍,要买全,买齐,在中国大陆就买影印版、翻译版,在美国就买二手,把这些书整整齐齐放在孩子书架上,

孩子靠兴趣驱动学习时,是跳跃的、离散的、冲动的,也许孩子感兴趣两个月看完一本书,也许两年翻不了三页,也许一本书只看零星两三章,

但是一定要给孩子买齐、买全、买完整,放在卧室书架上,吃饭看、拉屎看、无聊看、闲着看、呆着看、随机翻看、边看边敲、边看边在claude code里实现,

一定要买齐,在家里置办“党哥家庭图书馆”,孩子对知识的欲望和冲动,远比对抖音直播间大美女的冲动要强烈得多,当冲动来临的时候,一定要做好图书、知识、教材的准备,让孩子凭借一腔热血和兴趣,选择随机翻阅和学习。

当然,另一件重要的事情,就是听党哥的话,围绕北京、上海、深圳、bay area全球四大科技中心,每个月至少带孩子逛一次电子展、科技展会、技术大会、学术会议、路演日、融资大会、技术沙龙、AI线下沙龙、技术公开课、产业大会、学校技术论坛、学校学术展览,逐步建立世界观,认识技术,认识产业,认识各行各业,形成自己的价值观,并且开始接触一些各行各业的“大朋友”,

当这些大朋友给孩子一些启发的时候,孩子就会立刻回家,闷头去翻这些教材,继续去补充营养,带着白天开会时满脑子的小问号,回到自己的“党哥家庭图书馆”,去狠狠恶补自己欠缺的知识。

记住,除了买mac mini+大显示器+人体工程学座椅+订阅coding plan,除了开会,家长首先要做到的,就是买书,买书,买书,买书,买书,买书,买书,买书,买书。

如果要看视频课程,可以按照csdiy wiki上面课程去学习,可以上B站上中文平替,

先从哈佛CS50起,接下来看UC Berkeley CS61A、CS61B、CS61C,然后按照csdiy wiki上面对应的视频课程自己去学习,一般youtube有原版,B站有翻译版本。

让孩子自己自由随机去挑感兴趣的视频去跟着学。

但是一定他妈给我把这些书给孩子买齐,买全,加起来没有几千块钱,能改变你的孩子一辈子的命运,让他获得一个9位数的人生。

现在就给我去京东天猫拼多多闲鱼去下单,给我买!!!!

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

Higgsfield 又制作了一部完整的超长AI电影

这可能是 AI 电影真正开始进入长片时代的一个节点

片长110 分钟
4 周制作
花费200 万美元

Higgsfield 和 Cully Games 用 AI 完成了一部长达 110 分钟的剧情长片

有趣的是该AI电影由真人演员参与,但是并不是真的参演, 而是使用了他们的肖像授权进行AI制作…

而且他们将整个制作过程完全公开:每条Prompt、每个素材、每个角色,免费可复用。

过去两年,我们看到的大多数 AI 视频还停留在几秒、几十秒的 Demo,大家比的是画质、运镜和生成效果。

现在,AI 已经开始进入一部完整电影的生产流程,而且还有真人演员参与。

从「生成几个惊艳的镜头」,到「真正制作一部 110 分钟的电影」,这个变化来得比想象中快得多。

AI 电影真正值得关注的阶段,可能才刚刚开始。

引用 Higgsfield AI 🧩 @higgsfield_aiWe made the first 110-minute AI feature film with a real cast, The Cully Hill Boys, on Higgsfield for $2,000,000. Starring @N3onOnYT , @stylebender , @RampageJackson , and @MKIATPIS It's 100% open-sourced on Higgsfield: all prompts and assets are public now. Made with Seedance on Higgsfield. Watch full film below 👇查看被引原帖 ↗
Guillermo Rauch@rauchg · 创始人 · 1 天前Guillermo Rauch,Vercel 创始人兼 CEO

Vercel 上的部署自今年年初以来增长了 3 倍。前 6 个月增长了 2 倍。Agents 喜欢部署。

我们通过卓越的可靠性支持了这种增长,同时在这段时间内的公开 changelog 上推出了 19 项构建和部署性能优化。

Redis 自从我 10 年前创建 z̵e̵i̵t̵ Vercel 基础设施以来一直给力,服务于很多地方。但就像很多这样的情况,Dynamo 才是应对这种规模需求的正确选择。

引用 Vercel @vercelWe migrated the database behind every build at Vercel, one of our most critical services, while it handled 6,000 deployments a minute. The story of how it happened, our philosophy, and what we learned ↓ vercel.com/blog/how-we-migra…查看被引原帖 ↗
查看英文原文
Deployments on Vercel are up 3x since the start of the year. And 2x the 6 months prior. Agents love deploying.

We supported this growth with excellent uptime, while also shipping 19 build & deploy performance improvements on our public changelog in this period.

Redis served us well since I first created the z̵e̵i̵t̵ Vercel infrastructure 10 years ago, in numerous places. But as many such cases, Dynamo is the right beast for this size of demand.
Ethan Mollick@emollick · 创始人 · 1 天前沃顿商学院教授,AI 应用研究权威

我喜欢现在所有 AI 评论员都得假装自己从来深刻理解那些只有专家听过的一堆未解数学问题的区别:"Gromlach 猜想对 r 维矩阵是错的,哇,这比上周对受限样条的 Erdos 问题 444 的解决方案还牛逼!"

查看英文原文
I like that all AI commentators now need to pretend they have always had a careful nuanced grasp of the difference between a bunch of unsolved mathematical problems that only specialized experts had heard of: “The Gromlach Conjecture is false for r-dimensional matrices, wow, that is more impressive than last weeks solution to Erdos Problem 444 for restricted splines!”
el.cine@EHuanglu · 博主 · 1 天前
连环推 ×2

这个新的 AI 电影工作室真的很牛。你可以控制面部特征获得逼真的人物情感,生成详细的角色表。在 AI 时间线上编辑、添加 VFX 和 SFX,加入等候名单获取早期访问码和免费额度:filmera.ai

引用 el.cine @EHuangluAI video is now undetectable查看被引原帖 ↗
查看英文原文
this new AI film studio is next level

you can control facial features to get realistic human emotions, generate detailed characters sheet

edit, add vfx and sfx on AI timeline, join waitlist to get early access code and some free credits:
filmera.ai
this is the first ai platform that has professional filmmaking workflow

the ai agent assist you to write prompts, manage assets in media page and edit on AI timeline

join waitlist to get early access code and free credits:
filmera.ai
◔ 4.5 万 次浏览♥ 333⇄ 37▶ 含视频演示看原帖 ↗
Bilawal Sidhu@bilawalsidhu · 博主 · 1 天前
连环推 ×2

如果你拍的每段视频都能在原地变成永久的3D全息影像呢?我先用照片试过了,现在视频也可以。这是我几天前用iPhone拍的母鹿和小鹿,现在作为地理位置锚定的全息影像漂浮在拍摄地点。

查看英文原文
What if every video you ever shot could turn into a permanent 3D hologram in the exact place it happened?

I did it with photos first. Now I can do it with any video.

Here’s a mama deer and her baby deer, filmed with my iphone a few days ago -- now floating as geo anchored holograms exactly where I captured them.
Spatial memory palace of a large number of photos:
◔ 4.4 万 次浏览♥ 1,096⇄ 108▶ 含视频演示看原帖 ↗
Ethan Mollick@emollick · 创始人 · 1 天前沃顿商学院教授,AI 应用研究权威

有趣的研究表明,单从一个数据源判断谁赢了要谨慎啊...OpenRouter 的数据显示开源模型越来越猛,但 Pangram 上提交的作品大多是 ChatGPT,不过 Claude 的比例越来越高。pangram.com/blog/pangram-s-m…

查看英文原文
Interesting research suggests caution in determining which AI company is winning by looking at any one source..

OpenRouter seems to show open weights winning over time, but work submitted to Pangram is overwhelmingly by ChatGPT or increasingly, Claude.
pangram.com/blog/pangram-s-m…
宝玉@dotey · 中文博主 · 1 天前宝玉,中文圈 AI 翻译与科普大 V

我一直有个观点,就是:
人不用去跟汽车比谁跑的快,不用跟 AI 比谁更聪明博学;我们驾驭好汽车到达目的地,用好 AI 完成任务,就很好了。

引用 Jackywine @Jackywine我有一个绝望的观点: 人类已经彻底跟不上 AI 的发展速度了 但是在自己力所能及的情况下,还是尽可能多的学习一点 我的 Agent 启蒙是 anthropic 的官方教程,再加上一些其他的 AI 智能体官方文档,交叉之后发现 Agent =LLM +Loop 当然,人的参与也很重要查看被引原帖 ↗
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

关于 Anthropic 和水印问题:什么是真的,什么不是,真正的问题在哪。

根据我的了解,对 Claude 新文本水印的反对不是关心秘密追踪,而是关于作者身份认定、质量影响,和谁该为这个不完美的检测系统承担代价。

记住:Anthropic 计划在 Claude 模型生成的文本中嵌入一个不可见、机器可读的信号。

第一个问题是解释:

检测到的水印并不能证明是 Claude 写的文档。Anthropic 说,它也可能出现在有人用 Claude 校对、翻译或改进人类原创文本时。

但雇主、大学、客户或出版商能理解这个区别吗?

"由 Claude 处理"很快就会被理解成"由 AI 写的"。这是个真问题!

第二个问题是质量:

文本水印通过影响模型的 token 选择来创建可检测的统计模式。Google 的 SynthID 研究表明这可以在不明显降低质量评分的情况下完成,虽然某些配置会降低响应多样性。

但 Anthropic 还没有发布足够的技术细节或独立测试来证实"质量不受影响"的说法。这又是个问题!

第三个问题是有效性:

坚持不懈的用户可以通过大量改写、翻译、用其他模型或未标记的开源系统来减弱水印。而普通用户让 Claude 润色邮件、编辑文档或帮助写代码时,更可能保留水印。

EU 法规甚至豁免不实质改变输入或其含义的标准编辑。但 Anthropic 更宽泛的(!)实现似乎覆盖范围比这个最低要求还广。

所以真正的问题不是什么隐形间谍在监听,而是一个技术上有限的来源信号,可能被当成对人类作者身份的最终判定。这个担忧是合理的。

(顺便说,关于隐私:混淆来自 Anthropic 说水印在"模型级别"工作。

这不是说 Claude 能追踪文本回到你的账户、身份或之前的对话。

这是说标记是在模型生成文本时添加的。所以无论你用 Claude.ai、Claude Code、API、AWS 还是其他平台,标记都可能出现。这是在 Claude 产品中工作,而不是追踪你的个人会话。)

简版:水印确实有问题,我对这些问题有严肃的关切(我也批评 EU 想再次监管这个本不需要监管的东西),但问题不在隐私上,而是如上所述的其他地方。

查看英文原文
Regarding the Anthropic and Watermark issue: What's true, and what's not and whats the real problem.

From what ive read, the backlash to Claude’s new text watermark is not about secret tracking but about authorship, quality and who bears the cost of an imperfect detection system.

Remember: Anthropic plans to embed an invisible, machine-readable signal directly into text generated by supported Claude models.

The first problem is interpretation:

A detected watermark does not prove that Claude wrote a document. Anthropic says it may also appear when someone uses Claude to proofread, translate or improve a text originally written by a human.

But will an employer, university, client or publisher understand that distinction?

“Processed by Claude” could quickly become “written by AI.” This is a real problem!

The second concern is quality:

Text watermarking works by influencing the model’s token choices to create a detectable statistical pattern. Google’s SynthID research suggests this can be done without a measurable drop in normal quality ratings, although some configurations reduce response diversity.

Anthropic has not yet published enough technical detail or independent testing to evaluate its own claim that quality is unaffected. Thats another problem!

The third problem is effectiveness:

A determined user can weaken a text watermark through extensive rewriting, translation, another model or an unmarked open-source system. Ordinary users asking Claude to polish an email, edit a document or help with code are more likely to retain the mark.

The EU regulation even exempts standard editing that does not substantially change the input or its meaning. Anthropic’s broader (!) implementation appears to cover more than that minimum.

So the concern is not an invisible spy inside every Claude response but a technically limited provenance signal that may be treated as a definitive judgment about human authorship. That concern is legitimate.

(Btw re privacy: The confusion comes from Anthropic saying that its watermark works "at the model level."

That does not mean Claude can follow a text back to your account, identity or previous conversations.

It means the mark is added while the model generates the text. So it can appear whether you use Claude. ai, Claude Code, the API, AWS or another platform. It works across Claude products, not across your personal sessions.)

tl;dr: There are real problems with the watermark and I have serious concerns about them (and I criticize the EU for wanting to regulate something again that is unnecessarily regulated), but they don't lie in privacy, but rather elsewhere as described above.
Ethan Mollick@emollick · 创始人 · 1 天前沃顿商学院教授,AI 应用研究权威

如果LLMs对科学的贡献仅限于数学领域——以新颖方式跨子领域整合思想——那也将是革命性的。在LLMs出现之前,科学在知识的重压下停滞不前,信息太多难以吸收,工作也因此僵化。

引用 Ethan Mollick @emollickThe paradox of our Golden Age of science: more research is being published by more scientists than ever, but the result is actually slowing progress! With too much to read & absorb, papers in more crowded fields are citing new work less, and canonizing highly-cited articles more.查看被引原帖 ↗
查看英文原文
If LLMs did nothing else for science than what they have been doing in math - combining ideas across subfields in novel ways - it would be revolutionary. Before LLMs, science was stalling under the burden of knowledge, there is too much to absorb & work was ossifying as a result
ollama@ollama · 公司官方 · 1 天前本地跑大模型的热门工具

很高兴 Ollama 能够与 @JensenHuang 和 @NVIDIAAI 团队合作推出开源模型。

开源模型没有边界,让我们继续合作使这个生态系统更好!

引用 Jensen Huang @JensenHuangLightning strikes for continuous and long-run agents! Nemotron 3.5 Lightning is smart, fast, efficient and open.查看被引原帖 ↗
查看英文原文
It's been exciting for Ollama to partner with
@JensenHuang
and the
@NVIDIAAI
team on launching open models.

Open models have no boundaries, and let's continue to work together to make this ecosystem better!
宝玉@dotey · 中文博主 · 1 天前宝玉,中文圈 AI 翻译与科普大 V

Manus 还是挺了不起的,期待再创辉煌

引用 晚点 LatePost @latepostnews《Manus 第二季,浪子回头》 8 月 11 日晚间,Manus 网站上挂了小半年的那句 “Manus 现已成为 Meta 的一部分”,变成了 “Manus 即将恢复独立运营。” 从被收购到确认拆分,中间不到八个月。 2025 年 7 月,肖弘离开了长居的武汉,和整个团队搬去新加坡。同年 12 月 29 日,他决定将这家公司以超过 20 亿美元卖给 Meta,出任 Meta 副总裁。 这是大模型应用领域当时最大的收购案,交易从接触到签约仅约 10 天;也是 Meta 历史上第三大并购案,仅次于 WhatsApp 与 ScaleAI。 交易宣布一周后,中国商务部新闻发言人在例行记者会上说,将会同相关部门对此项收购与出口管制、技术进出口、对外投资等相关法律法规的一致性开展评估调查。 戏剧性转变引发了看戏的热情,社交网络上参与讨论的人远远多过用过 Manus 的人,“香槟开早了”“20 亿美金一夜归零” 被复读。也有人将其视为转折点——在中国创业,通过美国资本退出的终结。 这期间,曾经会从任何缝隙挤进来,争取股份的风险投资机构谨慎地保持着距离。肖弘、季逸超和张涛每日跟留在新加坡的团队打视频电话,同步进展,安排工作。 风波中,Manus 产品依然保持每周 1-2 个功能更新,跟 2025 年底的高峰相当,快过一年前。但没有大版本号变动,产品停留在 1.6,更多是功能完善,增加连接器(接入 GMail、Shopify、Slack 等全球产品)、上线客户端。一位 Manus 人士说,开发节奏如常,2.0 版本已经开发完,将在交易完全结束后再发布。 4 月 27 日,国家发改委发布决定:“对外资收购 Manus 项目作出禁止投资决定,要求当事人撤销该收购交易。”Meta 随后在发给媒体的一份声明中回应称,交易完全符合适用法律,并期待审查得到妥善解决。交易撤销(deal unwind)——通常是指,撤销已经完成的交易并尽可能恢复交易前的状态。 几天后,Meta 与 Manus 实际拆分开始。据媒体报道,两边停止了数据共享,Manus 员工也不再能访问 Meta 内部系统。 到了六七月,保持距离的投资机构开始意识到这是一个有利可图的机会。有一位投资人告诉我们,如果还有资金缺口,他们很想参与进来。 一些嗅觉敏感的 VC 和企业家开始行动,他们准备好资金,希望参与 Manus 回来之后的首轮融资。 因为在风波发生的这三个月里,Manus 收入增长了约 5 倍,从每日 30 万美元上涨至近 150 万美元,一度超过了 DeepSeek。 其间,各种传言不断。7 月,有报道称腾讯将控股 Manus。于是,Manus 北京与新加坡团队每天保持沟通,安抚情绪、稳定团队。几个月的动荡中,这家 150 人的公司无一人离职。 在北京的人有了新的生活节奏。肖弘重新学数学、研究大模型。张涛学习攀岩、练习魔方,花了半年把体重从 150 斤减到 120 斤,每天睡够 7 小时,连血压都降了下来。 这几个月重画了这家创业公司的轨迹,也让它的创业者想清楚自己到底想要什么。 回到 2025 年 12 月,Manus 正在融资,Meta 来人说想要见见核心团队,看是否有业务合作机会。几轮沟通之后,在一次核心团队全员参加的会议中,猝不及防地,Meta 当着一群人的面表明了收购意图。到这时,多位合伙人、核心技术人员已经跟随肖弘多年,经历过两次创业。卖掉公司,是对所有人的交代和回报。 收购结束后,2026 年元旦,肖弘走在新加坡海边,觉得曲折的十年旅程已经走到终点,接下来要去大公司上班了。但真正抵达这个看似圆满的终点时,他心里反而有些失落。 而在收购明确将被撤销后,外界眼中的巨大反转和打击,反而让这几位连续创业者重新有了期待:他们的创业故事进入了第二季。 据一位接近 Manus 的人士说,肖弘在团队电话会上表示,他不想旁观或回忆这个时代,而要继续置身浪潮、参与其中。否则,他只会变成一个写自传回忆当年的人。 Manus 是一个经典的科技创业故事。有 ChatGPT 之前,肖弘用了更早发布的 GPT-3 大语言模型,看到技术的可能性,拉季逸超、张涛入伙,给 GPT 做了个浏览器的壳 “Monica”。 浏览器很难让人看到下一个超级产品的可能性。一位创业者在 Manus 发布前一年见到肖弘的一位合伙人,聊到已经发布的 Monica,觉得对方颇有些不自信。 开发 Monica 期间,伴随大模型的进步,这个团队在试错中找到了超前的产品形态,让 agent 像人一样全自动工作。到 2024 年,Cursor、Copilot 乃至 ChatGPT 都还在假设人盯着屏幕,不断发提示词指挥 AI 工作。当时模型推理成本比现在高一个量级,但 Manus 已查看被引原帖 ↗
Mustafa Suleyman@mustafasuleyman · 创始人 · 1 天前微软 AI CEO,DeepMind 联合创始人

我们最新的代码模型速度快 25%,质量更好,成本仅为 6 月版本的四分之一。现在已在 GitHub Copilot 中上线 - 去试试吧。

查看英文原文
Our latest code model is 25% more efficient, higher quality, and a quarter of the cost than our model launched in June. Now live in GitHub Copilot - try it out.
Matt Shumer@mattshumer_ · 博主 · 23 小时前HyperWrite CEO,AI 实战技巧分享

看起来 @SpaceXAI 在追赶最前沿。
如果这个模型像基准测试显示的那样好,他们很快就能成为领先实验室...
真是 AI 竞赛中激动人心的搅局!

引用 SpaceXAI @SpaceXAIIntroducing Grok 4.6. It delivers frontier intelligence and is a significant improvement over Grok 4.5 at the same price.查看被引原帖 ↗
查看英文原文
Looks like
@SpaceXAI
is catching up to the frontier.

If this model is as good as the benchmarks say, they're on track to be a leading lab very soon...

Super exciting shake-up in the AI race!
Gary Marcus@GaryMarcus · 博主 · 1 天前

👇👇👇

注意看:这就是AI发展的转折点——从威胁科技公司,变成威胁整个银行系统和经济体系。那些向来擅长把烂账打包给别人的银行,这回开始玩起"AI接盘侠"游戏了。

下面是个恐怖的长推文链

引用 Porter Stansberry @porterstansbHere's the real reason the A.I. boom is going to run out of capital. It's the same reason why rates are rising. And it's exactly why there's suddenly a mad scramble for capital in A.I. Equity values will fall as the cost of capital increases 20%-30%. But that's only the beginning👇查看被引原帖 ↗
查看英文原文
👇👇👇

“Pay attention: This is where the A.I. build out goes from something that will threaten tech companies to something that will threaten our entire banking system and our entire economy. Banks, ever ready to package dodgy debt for other people to hold, have begun operation “A.I. bag holder.”

long, terrifying thread below
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

好的,听起来 Composer 3 要来了。真的很想看评测结果

引用 Ricky Doar @rdoar4我认为Cursor没有经历过更大的一周,理由有三个。查看被引原帖 ↗
查看英文原文
Ok, sounds like composer 3 is incoming. Really curious for the evals
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

这就是我喜欢看到的研究:Google 让 Gemini 经历了一次模拟医学实习,它在临床咨询上明显变强了。

ResidencyRL 在 49,870 个模拟远程医疗对话中训练了 Gemini 3.5 Flash,覆盖 81 种疾病。

AI 患者会隐瞒症状、拒绝建议、提出不合理要求,迫使模型在长达 60 轮的对话中收集信息。

训练后,对抗性条件下诊断准确率从 81% 升到 88%,漏诊关键症状的情况下降 31%。在 97 个病例的盲测中,临床医生在 87.6% 的情况下更倾向于训练后的代理而不是基础模型。

这个改进还转移到了未见过的肿瘤学案例和外部基准上。

AI 健康代理要来了 <3

引用 Samuel Schmidgall @SRSchmidgallDoctors don't become doctors from textbooks. They become doctors through residency, requiring years of practice across thousands of patient encounters. Can AI learn the same way? Introducing ResidencyRL 🧵查看被引原帖 ↗
查看英文原文
Thats the reaserach i love to see: Google put Gemini through a simulated medical residency, and it became markedly better at conducting clinical consultations.

ResidencyRL trained Gemini 3.5 Flash across 49,870 simulated telehealth encounters covering 81 conditions.

The AI patients hid symptoms, resisted advice and requested inappropriate treatments, forcing the model to gather information over conversations of up to 60 turns.

After training, diagnostic accuracy under adversarial conditions rose from 81% to 88%, while missed red flags fell by 31%. In a blinded evaluation of 97 cases, clinicians preferred the trained agent over the base model in 87.6% of comparisons.

The improvement also transferred to unseen oncology cases and external benchmarks.

AI health agents incoming <3
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

微信官宣了自己的模型

目前其微信里的小微助手就是由该模型驱动的

两款:

WeLM-80B

总参数量:800 亿
每次推理激活30 亿
已部署到微信原生 AI 助手小微中,可以调用微信原生功能和小程序

用户说一句话 → WeLM 理解需求 → 调用微信功能/小程序 → 帮用户完成具体操作。

WeLM-617B

总参数量:6170 亿
每次推理激活参数:230 亿

◔ 3.6 万 次浏览(2 条合计)♥ 68⇄ 13新品看原帖 ↗
François Chollet@fchollet · 创始人 · 1 天前
连环推 ×2

Expedia 前不久把排名模型迁移到了最先进的 Keras 3 架构。效果:训练快了 30%,推理延迟降低了 70%。

阅读他们关于此升级的文章:

medium.com/expedia-group-tec…

查看英文原文
Expedia recently moved its ranking models to a state-of-the-art Keras 3 setup. Results: 30% faster training, and inference latency decreased by 70%.

Read their writeup about the upgrade:

medium.com/expedia-group-tec…
The post has a lot of great technical insights on performance optimization for serving. If you're looking to serve large-scale ranking and recommendations models at ultra-low latency, definitely take a look.
Guillermo Rauch@rauchg · 创始人 · 1 天前Guillermo Rauch,Vercel 创始人兼 CEO

“你行的,再努力点,坚持下去”这类提示词背后有个很妙的地方,这就是工程管理。

你知道结果就在眼前,所以决定投入更多时间和金钱去追。你鼓励前行。但你也需要智慧,明白自己可能是在追一个死胡同,必须调头止损。

这是Agent管理。

查看英文原文
There’s a cool meta to the “you’ve got this, try harder, keep going” prompts. That’s engineering management.

You know the outcome is within sight, so you decide to spend more time and money chasing it. You encourage. And you also need the wisdom to know you’re chasing a deadend and have to pivot and cut losses.

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

Google 在 Gemini Enterprise 上测试聊天和任务的新切换功能。看起来很熟悉 👀

查看英文原文
Google is experimenting with a new toggle between Chat and Tasks on Gemini Enterprise.

This looks familiar 👀
宝玉@dotey · 中文博主 · 1 天前宝玉,中文圈 AI 翻译与科普大 V
连环推 ×2

赞,正想找个 Rust + GPUI 的代码库学习下👍

引用 EGOIST @localhost_5173Introducing ⚡Waku, a fast and native desktop app (Rust+GPUI) for all your coding agent CLIs. waku.sh is free and open source.查看被引原帖 ↗
Guillermo Rauch@rauchg · 创始人 · 1 天前Guillermo Rauch,Vercel 创始人兼 CEO

用你对 Vercel 仪表板的批评反馈回复我们,我们会修复!

引用 wits @witsdevI lead the dashboard team at Vercel every day we ship some changes based on a feedback channel in our Slack reply here with your feature requests, bug reports, or any other feedback and I'll forward it to that channel查看被引原帖 ↗
查看英文原文
Reply with your critical Vercel dashboard feedback and we will fix!
yihong0618@yihong0618 · 中文博主 · 1 天前

没想到 minimax 不卷 coding 靠 video 杀回来了。真厉害。

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

哇!话想不到我之前的这个项目还有 fork!当时大模型太慢了,效果不太好,就搁浅了这个项目,想不到还有后续。哭死😭

引用 码妞 @niuniu_8964@yetone 还是太超前了 顺便打个广告,这是我和 @XYen0n 共同维护的 fork github.com/XYenon/smart-sugg… yetone 可以把重心放在探索下一个奇点上🫡查看被引原帖 ↗
Sebastian Raschka@rasbt · 博主 · 1 天前
连环推 ×2

《Build a Reasoning Model (From Scratch)》终于也在Amazon上架啦!真的要感谢所有预订的各位!

特别提醒印度的朋友们:最近千万别从@amazon India(@amazonIN)下单,他们在卖假货(黑白版)。正版这次是彩印的,还配有Manning网站的注册码。如果你在印度,我建议直接从Manning官网订购。(要是收到假货的话一定要退货啊,真的很抱歉给你添麻烦。)

这些假冒销售直接伤害了我这个独立作者的利益。而且从Amazon India收到的假货评价也会显示在全球Amazon.com页面上,这样正版的评价看起来也跟着有问题了。

根据昨天从Amazon.com收到书的读者那儿听到的反馈,Amazon.com本身应该是安全的。我自己也从Amazon.com订了一本,可以确认是正品。

如果你收到的是正版,并且已经开始看了,我真的很希望你能在Amazon.com上留个评价,哪怕就一两句。这真的很有帮助,尤其是因为Amazon India假货的评价也会显示在Amazon.com的评价页面里。谢谢大家!

查看英文原文
Build a Reasoning Model (From Scratch) is now finally also available on Amazon! Thanks so much to everyone who preordered!

One important note for readers in India: please avoid ordering through
@amazon
India (
@amazonIN
) for now since they are selling counterfeit black-and-white copies. The genuine book is printed in color this time and includes a code for registering the book on Manning’s website. If you are based in India, I recommend ordering directly from Manning. (If you received a counterfeit copy, please return it. I am very sorry about the hassle.)

Unfortunately, these counterfeit sales also directly hurt my work as an independent author. And reviews from readers who received the fake copies from Amazon India also appear on the global Amazon. com page, which makes it look as if the genuine copies on Amazon. com have the same issues.

Based on what I have seen and heard from readers who received the copies from Amazon. com yesterday, Amazon. com itself appears safe. I also ordered a copy through Amazon. com myself, and I can confirm that it's genuine.

If you received a genuine copy and have had a chance to read it, I would be very grateful if you considered sharing your experience in a review on Amazon. com, even a short one. It would really help, especially since reviews for the fake copies from Amazon India also show up on the Amazon. com page. Thank you!
One clarification about the previous 2024 Build a Large Language Model (From Scratch) book, not the new 2026 Reasoning From Scratch one: the genuine edition of the 2024 book is printed in black and white. Counterfeit copies can therefore be harder to spot, but they are usually lower quality and do not include the Manning registration code.
◔ 2.6 万 次浏览♥ 652⇄ 59▶ 含视频动态看原帖 ↗
clem 🤗@ClementDelangue · 创始人 · 1 天前HuggingFace 联合创始人兼 CEO

本地 AI 爆炸式增长!

我们在 @huggingface 过去三年一直在开发的 Transformers.js,现在已经成为在浏览器中直接运行 AI 模型最受欢迎的开源库,月下载量突破 1000 万(!!),比半年前增长了接近 10 倍。

更广泛地说,我很高兴看到有意义的一部分 AI 工作负载迁移到本地,因为它既免费又完全私密!在计算资源短缺和网络攻击风险增加的时代,我们比以往任何时候都需要它。

一起来!

查看英文原文
Local AI is exploding!

Transformers.js, that we've been building
@huggingface
for the past three years has now become the most popular open-source library to run AI models directly in your browser and crossed 10 million (!!) monthly downloads, close to 10x from just six months ago.

More generally, I'm excited to see a meaningful percentage of AI workloads move to local because it's free & fully private! At a time of compute shortage & increased cyber-attack risks, we need it more than ever.

Let's go!
Unitree@UnitreeRobotics · 公司官方 · 1 天前

Unitree 人形机器人
累计生产和下线数量:约 18,000🥳
仅统计仿生双足人形机器人,不包括其他人形机器人类型和轮式人形平台

查看英文原文
Unitree Humanoid Robots
Total units produced and offline cumulatively: Approximately 18,000🥳
Bionic Bipedal Humanoid Count Only, Excluding Other Humanoid Types and Wheeled Humanoid Platforms
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

当前企业 AI 领域存在一个奇怪的问题:公司花费数百万美元来集成他们的竞争对手五分钟后就能访问的同样模型。

这可能有用,但这不是护城河。

Oumi 今天推出的东西很有趣,因为它把 AI 看得不像是部署一次就算完事的软件,而更像是应该从它实际做的工作中学习的系统。

构建一个专用模型,部署它,看它在哪里失败,把那些失败转化成训练信号,改进它,再部署。

一个循环,而不是一次发布。

公司保留对所创建的模型、数据、评估和配方的访问权。甚至编码 agents 也可以通过 Oumi 的 CLI 操作工作流。

如果访问强大模型变成了商品,真正的优势将是围绕它们的学习循环。

这就是 Oumi"复合 AI 工厂"背后的大想法:公司不应该只是租用智能。他们应该积累它。

引用 Manos Koukoumidis @Koukoumidis企业AI面临矛盾困境:既要差异化竞争又租用相同模型;需求专业化却依赖通用模型;为大模型支付高价但仅用1%功能;要求控制主权却租用核心业务智能。Oumi认为下一代企业AI应由企业自建、拥有并持续改进。企业应通过自有数据与经验构建专业智能作为竞争优势,而非租用通用AI。查看被引原帖 ↗
查看英文原文
There is a strange problem with enterprise AI right now: companies spend millions integrating the same models their competitors can access five minutes later.

That may be useful, but it is not a moat.

What Oumi launched today is interesting because it treats AI less like software you deploy once and more like a system that should learn from the work it actually does.

Build a specialized model, deploy it, see where it fails, turn those failures into training signals, improve it, and deploy it again.

A loop, not a launch.

Companies retain access to the models, data, evaluations, and recipes created along the way. Even coding agents can operate the workflow through Oumi’s CLI.

If access to powerful models becomes a commodity, the real advantage will be the learning loop around them.

That is the bigger idea behind Oumi’s “compounding AI factory”: companies shouldn’t just rent intelligence. They should accumulate it.
◔ 4.7 万 次浏览(4 条合计)♥ 108⇄ 2▶ 含视频观点看原帖 ↗
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

Brad Lightcap 是 OpenAI 最重要的非技术型建设者之一,工作八年后离职创业。他 2018 年加入,建立了 OpenAI 的大部分运营体系:财务、法律、人事、安全、市场销售、政府关系和合作。作为首席运营官,他帮助把一个小型研究实验室转变成为近 10 亿用户服务的公司。今年 4 月他从首席运营官转到负责复杂交易和投资的特别项目。为什么离职?Lightcap 说他一直在思考"当 AI 进入下一阶段时世界需要做对的一些重要事情"。这是巨大的损失。祝他一切顺利!

引用 Brad Lightcap @bradlightcapOpenAI 团队成员宣布离职创业。他加入 OpenAI 已 8 年,见证公司从研究实验室发展为最具影响力的公司之一,构建了多个关键部门,服务超 10 亿用户。虽然离职,但对 OpenAI 团队能力和未来充满信心。查看被引原帖 ↗
查看英文原文
Brad Lightcap, one of OpenAI’s most important non-technical builders, is leaving after eight years to start something new.

He joined in 2018 and built much of the machinery behind OpenAI: Finance, Legal, People, Security, GTM, Government and Partnerships. As COO, he helped turn a small research lab into a company serving nearly one billion users.

In April, he moved from the COO role to special projects involving complex deals and investments.

Why leave? Lightcap says he has been thinking about “a few important new things the world will need to get right” as AI enters its next phase (whatever that means).

Big loss. I wish him all the best!
◔ 4.8 万 次浏览(3 条合计)♥ 184⇄ 4动态看原帖 ↗
v0@v0 · 公司官方 · 1 天前

v0 新增了侧边栏功能。• 聊天按项目分组 • 实时显示流式传输、等待和就绪状态 • 悬停卡片预览网站、已修改文件和 git 分支

查看英文原文
v0 has a new sidebar.

• Chats grouped by project
• Real-time status indicators for streaming, waiting, and ready states
• Hover cards with a website preview, changed files and git branch
◔ 2.1 万 次浏览♥ 211⇄ 10▶ 含视频新品看原帖 ↗
OpenAI Developers@OpenAIDevs · 公司官方 · 1 天前OpenAI 开发者平台官方

ChatGPT 桌面应用现在可以预览使用了,支持以下 Linux 发行版的桌面版本:

• Ubuntu 24.04 LTS 和 26.04 LTS
• Debian 13
• Fedora 43 和 44

通过 .deb 或 .rpm 包安装,支持 x64 或 ARM64 架构。

openai.com/codex

查看英文原文
The ChatGPT desktop app is now available in preview for desktop variants of these Linux distributions:

• Ubuntu 24.04 LTS and 26.04 LTS
• Debian 13
• Fedora 43 and 44

Install with .deb or .rpm packages for x64 or ARM64.


openai.com/codex
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
连环推 ×2

Google I/O 大会

来中国了…

有在现场的兄弟吗?

Tinyfool@tinyfool · 中文博主 · 1 天前

一分钟快7000万,真是燃烧啊

Dan Shipper 📧@danshipper · 博主 · 1 天前

这张地图上有个明显的遗漏👀

引用 Hassan @nutlopeAnnouncing the NYC AI Atlas! An open source 3D map of the top AI startups in New York City. nycaiatlas.com查看被引原帖 ↗
查看英文原文
there is one glaring omission on this map 👀
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

又有羊毛薅了,Manus 从 Meta 剥离以后开启了限免。

如果你之前自己买过 Manus 会员,或者是通过 Lenny's Newsletter 领过会员的话,最近可以用一用。

仅限 Manus 1.6 和 Manus Lite 两个模型

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
连环推 ×2

小红书发布了他们的 Read Skills Top 100 排行榜,这个有专门的页面,而且会自动更新。

结果发现,藏师傅居然有两个 Skills 在前十的位置,其中一个居然排在第二!

他们这个也挺有意思的,排名第一的居然是一个占星应用。

Tibor Blaho@btibor91 · 博主 · 1 天前逆向挖掘 AI 产品代码的爆料专家

新功能:ChatGPT 赠送额度 - 符合条件的用户可以购买额度作为礼物送给他人,然后分享一个一次性领取链接,收礼人可以将额度兑换到个人 ChatGPT 账户

查看英文原文
New: Gifting credits in ChatGPT - eligible customers can buy credits as a gift for another person, then share a one-time claim link that the recipient redeems to a personal ChatGPT account

(
help.openai.com/en/articles/…
)
ollama@ollama · 公司官方 · 1 天前本地跑大模型的热门工具

GitHub Copilot 在 JetBrains 里现在支持 Ollama 了。

查看英文原文
You can now use Ollama as a provider in GitHub Copilot for JetBrains.


github.blog/changelog/2026-0…
Lisan al Gaib@scaling01 · 博主 · 1 天前高频 AI 模型测评与爆料博主

OpenAI:我们增强了 GPT-6 推理效率
GPT-6:已完全非语言化
那还费劲说话干什么呢,不说话不是更高效

查看英文原文
> OpenAI: we increased reasoning efficiency of GPT-6
> GPT-6: completely non-verbal
> why say any words at all wen no words do trick
DeepLearning.AI@DeepLearningAI · 公司官方 · 23 小时前

AI 编码代理默认带着一些设置:用哪个模型、要花多少钱、什么数据会离开你的电脑。你可以把这些默认设置改成自己选择。

我们的新短期课程《AI Coding Workflows: From Cloud to Local》是与
@JetBrains
联合开发的,由 JetBrains 开发者倡导 @paulweveritt 讲授。你会在云端、混合模式和完全本地的环境中重建同一个应用。

过程中你会学到怎样在子代理之间分配工作、用更便宜的模型来处理日常任务,最后用在你自己机器上运行的模型来做最后的工作。

免费报名:
hubs.la/Q04sM4rQ0

查看英文原文
Your AI coding agent comes with defaults: which model runs, what you pay, and what leaves your machine. You can turn those defaults into choices.

In our new short course, AI Coding Workflows: From Cloud to Local, built in partnership with
@JetBrains
and taught by
@paulweveritt
, Developer Advocate at JetBrains, you'll rebuild the same app across cloud, hybrid, and fully local setups.

Along the way, you'll split work across subagents, put cheaper models on the routine tasks, and finish with models running on your own machine.

Enroll for free:
hubs.la/Q04sM4rQ0
◔ 1.5 万 次浏览♥ 129⇄ 18▶ 含视频教程看原帖 ↗

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