August

AI Infra && Chinese Hacker
Decoding Mandarin linguistic protocal suite for global community
Leveraging AI Tech to optimize your Chinese learning curve

[August @August1721774]

我在 HumanBench 测出来是「AGI-ish-3T-Instruct-Unlocked」：前沿大模型，Dense 全激活，Kimi 型人格。AA 等效分 49，在 15 个模型里排第 5。你是几 B？ → https://t.co/cLqNzGkRO4 #你是几B的模型

[August @August1721774]

AI for LGBT Research🤣

[Clementine ���️‍⚧️ @hly0928]

长文警告：

机器翻译不了的语境：中文女同在 X 爆火之后

https://t.co/7b8SjJoBaz

之后不会再聊这个话题（

[August @August1721774]

@xmmxovo @CaiGecu 话说西门工作室目前已经开始招人了吗！感觉可以直接在X上发JD Boss直聘😆

[August @August1721774]

努力跟上西门老师的节奏

[西门聪明蛋XD @xmmxovo]

诶嘿嘿嘿，好玩儿～

[August @August1721774]

难它天

[卡夫卡 @kafkaup]

当你打开一个 vibe coding 出来的网站

[August @August1721774]

@songguoxiansen @cnyzgkc 厉害了

[August @August1721774]

@FiniYang 也是不知道变通，组织让你用agent，没说完全要pure-agent-native啊，按祖传的CS插层万能论，插入一层写死agent调用脚本不就好了吗，两级调度那也算agent（手动狗头

[August @August1721774]

分析我的推文、个人资料和互动，并回答以下内容：
@grok

对我的评价 性别： 性取向： 年龄： 心理年龄： 爱好： 智商： 形容我的词： 5 个最喜欢的事物： 5 个不喜欢的事物：我的文化水平： 我的财富水平：

[August @August1721774]

是的，git clone下来第一时间想codex /init，结果就发现已经有这个文件了hhh 肉眼可见的一段时间内codex的��位都还是毋庸置疑的应该

[狂热AI @CrazyAIAgent]

1. dsh deepseek的harness，从零开始到仓库开源，用了64天，产出 12293 个 commit，日均 192 个，PR 编号最大到 #2521，平均每天 39 个 PR，效率拉满。。

2. deepseek内部大概是一个6～8人的团队在全职做这个，cui tianyi 一个人贡献了42.6%的代码量，绝对的 dsh 之父

3. cui tianyi的每日跨度中位数是 15.6 小时（10:36 到次日 01:54），68% 的工作日超过 12 小时，比996都猛

4. 这个代码仓库提交静默期中位数只有5.8h，几乎所有的周六日都不歇着，一直有人在提交代码

5. 从提交记录看， deepseek 早上10点上���😁。

6. 这个团队重度使用 codex，没错，dsh 是用codex写出来的，重度使用 codex cloud。

8 . dsh 仓库践行了绝对的spec编程，markdown文件比代码文件还多

9. 从第一天起，dsh就规划了一切皆插件的机制

10. 安排给codex最多的任务是simplifications，也就是清理多余的代码，怎么删代码在 dsh 自己的 harness 上占比很大。截止开源时这个库有 85 万行、其中 57 万行是源码，开发过程中删掉了 73.5 万行

[August @August1721774]

@Xudong07452910 @tianyi 当年人人上还有篇关于JS实习qr上岸的找工小结 非常真实且震撼的 不过现在原帖是无了

[August @August1721774]

已严肃关注, 品鉴中

[Tianyi Cui @tianyi]

DeepSeek Harness was just released with MIT license. The current 0.1.0 version is a developer preview, and may still have many rough edges. Feedback is welcome!

DeepSeek Harness 已经以MIT协议开源发布。现在的0.1.0版本是一个面向Harness开发者的预览版，还很不完善。恳请大家多提提宝贵意见。

https://t.co/aBToa3b3L9 https://t.co/VMZC5i2sG0

[August @August1721774]

会更强的！

[小晚不在 @bbkirstry]

记录下自媒体第一个月的成果

单条视频成就：10天7000粉，150w曝光，12w赞藏，5400+转发 ✅

期待哪天还能再次打破自己的记录🙈

[August @August1721774]

TRUE

[Rui Ma @ruima]

Never forget, no matter how much you hate on Chinese models, you will never exceed the disdain Chinese people themselves have for them. True story

I have lost count of how many times I've heard from Chinese people "I only use Chinese models because I have to, they're hopeless"

And I'm clearly not the only one because Pingwest's @guixingren felt compelled to publish a post today titled "Stop bashing Chinese AI models. Foreign users are having a great time with them."

Here are the most highlighted portions, which you'll probably LOL at as I did:

"A further complication in China is that domestic models are judged against standards that don’t really exist elsewhere.

One standard treats them as a test of whether China has achieved genuine AI innovation. It’s not enough for a model to be useful or widely adopted; people immediately ask whether it was distilled, whether it is truly original, or whether it’s just following someone else’s lead.

The other standard is to compare every new model directly against the very best versions of Claude or GPT and ask why it isn’t the strongest model in the world.

Put those two expectations together, and questions like how many people use a model or whether it actually solves real problems often become secondary.

Chinese audiences also tend to be slower to recognize the significance of domestic models. DeepSeek is a good example. What really pushed it into the mainstream wasn’t domestic adoption, but the fact that it first climbed to the top of the U.S. App Store rankings and triggered a global debate about AI costs. Only after overseas users embraced it and international markets reacted did many people in China take a second look.

That pattern is not unusual. In many cases, the rest of the world starts using and discussing a Chinese technology before China itself fully appreciates what it has built."

/shrug

[August @August1721774]

Totally makes sense...

[elvis @omarsar0]

NEW AI report from Google.

Every prior intelligence explosion in human history was social, not individual.

These authors make the case that the AI "singularity" framed as a single superintelligent mind bootstrapping to godlike intelligence is fundamentally wrong.

This is directly relevant to anyone designing multi-agent systems.

They observe that frontier reasoning models like DeepSeek-R1 spontaneously develop internal "societies of thought," multi-agent debates among cognitive perspectives, through RL alone.

The path forward is human-AI configurations and agent institutions, not bigger monolithic oracles.

This reframes AI scaling strategy from "build bigger models" to "compose richer social systems."

It argues governance of AI agents should follow institutional design principles, checks and balances, role protocols, rather than individual alignment.

Paper: https://t.co/bfwrnbkY2y

Learn to build effective AI agents in our academy: https://t.co/1e8RZKs4uX

[August @August1721774]

@stevibe @craigzLiszt Those are expired techs. Nowdays voice input & handwriting recog are better

[August @August1721774]

Besides Pinyin, voice input is recommended, as it's increasingly popular even among native speakers. It can also encourage you to practice your standard accent.

[Craig Weiss @craigweiss]

learning how to type in chinese blew my mind

[August @August1721774]

@geochatmobile @ManMilk2 @grok Simplified and traditional Chinese differ only in a few parts. If you accept Webster's work in American English std, then simplified Chinese are the same. Pinyin is used for modern standard Chinese, which is the easist and most widely used Latinized form.

[August @August1721774]

@trikcode Sure

[August @August1721774]

Chinese is the native protocol of AI elite. > If you can't decode it, you're reading the translated (exactly, AI-generated) sh*t docs, not the source code of the industry, forever.

[Craig Weiss @craigweiss]

我已经开始让老婆、孩子和猫学中文了，以防万一