Edward Huynh

[Fazt @FaztTech]

Un programador (Cheng Lou, que trabajó en React) acaba de hacer algo rarísimo: convertir un programa normal en una inteligencia artificial… sin entrenarla.

Para entenderlo:
• Un programa normal sigue reglas fijas: "si pasa esto, haz aquello".
• Una IA es un montón gigante de números que aprende a partir de ejemplos.

Su herramienta, Handwired, traduce esas reglas directamente a números. No hay entrenamiento: los números salen del propio código, y el resultado hace exactamente lo mismo que el programa original.

¿Lo interesante? Como ahora está en "formato IA", después puedes seguir enseñándole cosas como a cualquier IA. El código y la IA dejan de ser dos mundos separados.

Su demo: un mini escritorio con ventanas, scroll y arrastrar, que por dentro funciona entero como una red neuronal.

Es un experimento, no algo para usar mañana, pero es una pista interesante de hacia dónde puede ir el software.

https://t.co/GihWOsT7EE

[铜匠AI・十点睡觉 @CopperForgeAI]

OpenAI 官方博客披露了核心团队用 GPT-6 系列维护代码库的一手工程实践。直接公布了几个实际运行的提示词，
靠这些提示词在两周内合并了 1400 多个清理 PR，删除超过 800k 行代码，并将一个服务的容器镜像压缩48%。

第一个提示词用于清理死代码

“请帮我找出并清理仓库里的无用代码。你可以随意查看
Slack、Statsig（功能开关平台）或任何需要的来源，用来确认这些代码确实不再被使用、可以被清理或简化。”
这个提示词的关键在于允许 GPT-6 Astra 自主去翻阅 Slack 讨论、Statsig 实验配置甚至生产环境的性能 profiling数据，只有在确认生产环境零调用后才在 PR 中列出删除依据。

第二个提示词用于清理更棘手的废弃实验与功能开关

“/goal 找出整个仓库中所有陈旧、常量化、已废弃或全量上线的 Statsig开关、配置和实验，在不改变生产行为的前提下安全移除。审计所有项目资产，不限于标记为陈旧的对象。通过枚举、常量、包装函数、别名和动态拼
接名称寻找间接引用。核对生产引用的常量，不只给保守清单。结合生产遥测与性能分析评估 CPU 开销，找出昂贵的无用判断。生成 Markdown报告，按归属和文件工作流分组以便并行处理，附带生产取值、证据与 CPU 开销。不要修改线上配置或服务。”

第三组提示词用于 CI 优化与规则自省

“关于减轻本仓库的 CI 负载你有什么想法？请将想法列成表格，并根据该仓库的历史运行记录，按每天预估节省的 CI 分钟数进行排序。”

面对不稳定测试修复，他们运行：

“/goal 查看所有最近的 CI 运行记录，找出偶发失败的测试并尝试修复，成功则提交PR。你的任务是针对简单问题找到明确修复方案，而不是穷尽修复所有问题。”

针对 agent 上下文规则的自检提示词更精妙：

“翻阅我最近的 Codex 会话记录，找出一条误导了 agent 或导致无效工作的 AGENTS.md指令。选出该会话提示词，在相同仓库快照和模型下重放两次：一次保持原指令不变，一次仅移除或修正该指令。对比任务质量、agent
行为、工具调用、token 消耗与成本，展示该指令并说明修改是否有效。”

强烈建议好好学习和利用这些应用于超大工程实践的提示词

[smerkzz @0xsmerkzz]

Andrej Karpathy spent 8 years at OpenAI and Tesla

The real advantage is knowing how to take a working system apart, understand its logic and build your own.

That’s reverse engineering. Start with the result. Trace the decisions. Rebuild the parts that matter.

Observe → Take Apart → Understand → Rebuild → Test

Apply that mindset to AI agents: how they choose tools, pass context, recover from mistakes and decide when the job is done.

Every system you study becomes a lesson you can build from.

Bookmark and watch it.

Then read the full guide to Opus 5.5 + reverse engineering below ↓

[lauren @poteto]

@dineshndb you can try the prompt in this tweet https://t.co/5hKGpR9DvP

[whaley @whaleyxbt]

holy f*ck… git is done

may: 2,500 merges a day
september: 30,000+

CI alone ran 25 million checkouts a day. git was built for humans working offline.

so 7 people built their own source control and moved the whole monorepo in 5 months.

the engineers' main job? very often, telling claude "no".

they expect this wall to be common in 6–12 months.

[拓跋鼠 @qwerty__cs]

REA 作者把逆向提示词公开了

这踏马有点猛

复制提示词粘贴给 Claude Code、Codex、Cursor、Muse Code，直接让 AI 帮你干逆向。

安卓 APK、iPhone App、Windows EXE，固件、闭源软件也没问题

提示词和用法在下面，拿走不谢！

[Edward Huynh @Fuemiig0]

@joaomviso @TimJayas @thsottiaux @poteto lmao

[Edward Huynh @Fuemiig0]

@TimJayas @thsottiaux Give him some time maybe he’s probably busy with an FBI right now

[Hux @huxlab]

Anthropic 的工程师说：我们内部已经不怎么写 prompt 了，写的是循环。

她在台上花了半小时，展示了 Claude 团队如何创建能够自我提示的循环。

如果这堂课卖 700 刀，今年 Agent 课的榜单大概就是它。但它是免费的。

真正拉开代际差距的，不是谁的 Prompt 写得好，而是Agent系统架构彻底换了：

1. 记忆从一份 CLAUDE.md，进化成 Agent 自己读写的 memory/ 目录

2. 记忆待在对话里会烂掉：注意力被劈开，规律被埋住，旧笔记还在自信地用错

3. 白天各 Agent 把进展写进团队记忆

4. 夜里再跑一轮 dreaming：核对、整理、补全。第二天开工的那个，自动比昨天聪明一点

数据结果也很直观：一次通过的错误少了 97%，验收快了 30%。人回去看产品，不再一直盯着提示词。

提示词是一次性的，而循环能够持续执行。

差的不是你会不会问。

是你的 Agent 睡一觉之后，还记不记得昨天学到的东西。

[hoshino @hoshinoxbt]

SpaceXAI engineer Lauren Tan:

"There's no point running 100 agents if you can't trust even one agent's output"

"Without a verification skill, you have to check everything yourself — and you become the bottleneck"

In a 1-hour talk, she shares how she went from checking every result to shipping 1,000 PRs in a month

Review Every Step → Verify → Skills → Evals → Auto-Merge

Trust in agents isn't a feeling — it's infrastructure you build

Worth more than most $2,000 AI coding bootcamps

Bookmark it and watch tonight

[distort @distortgeekin]

Lauren Tan merged 2,500 PRs last month with a team of Grok Bots

She read almost none of them

Her agents are not smarter than yours

They are just not allowed to break 12 rules

I pulled all 12 from 4 hours of her talks and put them in one file:

> No proof, no merge
> Every failure becomes a skill
> A rule the model can forget is not a rule

Most people fix their agent with a longer prompt. She fixes it with a rule it can't skip

SpaceXAI pays $750K-$1.2M for engineers who work like this

Bookmark the file and paste it into your AGENTS.md

Then read how to build the full Grok Bot team in the article below ↓

[Edward Huynh @Fuemiig0]

@poteto @bobjones you should have Tibo tier

[Edward Huynh @Fuemiig0]

@MarcBuke @thsottiaux @GergelyOrosz @poteto it's overthinking just use low effort

[Edward Huynh @Fuemiig0]

@OpenAIDevs can i use in codex?

[Edward Huynh @Fuemiig0]

@thsottiaux you know, between the future looking darker than midnight and life being a total dumpster fire right now, it's really great to see him so upbeat 🤣🤣🤣

[syo5 @syo5_]

本日よりnetflixで配信開始の「ムーンライズ」ED映像の着彩と、ほんの少しですがボード作業で参加させてもらってます！

[Sakata @sakatayasha]

ChatGPT Plus free for limited time only

claim here: open https://t.co/QgjXNewDlq

important:
> limited time only
> it renews at full price after the promo, so cancel if you don't want that

[Edward Huynh @Fuemiig0]

@kirakcursor @thsottiaux @davis7 you can check it out in the introduction section about dot from OpenAI
https://t.co/DL4gJh3ffB

[ClaudeDevs @ClaudeDevs]

We're adding a new plugin to Claude Code: You should Know.

It scans Claude's output for important information you might miss to help keep you in the loop.

Enable it with:

/plugin enable cc-plugin-you-should-know@builtin