Joined July 2016
Couldn't help myself
313
922
209
15,556
1,366,226
鲁晓宇 retweeted
Tibo
@thsottiaux
Sep 12
Hi Astra users. A reset and a quick update on quality issues that have been posted around. Working with some of you, we have found and fixed the following issues: - Some skills written for previous models were triggering too often or preventing the model from checking its work. - An opt-in context management experiment that could cause early stops or replies to older messages. We've disabled it. Our rough estimate is that 4-5k users were affected by this experiment. - We've also removed some badly configured engines that resulted in a measured quality degradation for a long tail of traffic flowing through them. We’ve also made some more minor improvements and things should feel significantly better across the board. More consistent follow-through, better tracking of your latest message, and better checks on the work as it’s going through the motions. The examples posted and all the users who worked directly with us were incredibly useful in helping fix things quickly. Always grateful for this incredible community. And of course, a reset is also landing by midnight today.
3,951
1,611
1,970
28,358
8,595,888
鲁晓宇 retweeted
GPT 6 Astra 成功通关了“我不是机器人”游戏的所有 48 个关卡 所有网站的验证码防护都失效了...😄
314
147
50
1,563
268,112
鲁晓宇 retweeted
DeepSeek Harness 用了半天,我已经快要跟这个玩意的作者对齐了 我来给你们讲讲他的心路历程 首先他认为,现代软件是代码的产物,所以一个能在运行中改自己代码的现代软件,就可以实现自进化 在这个基础上,如果改炸了怎么办呢?做成可插拔,所以要在现代软件里面引入天然的隔离墙 打开之后是网页,网页里面带个 coding Agent 干活的区域,然后用 coding Agent 修改自己所在的环境,改炸了就回退隔离墙里面的东西,这不就自进化了吗? 然后这个时候底下有人举手了,“老大,这个好像叫 vs code 和它的内置浏览器” 哥们一拍大腿,“对,我们就是要干翻 vs code,一切皆软件,vs code 那不就是 Harness 嘛”,遂开发并写作论文 考虑到 vscode 叫 extension,为了避嫌,他把这个可拆卸组件叫 plugin(可能也有别的原因捏比如从 Koishi 里面带下来的,反正无所谓了) vs code 桌面端支付了很多适配债,他不想支付,反正 vscode 自己也是 electron,那我开个网页不就得了 随着身体一阵抽搐, DeepSeek Harness 魔童降世,开启了属于自己的青春氛围马具物语 --- 后记: 就事论事的说,让一个coding Agent 热更新自己的组件和前端,是不是有新意?当然是有的 但是,既然都这样了,为什么不直接用 coding Agent 写网站或者写脚本给自己用、而是一定要把简单的模式包装在 Plugin 这个超级名词里 事实上,dsh的这种结构论亲缘关系跟 Notion 很近。而 Notion 做的并不好 顺便一提,就连 DeepSeek 本家的模型都有点搞不清 plugin 该怎么用,经常会跑去直接改自己的代码,毕竟这个快,效果也差不多 本文不含任何推广和情绪,纯粹是用了两小时的体验
163
68
14
926
149,742
鲁晓宇 retweeted
Kimi Code 实验性功能 /tower: 多个 agent 同时维护 1 个 repo,互不“打架”。 怎么理解🤔:想象一下 PM 找不同的人沟通事情。沟通时间、冲突情况都是不保证的,但沟通内容都是 team 内的事情。内部有个塔台(tower)来分发任务、协调冲突、指定 reviewer、合并代码、清理残留。 求体验求拍砖,PR 在这👉: github.com/MoonshotAI/kimi-c…
11
5
2
90
18,073
鲁晓宇 retweeted
Kimi Code Changelog 0.36.0 Upgrade the experimental subagent model setting to a model pool: the [secondary_model] section can now hold a set of candidate models with descriptions, and the main agent picks from them per spawn based on the task.
20
29
5
513
47,720
鲁晓宇 retweeted
Elon Musk
@elonmusk
Aug 10
Replying to @ARKInvest
China is awesome. I strongly encourage people to visit.
4,076
2,131
1,163
29,091
6,048,231
鲁晓宇 retweeted
Kimi K3 is now live on @databricks!
Moonshot AI's latest open-weight model, Kimi K3, is now available on Databricks through Unity AI Gateway. @Kimi_Moonshot Run Kimi K3 where your data already lives - governed, secure, and ready for custom AI apps and agents built with the data in your Lakehouse. Unity AI Gateway lets you deploy Kimi K3 with enterprise-grade access controls. Govern every call, monitor performance, and scale securely across your AI apps. Test Kimi K3 alongside other frontier models without changing your application code. The open-weight frontier just arrived on Databricks. Try it today. databricks.com/blog/kimi-k3-…
51
60
8
1,225
140,371
鲁晓宇 retweeted
Kimi Delta Attention. — a minimal implementation of the Kimi K3 architecture. k-a.in/KDA.html
14
89
2
865
46,905
鲁晓宇 retweeted
AI at the final frontier 🚀
Elon Musk
@elonmusk
Aug 4
SpaceX has committed to using Nvidia GPUs exclusively because they are the best
870
1,458
165
25,359
1,701,690
鲁晓宇 retweeted
实测下来,Kimi 是真开放: 它的 Computer Use 走 MCP,能对接任意 harness。 我用 DeepSeek v4 Flash——连视觉都没有——照样用Kimi Computer Use跑通 完成任务。 安装稍麻烦:需要在Kimi Code 里敲 /plugin 装上,系统立刻多一个 Kimi CU 的 app,还能一键装给 Claude Code 和 Codex,其他 agent 也给了安装教程。 反观 OpenAI:Codex CLI 嘴上开源,context 压缩却是黑盒;Computer Use 更绝,只许自家 Codex app 调用,其他一律没门。 谁在开放,谁在圈地,一比就知道。
Kimi 放大招了!刚发布了两个重要插件, /plugin 命令安装, pi Agent 里应该也能用。 简单说: . WebBridge ≈ Codex 的 Chrome 插件 . Computer Use我实测支持后台操作 + 虚拟光标,走 MCP 通道,强 最狠的是它不止在 Kimi 里能用——Claude Code、Codex app 这些 harness 都能接,Pi Agent 大概率也行。 感谢Kimi!
19
45
2
289
55,274
鲁晓宇 retweeted
📣 Kimi K3 is now available in GitHub Copilot for @code! Try Kimi Moonshot's latest open-weight model for agentic coding, now hosted by @FireworksAI_HQ. 📖 Learn more: github.blog/changelog/2026-0…
109
347
36
3,204
1,609,459
Kimi K3 scores nearly TWICE Claude Fable 5 on Harvey LAB-AA's hard autonomous legal tasks
18
17
2
171
30,298
鲁晓宇 retweeted
Glenn Diesen
@Glenn_Diesen
Aug 5
I cannot remember the United States ever apologising for "overcapacity" when it was the world's leading industrial power. Should Italy apologise for its overcapacity in fashion, France for its overcapacity in wine, the UK for its overcapacity in financial services, or Belgium for its overcapacity in chocolate? *** The real challenge of our time is that centuries of Western dominance have come to an end, and there is a need to adjust to a multipolar world.
302
1,998
135
7,983
248,469
鲁晓宇 retweeted
Tomorrow will be my last day at Google after 27 years, and watching it grow from 25 people to 190,000+ has been an amazing journey. Below is a note I shared with many people internally at Google today. An excerpt is: It has been an absolute pleasure to work with you and to help build some of the most widely used and impactful products of all time. As a kid, I dreamed of helping build software that would be used by many people, and Google now has thirteen products used by more than a billion people (amazing!). Our work has had a tremendous impact in the world, and I have been lucky enough to collaborate and form friendships with many colleagues that I deeply admire, respect, and enjoy. It still brings me joy every time I see people out in the world using our products to find information, handle email, translate documents, watch videos, learn new things, navigate and understand the physical world, browse the web, use their phone, run large-scale computations on our infrastructure, ride in an autonomous vehicle, or perform complex tasks with the help of our AI systems. I hope you all share this sense of joy, because it is a shared accomplishment! Thank you to all of my colleagues at Google over many years! Now I'm excited to go start @DiscoLoopAI with my longtime friends and colleagues @Sanjay_Ghemawat, @OriolVinyalsML, and @quocleix. (Updated post: slightly redacted to not have some personal info)
846
2,764
757
33,731
11,896,062
👏Free👏Kimi K3 has been extended through August 12! Quick setup: → tokenrouter.com → get API key → model: moonshotai/kimi-k3-free → base URL: api.tokenrouter.com/v1 → real access to a top-tier model
More GB300 online. The Kimi K3 free tier just got bigger, and it runs through Aug 5. Try it now: tokenrouter.com/models
51
71
9
836
5,269,427
鲁晓宇 retweeted
A Brief Look at K3's MoE and Attention kexue.fm/archives/11848
1
63
3
401
33,520
鲁晓宇 retweeted
BREAKING: Anthropic CEO Dario Amodei is concerned that the download button of kimi k3 is still operational.
71
86
10
2,757
203,335
俺上电视啦! 聊聊 kimi-k3 这么猛为什么还要开源 上周有幸接受了深圳卫视-科创最前沿栏目的采访, 主要聊了聊 Kimi-K3 这次取得如此牛的成绩背后的意义. 以及SOTA级别的模型就这样开放权重了, 不怕别人抄吗? 在我来看是真不怕的, 想象一下一个武林高手, 把自己的秘籍写成书, 免费送. 那是不是就意味着这些武林秘籍不是谁都练的. 大模型的情况也是这样的, 虽然权重也开源了, 训练技术也开源了, 但是, 最重要的, 训练使用的数据和经验, 这个是没开源的. 举个最典型的例子, 现在个人拿到kimi-k3的模型权重, 然后自己租算力微调, 大概率是会把模型越调越拉跨的. 就是因为既没有优质的后训练数据, 也没有优秀的后训练能力. 而现在大模型训练的数据其实大部分都是AI合成的了. 护城河已经从“静态的代码和模型权重”转移到了“持续的进化能力和数据飞轮”了. 所以即使全面复刻了 kimi-k3 的技术架构, 但是没有高质量合成数据流和持续的后训练体系, 也是没有办法竞争的. 但是 kimi-k3 开源获得的收益却是事实存在的, 不但能获得超高的曝光, 而且还能提升自己的社区影响力, 并且还能收获社区的正向反馈. 开源出来的训练工具链, 社区优秀的工程师在使用中也会不断的帮助反馈和提升. 形成整个生态的进步. 得道者多助这个道理是不会变的. #kimik3 #深圳卫视 #科创最前沿
54
6
126
17,902