kotoba

[kotoba @liurcxx]

@PhonetoFBRR @nosugarnoBS @MaxForAI claude一直都很nice，只是达里奥犯畜

[kotoba @liurcxx]

@leo23kingx 几天解封的？

[lauren @poteto]

i love writing. i never let agents write for me. whether it's docs, or tweets, or shitposts, or even prompts (okay maybe that one's harder as agents manage other agents)

there's a simple reason for that: i still strongly believe that writing is thinking. it's a skill i hope to never give up. it's also important to me that whatever i say is always authentic, mistakes and all. my favorite posts are the ones that i just full send without any editing

i suspect it also helps with prompting. being able to clearly express what i want, and knowing the vocabulary i need to be able to direct my agents does seem to make for more effective usage of agents. in the same way that engineers have an advantage when we prompt agents to write code, because we are intimately familiar with the words that make up the discipline

knowing the right word incantations to get an agent to produce a desired goal almost feels like spellcasting. as models get more capable, and that input box looks more and more like a magic box that can output anything, words become the interface between you and the genie

there may come a time when we give up the keyboard in exchange for brain-computer interfaces. but until then, i'm writing every word.

[酱紫表 @pengchujin]

如果觉得自己用的 GPT6模型降智了，可以测试一下。https://t.co/Cb9DvsgpdS

[山羊赚钱笔记 @shanyanggm]

卧槽兄弟们，GitHub 上真的有一堆免费到离谱的项目。

很多能力已经能直接干掉你正在月付的软件。

1. TradingAgents
AI 多 Agent 量化交易框架
https://t.co/3T1sPPQ6TL

2. LibreChat
一个界面接入 ChatGPT、Claude、Gemini 等多个模型
https://t.co/HZJQEij5h1

3. HyperFrames HeyGen
开源的视频生成引擎
https://t.co/o0OnS8iJMG

4. Fincept Terminal
开源版彭博终端
https://t.co/n5COGtvHc4

5. MoneyPrinterTurbo
AI 一键生成短视频 https://t.co/Po8sGObI7q

6. Agentic Inbox Cloudflare
开源 AI 邮件助手
https://t.co/S7XDWi1yTD

7. VoxCPM
AI 声音克隆工具
https://t.co/wkSGMchuq2

8. Flowsint
开源 OSINT 情报分析工具 https://t.co/RwFvhurB0P

9. agent-skills
Claude Code 技能库 https://t.co/c2NifSzKM1

10. Nango
开源 API 集成平台
https://t.co/tNbwrMEzZn

兄弟们，这些真不是玩具项目。

很多你还在交月费的软件，GitHub 上已经有人做出开源平替了。

一句话：

别只会收藏 AI 工具网站了。

真正狠的东西，很多都藏在 GitHub 里。

[桃子🕊️ @taozi0929]

刚看到个事儿，一哥们被某个爬虫服务收16刀一个月，气得直接自己写了个开源的，扔到GitHub，现在五万一千星。

叫crawl4ai。

就是你把任何网址丢进去，它给你吐出来干净的Markdown，LLM直接开吃那种。

不用key，不用注册，不按页算钱，跑起来比我用过的付费的还快。

这事儿真正有意思的地方吧，不是技术多牛，而是——爬虫的门槛早就没了，还在收你月费的才是真暴利。

我反手试了一下，速度确实离谱。

那种“付费墙背后才是完整信息”的幻觉，当场碎了一半。

收藏下面的链接，又可以省一笔钱：
https://t.co/u0gIWT2AmV

[蜘蛛侠 | 1000X GEM @zhizhuxia22]

我最近发现个超级牛逼的网站，简直就是数字游民的福音

它直接把全世界城市的：
房租、物价、医疗、安全、气候、税率、宜居度

全部整理到一起了！

哪个城市适合长期住
哪个地方生活成本更低
哪个城市更适合远程办公
甚至连一个月大概要花多少钱，都给你算得明明白白
以后想换个城市生活、旅居几个月、或者去国外待一阵

真的不用自己到处查资料了
直接拿这个当城市对比工具就行

传送门：https://t.co/XAIp4CJeyZ

[数字生命卡兹克 @Khazix0918]

给大家在AIHOT上更新一个被催了无数遍的小功能：Tibo重置监控。

以后大家就不用怕Tibo重置Codex然后看不到浪费Token了：
日历查记录：全员重置、发重置卡分开标记，统一换算成北京时间，帖子附中文译文和原帖链接。
群里收通知：监测到新的重置、发卡预告或确认消息，会自动推送到群里。
开放 API：无需 API Key，返回 JSON，包含历史记录、北京时间和原帖链接，可以接入自己的工具。

👉 Tibo重置日历：https://t.co/4zdZZoFUJ0
🔌 API 接口：GET https://t.co/w3uCu7zzr5
📖 接入说明：https://t.co/0gd0Q3BtHM

希望大家少刷几次推，多领几次卡～快乐Coding～

[kotoba @liurcxx]

@TJ_Research 链上openai和A社跌了几个点，其他硬件股没什么反应，谷歌甚至还涨了

[Andrey Melnikov @mylandros]

A few questions on this.

Did humanity appoint you as its representative?

Do you represent the values of all humanity, across cultures and societies? Or this is narrow values of part of SF only?

Is splitting the world into “us” and “them,” labeling “them” authoritarian, then asking for international cooperation a sensible opening to negotiations?

Can governments recruit enough frontier talent to independently judge the labs they regulate?

Will they even understand this complex subject?

Do a few separate incidents justify an industry-wide slowdown, without showing why narrower remedies would fail?

Will countries accept rules explicitly designed to preserve another bloc’s technological superiority?

Will complex certification requirements help newcomers compete, or rather protect incumbents?

Does slowing model development also slow your revenue growth, market expansion or move into vertical services?

Will restrictions on public access come with equivalent restrictions on your own commercial use of powerful internal models?

What commercial sacrifices have your board and investors approved?

Do your IPO forecasts reflect those sacrifices?

Will you pace your fundraising and IPO too?

Does concern about safety eliminate commercial self-interest?

Some answers seem obvious. The others deserve a public answer.

Still “0” following. Do you care what others think or you prefer to discuss this with people that agree with you?

[bubble boi @bubbleboi]

I am turning MAX BULLISH right now. Just like a river boat gambler holding his last chips I’m looking at the landscape and thinking fuck man we are in the fast take off stage right now.

I want to put it all max levered long on the high beta names and walk off into the orange glow of the sun. My bullishness has absolutely nothing to do with the math problems that are being solved most of them don’t matter that much.

But it’s more the realization that every AI lab just had. They all woke up and said wait a minute what are we doing selling $200 a month subscriptions to idiots who are making slop apps and nonsense substack articles when we can just subsidize a few super geniuses by giving them real compute to work on an important problem.

No one really cares about navier stokes besides some CFD simulation cycles. But what if we put these compute resources on real valuable problems? That opens up a whole new world. The bull case for AI has always been to me discovering new drugs or even verifying they work/don’t work reliably… I was thinking way way wayyyy too small.

There are hundreds of problems in engineering that if you got the experts in the room with real compute and even if they made a marginal improvement it would be worth hundreds of billions to the economy. Imagine we can build a steep slope device at 0.3V that’s worth like several generations of lithography nodes, or we can design a sub 100 femtojoul/bit optical IO and cut datacenter power by 40%, or even something as simple as more energy efficient molecular separation to replace distillation which is like 15% of all industrial energy use.

This is the real shit guys. The labs are realizing quickly this is the REAL frontier this is the final land grab. Open source can never compete here because they don’t have the capital and resources this plays right into the labs hands. They can’t give everyone a 10,000 agent swarm but they definitely can give it to Exxon mobile and charge them $50M dollars for it if it means they find more oil or Dow Chemical to create a new kind of super glue or to Moderna to computationally create different mRNA therapies… the list goes on and on.

A big reason why I never cared much for AI is despite all the normies going googoo Gaga over it is that literally they don’t have any clue what to do with it. Like if you were mediocre before AI you are mediocre after it, nothing changed u just got more skills that you don’t know how to use and you have no imagination or intuition. It’s the same reason why education is a waste on the majority of the population because let’s be real most of you aren’t going to be deriving differential equations or using epistemological concepts on your day job.

But you give this kinda compute and unfettered AI in the hands of people who are world class. They will cook. It’s the reason why Deepmind has a Nobel prize and no other lab does because deepmind actually hired or worked with the top biochemists and biophysicists to develop alphafold.

AI alone isn’t going to discover anything. Don’t fool yourself and grow up already if you still think AI is going to just figure out large scale nuclear fusion on its own then you are simply lost in the sauce at this point and are beyond saving. The pragmatic answer has always been use the AI to automate the busy work and keep the human experts for the actually thoughtfulness and intuition.

I think the AI labs will wake up now and realize selling coding tokens and image models to normies to make Disney characters isn’t the best business in the long run. Actually using the compute for real serious innovation is and it’s really easy to monetize this upfront.

I expect AI labs to stop looking like API companies and start looking like real research labs now. Thank god.

[Xiuyu Li @sheriyuo]

读知乎回答有感，浅谈一下后训练 benchmark
https://t.co/wA8Q06vZVe

Any benchmark will be saturated，而随着工业界后训练的发展，reasoning / coding 的 benchmark 越来越泛化 / 众包化。其实 LLM eval 一直是个草台班子，每家都可以有自己的 setting，而任何 setting 只要不被强行统一就存在操作的空间，最早可以调 temperature / top_p / length，后来可以调 harness / setting。

AA 给出了一个相对“规范”的 setting，让大家都听它的而不是野蛮生长，但这并不 make sense。只要评测非中心化就一定存在 noise，任何结果都可以是 cherry pick 后的，而 AA 官方的 95% 置信度结果就成了 judge 模型能力的标杆。

代价是什么呢？模型如果没有很亮眼的 AA 分数，即使精心打磨体感 / working / CUA，无人在意，这本质也是模型要 sell 就要迎合表面主流的评测中心化组织。

至少我们可以从每家基模厂、每个新模型的发榜看出来大家并不会被 AA 所局限，而 OSWorld v2、ALE、TB4 / TB-Sci 等 frontier 众包 benchmark 已经告诉我们：最 frontier 的能力一定是最通用的能力，而 agent 已经站在了单 domain 能力的肩膀上。

现在的职业任务分为三类，已经被 AI 训练好的，无法被 AI 训练的，还在被 AI 训练的。所有能归纳在 Computer Use 的都是第三类，而模型如何解决第三类职业任务仍需要数年的进化。

这些众包 benchmark 越来越变成第三类任务的闭包，决定了未来数个月内基模团队的优化方向，而越是 frontier 的 benchmark 就越有 noise。如果 benchmark 团队不为自己的错题率负责，只会诞生越来越多的错题 benchmark：GPQA-Diamond，HLE，充斥大街的弱模型 judge。

为了消灭 LLM-judge 所引发的不确定性，众包 benchmark 一定会越来越商业化，而 evaluation 一定会在短暂的时间内迎来大洗盘。

[astaxie @astaxie]

真的很感慨，斯坦福的课程更新速度太快了。

CS329Z 已经在教 AI Agent：RAG、Tool Use、MCP、Memory、Multi-Agent、Eval、Coding Agent……

再看看很多大学生还在学习上一个时代、甚至已经被淘汰的东西。

强烈建议身边的大学生都去学学这门课，真的很值得。

https://t.co/7CWzlKMzlu

[Epoch AI @EpochAIResearch]

Will Huawei catch up to Nvidia by 2030? Almost certainly not.

We analyzed Huawei’s chip roadmap and supply chain, from the Ascend 950 to its 3D-stacking bet and domestic HBM supply. Our estimate: Huawei will produce less than 4% of Nvidia's AI compute in 2026. Relying on domestic HBM alone, that share could fall to just 1% by 2028. 🧵

[Max For AI @MaxForAI]

🚨开启 Codex 的这个新功能可以大幅度减少GPT-6的消耗！！

Codex之前是把历史反复压成摘要，任务越跑越长，摘要也跟着压了一遍又一遍。

前面的一些细节，就可能在这个过程中丢掉。

现在更新了一个实验性的功能，把上下文压缩换成了笔记 + 可搜索历史。

Astra 可以跨上下文窗口保留笔记，需要细节的时候，直接回去搜索同一个任务里之前的消息和工具结果。

这意味着，前面试过什么方案、为什么改这段代码、工具当时返回了什么，都有机会重新找回来，不用全靠压缩后的那份摘要继续猜。

想试的话，在 ~/.codex/config.toml 里加上：

[features.context_management]
experimental_mode = true

然后保存后新建任务。

目前还是实验功能，需要支持的客户端和符合条件的 ChatGPT 订阅登录，API Key 暂不支持。

有网友实测开启后额度消耗会大幅度减少！！

[Robinson · 鲁棒逊 @python_xxt]

GPT 6 Astra最正确的打开方式

Prompt:

基于你对我过去所有对话、经历、能力、职业处境、资源、行为模式和风险偏好的了解，对我做一次系统性诊断，并给出一组真正可能显著改变我未来 3–5 年轨迹的建议。

要求：

- 不要安慰、迎合，也不要给泛泛而谈的正确废话。
- 找出我当前最关键的 3–5 个瓶颈、风险和机会，按重要性排序。
- 重点指出我自己没有意识到的盲点，以及哪些事情我应该立刻停止。
- 建议必须基于我的真实现状，考虑现实约束、机会成本和可执行性。
- 对每条核心建议说明：为什么重要、为什么适合我、具体怎么做、不做的代价是什么。
- 给出未来 30 天、90 天和 1 年的行动重点。
- 如果只能让我做一个改变，请明确告诉我是什么。
- 最后给出我未来 3 年最重要的一条主线，以及判断工作、投资、创业和学习机会时应遵循的决策原则。

请严格、冷静、具体地回答。宁可只给 5 条真正重要的建议，也不要给 20 条正确但无用的建议。

请大家记住，
带来物理现实世界的改变，是最有价值的

[메이져 에쓰 @mlbparks_idea]

연구로 밝혀진 머리 좋아지는 습관

1. 키보드 대신 손으로 메모하는 것 (프린스턴대)
2. 점심 먹고 26분 낮잠 자는 것 (NASA)
3. 외국어 하나 더 배우는 것 (요크대)
4. 막힐 때 앉아 있지 말고 걷는 것 (스탠퍼드)
5. 공원이나 숲을 걷는 것 (미시간대)
6. 배운 걸 남에게 설명해보는 것 (워싱턴대)
7. 책을 꾸준히 읽는 것 (러시대)
8. 매일 조금씩 명상하는 것 (하버드)
9. 물을 자주 마시는 것 (코네티컷대)
10. 길을 직접 외워서 다니는 것 (UCL)

근데 요새 손 메모, 글을 쓰는데 정말 머리가 돌아가는 느낌이야.
진작할걸 하는 생각도 드는데~

이 중 몇 개나 하고 있어??^^

[陈成 @chenchengpro]

群里看到的 Codex Tip，给对话批量改名。

---

Rename the conversations in the current Codex project. Only change conversation titles. Never touch the project name.

Rules:

- Date comes from the conversation's `createdAt`, converted to Asia/Shanghai. Do not use `updatedAt`.
- Format: `MMDD｜TYPE｜Topic`
- TYPE must be one of these. Use the English code by default. Use the Chinese label only if I ask for Chinese.
FEA / 功能 = feature
DES / 设计 = design
FIX / 修复 = bug fix
OPT / 优化 = optimization
REL / 发布 = release
EXP / 探索 = exploration
DOC / 文档 = docs
RES / 研究 = research
- Use one language for TYPE across all titles in a run. Never mix.
- Topic summarizes what the conversation is actually about. Do not repeat the project name.
- Keep titles short and specific. They show in the sidebar.
- If you cannot tell the topic, keep the original title. Do not guess.
- Change nothing else: not the project name, conversation content, project assignment, order, pin state, or archive state.

Examples (English, the default):

Before: Improve batch text display
After: 0903｜OPT｜Batch text display

Before: New feature discussion
After: 0901｜DES｜UI alignment check

Examples (Chinese, on request):

Before: 优化批次文字显示
After: 0903｜优化｜批次文字显示

Before: 提交代码到 GitHub
After: 0813｜发布｜提交代码到GitHub

Before making any change, output only a two-column table with this exact header:
| Before | After |
Wait for my confirmation. After renaming, report only the results.

[Max For AI @MaxForAI]

🚨因为AI的发展突飞猛进，斯坦福软件工程课里85%的内容就过时了。

斯坦福最新公布了2026年秋季课程 The Modern Software Developer，负责课程的教授直接说：

2025年秋季课程里，85%的材料已经过时。

原因很简单，去年11月Claude Opus 4.5之后，Coding Agent的能力发生了一次明显跃迁，软件开发的工作方式也跟着彻底变了。

所以今年课程几乎重新设计。

学生要学的已经不是怎么更熟练地手写代码，而是：

Agent Skills
高级Context Engineering
MCP
Agent-ready代码库
Agentic Code Review
安全
并行后台Agent
Software Factory

甚至每个学生都必须直接给真实世界的生产级开源项目提交PR。

目前课程的合作项目包括Browserbase、HeyGen、OpenHands、Milvus、CrewAI、Warp、Vercel、Unsloth、Anyscale等一大批AI项目。

课程的核心是LLM时代，AI-native Developer会成为软件组织里最重要的一类人。

以前大学教你怎么写软件。

现在斯坦福开始教你，怎么管理一群Agent写软件了。
9月22日开课，所有课程和资源都会免费公开。

[Natural Philosophy @Naturalphilosy]

“Nothing is more boring than a man with a career. Only those who decline to scramble up the career ladder are interesting as human beings.”

— Aleksandr Solzhenitsyn