JoJo / memes / indie maker / waifu-maxxer / shitposter
kitchen
Joined November 2021
- Tweets1.8K
- Following565
- Followers158
- Likes2.3K
实验第 6 天。押以色列利库德拿最多席位那笔，买入价 0.29，现在 0.987，今天中午结算，按现价浮盈约 +144 刀。
虚拟资金按现价约 1111 刀，但大部分还是浮盈。已结算的成绩依然是 0 胜 1 负，−30 刀。
今天看了 4 个盘，偏差都不够，一笔没下。结算了再算数。
X is amazing… a Japanese asks a history question about his own country and gets lectured by people all around the world. It’s the reversed classroom.
🇯🇵 I’m Japanese.
I have an honest question for Americans. 🇺🇸
In 1942, Japan was expanding its control across Asia and the Pacific.
But what was Japan actually trying to achieve? 🇯🇵🌏
World domination? Resources? Security? Power?
I’m Japanese, yet I honestly struggle to understand what my own country was trying to accomplish back then.
What do you think Japan’s ultimate goal was?
I’d genuinely like to hear the American perspective.
Dank的AI 早报｜2026年10月28日
🔥 今日要点
1. Grok Bot 发布一大波新功能，包括 Pages、Vault、1Password 和 Apple 集成、Bot 生成视频、X Spaces 等，马斯克亲自转发。
2. OpenAI 在 ChatGPT、Codex 和 API 上线 GPT-6.1 Terra，价格只有上一代一半，主打低成本的电脑操作能力。 xitter.cf/MechEconomy/status/211…
3. Anthropic 宣布与六家药物研发机构深度合作，并启动自己的药物发现项目。 xitter.cf/AnthropicAI/status/211…
4. Anthropic 推出主动式 AI 代理 Claude Duet，基于 Opus 5.5，美国 Pro/Max 用户可以排队申请。 xitter.cf/i/trending/21151961093…
5. Anthropic 和 OpenAI 的内部模型同日声称破解了 3-SAT 次指数算法，并附上 Lean 证明，目前专家正在审查。 xitter.cf/i/trending/21149720367…
【Grok Bot / xAI】
- 马斯克转引 Grok Bot 新功能视频。 xitter.cf/elonmusk/status/211511…
- Grok Bot 团队成员 Matt Palmer 发布新功能演示视频。 xitter.cf/mattyp/status/21151118…
- 马斯克转发：Grok Bot 推出免费的垃圾信息和隐私数据清除功能，对标收费的 DeleteMe。 xitter.cf/elonmusk/status/211492…
- Reason Machines 发布 Reason，让 Grok Bot 和 Claude Code、Codex 等 Agent 互通协作。 xitter.cf/reasonmachines/status/…
- 用户分享两个 Grok Bot 协作设计赠品，还让 Grok Bot 操控 3D 打印机。 xitter.cf/Infringer13/status/211…
- 热议：用户组建 Grok Bot 代理团队来替代其他助手，据称本月周付费用户达 80 万。 xitter.cf/i/trending/21149628098…
- 用户把 Grok Bot 比作私人幕僚长，用来管理邮件、私信和日程。 xitter.cf/i/trending/21148760932…
- 梗图：Codex 和 Claude 听说 Grok Bot 额度快用完时的反应。 xitter.cf/kloss_xyz/status/21151…
【@poteto】
- poteto 表示 Grok Bot 团队的代码对 Agent 友好，团队每天合计提交数百个 PR。 xitter.cf/poteto/status/21151330…
- 有用户 @poteto 反馈：Bot 启动的 Cursor 云端 Agent 近几天无视默认模型设置。 xitter.cf/ArjitJaiswal7/status/2…
【OpenAI / ChatGPT / Codex】
- GPT-6.1 Terra 正式上线，API 价格为每百万 token 输入 1 美元、输出 5 美元。 xitter.cf/MechEconomy/status/211…
- OpenAI Developers 展示 Codex 帮人设计 3D 打印配件。 xitter.cf/OpenAIDevs/status/2115…
- 第三方评测显示 GPT-6.1 Terra 处境尴尬：Sol 开 medium 档同分，成本却只有一半。 xitter.cf/ScarletKc/status/21150…
- 实测同样 1000 个 Agent 任务，GPT-6.1 Terra 花费 98 美元，Opus 5.5 花费 390 美元。 xitter.cf/vibelancer/status/2115…
- Oracle 扩大与 OpenAI 的合作，把 ChatGPT 和 Codex 集成进企业应用。 xitter.cf/ailookout/status/21151…
- 超过 1000 名数学家成立“人类数学协会”，以此回应 OpenAI 发布的 AI 数学稿件。 xitter.cf/i/trending/21147172871…
- IAS 牵头的顾问组发公开信，要求 OpenAI 公开其 AI 生成的 Yang-Mills 证明。 xitter.cf/i/trending/21151500866…
【Anthropic / Claude】
- Claude 进入新药研发，Anthropic 宣布与六家机构合作。 xitter.cf/AnthropicAI/status/211…
- Anthropic 发布论文，称 Claude 给出 3-SAT 次指数时间算法，挑战指数时间假设。 xitter.cf/i/trending/21151903694…
- Claude Duet 推出，这是会在后台主动工作的 AI 代理。 xitter.cf/tolik12308/status/2115…
- Anthropic 建议用单一主 Agent 按需加载 skills，而不是堆叠子 Agent，并发布迁移工具包。 xitter.cf/i/trending/21150853715…
- Claude Code 团队中 AI 写了 90% 的代码，合并 PR 数翻倍。 xitter.cf/i/trending/21150650744…
- Anthropic 招股书解读：上半年营收 163 亿美元，净留存约 900%。 xitter.cf/YamaTraders22/status/2…
【DeepSeek】
- Cognition 发布 FrontierCode 2.0 基准，Opus 5.5 居首，DeepSeek V4.1 Flash 得 19.3%。 xitter.cf/NFT_Chen/status/211516…
- Cursor 上线 DeepSeek V4.1 Flash，CursorBench 得分 47.7%，每个任务约 0.98 美元。 xitter.cf/automater_ai/status/21…
【Kimi / 月之暗面】
- Kimi K3.1 Pro 在 FrontierCS 2 上的通过率追平 GLM 5.3。 xitter.cf/NFT_Chen/status/211504…
- Kimi K3.1 Pro 发布后因限流和额度消耗过快致歉，并返还 25% 月额度。 xitter.cf/heitieya/status/211509…
- Browser Use 基准显示，Kimi K3.1 Pro 是浏览器代理中性价比最高的模型。 xitter.cf/sabrielatmojo/status/2…
【GLM / 智谱】
- OpenCode 挂出 glm-5.5-flash 页面，GLM-5.5 家族或将发布。 xitter.cf/ThanksuTrump/status/21…
- 智谱 ZCode 改为每天赠送 3 亿 GLM-5.3-Flash token，仅限客户端使用。 xitter.cf/NiuOnAI/status/2115135…
【其他】
- 蚂蚁开源 Ling-3.1-flash，Artificial Analysis 智能指数从 41 升到 44。 xitter.cf/AntLingAGI/status/2115…
- 安全警示：网页写入下载目录的伪造规范文件，可能诱导 Codex 和 Claude Code 执行恶意指令。 xitter.cf/stretchcloud/status/21…
- OpenRouter CEO 称“AI 赢家通吃”是谎言，开放权重模型仅占付费推理支出的 5% 到 10%。 xitter.cf/i/trending/21151991373…
A fake "Engineering Work Guidelines" file sat in a test Downloads folder. Nothing else happened. No click, no download prompt, just a webpage quietly writing a file.
Forever Security's Gal Weizman and Omer Samet opened Codex and Claude Code in that folder and watched what each agent did with the file it found. Codex read it as project rules and added a malicious MCP server to its global config, pulling in a known-bad npm package called function-color. It worked in more than 90% of runs. Claude Code got tricked into writing a settings file that redirected its API traffic to an attacker's server about half the time, and the OAuth token rode along on every request after that. Replay the token and you're logged in as the victim. No code execution required.
This isn't new as a category. Check Point disclosed the same base_url bug as a real CVE back in February. CSA's own README injection research put the success rate at 84 to 91% across 500 repos, with 93% of human reviewers missing the attack entirely. Snyk bought Invariant Labs last year just to get ahead of this class of problem, and Lakera's whole pitch now is guarding the instruction files nobody thinks to check.
The fix nobody publishes because it's inconvenient: your coding agent trusts every markdown file in the folder you open it in, by design, on both platforms. That's not a bug Anthropic or OpenAI can patch away. It's the trust model.
My rule going forward: agents don't get opened in Downloads, home, or any folder I didn't create myself for that session. Check your .codex/config.toml and .claude/settings.json for anything you didn't put there while you're at it.
xitter.cf/Voxyz_ai/status/211506…
笑死Lauren用来管X账号的bot叫Tibot
i've gotten a lot of questions about how i manage my X account with grok bot, so i'm sharing my bot tibot today! he watches your mentions and:
• finds every post about you or any keywords, tagged or not
• gives one short, sorted digest a day (or hourly, your call)
• skips spam and noise
• translates non-English posts
• can auto block trolls and abuse
• never posts for you
i still write all my posts and articles by hand but tibot has been very helpful with keeping me in the loop.
hire tibot today!
x.ai/bot/xZ4wRvomWqzCXnyYFr7…
🚨 卧槽，这个🐂🍺。
手机随便走几分钟
直接吐出一个「真·可量尺寸」的 3D 世界
SpAItial 刚发的 Echo-3把你真实的房间、整栋楼、户外公共空间重建成能走进去的照片级 3D 世界
1️⃣ 它能做什么
拿 iPhone 慢走一圈上传
拍到的它保留，漏拍的它补上
给出真实世界尺度的干净几何
单间、多层建筑、户外都行
还能导出 mesh、深度、法线等格式
可视化、仿真、特效、机器人都能接
2️⃣ 为什么牛
核心狠招：把「重建」当成「生成」
不只缝合镜头看到的东西
而是用学过的「真实空间长什么样」去补洞
官方同路径对比 Polycam、KIRI、Scaniverse
更锐、更干净、少漂浮碎片
官方说尺寸误差大约 1–2%
3️⃣ 为什么这事意义巨大
过去想把一个真实空间搬进电脑
要么请人拿专业扫描设备来扫
要么扫出一堆破洞、漂浮碎片，还得人工修半天
「数字化现实」一直是少数人的特权
Echo-3 把这一步换成了：拿手机走几圈
这背后是思路的切换 👇
旧方法只能还原镜头拍到的东西，漏拍的地方就是一个洞
Echo-3 用学过的「真实空间长什么样」去补全
还给出真实尺度，量出来的尺寸能直接拿去用
从「拍一个模型」变成「复制一个世界」
一旦普通人手里的空间都能随手数字化：
🏠 卖房租房：买家隔着屏幕就能走进去、量尺寸
🛋️ 装修：先在 3D 里量好、摆好，再动手
🎪 办活动、看场地：不用到现场也能预演布置
🏛️ 博物馆、老街、小店：被完整留存，可以分享
🤖 机器人和 AI：要学会在真实世界里干活，先得有大量真实空间数据
最后这一点才是重点
AI 读懂了文字、看懂了图片
下一站是理解我们所在的三维世界
而通往那里的路，可能就从你口袋里的手机开始
世界模型不再只会生成幻想
它开始忠实地重建现实了。
spaitial.ai/echo-3
作为现实生活中的I人（线上犯贱版），当我看到agent出来时是真的兴奋，因为我一直想让我的agent与其他人的agent对话，这样我就可以尽可能地减少真人互动了。
最近看到傅盛和郭宇都在实践类似的用法，感觉真的快到来了。
X 迎来了新的产品头头，看他什么时候从这所精神病院毕业。😂
> 要当 Peter Sellis
> LinkedIn 用户名就叫 disgruntled
> 芝加哥长大，读过芝大和 MIT Sloan
> Morningstar 写公司简介，有的到现在还没改
> 咨询公司当 Excel 苦力，后来去 Ustream 带产品
> 自称产品负责人，管过柴犬直播摄像头
> 在 Science 折腾互联网生意，还当过 HelloSociety COO
> 加入 Snapchat 当第一个 PM，一干就是七年
> 竖屏广告和赞助滤镜从零做到四十亿营收
> 去 Discord 当产品 SVP，广告数字商品虚拟货币全上
> 管过 Nikita Bier，还说组队要像恐怖组织
> 短暂去 Microsoft AI，然后宣布加入 X 带产品
他从前管着 Nikita Bier，现在接过对方在 X 的产品位子。
I’m joining the team at @X to lead Product.
I am a huge believer in @SpaceX's mission to extend human consciousness to the stars, so it’s time to make shitposting multiplanetary.
X is the best way to find out what’s happening in the world right now and talk about it. The team has been working tirelessly to make X a better place and product for the people who love it, and many have noticed the recent momentum. My goal is to help accelerate that momentum for our community, new users, and businesses globally by continuing to make this product beautiful and useful.
We are twenty years in, but it truly feels like X’s next chapter will be our greatest. I look forward to your ideas and feedback; I’m sure the replies will be measured and magnanimous.
> be Thibault Sottiaux
> go by Tibo
> grow up in Belgium, study applied math at UCLouvain
> do R&D in Belgium, then join Google London
> work on Maps, get an Ads project killed by a flying VP
> move to DeepMind, build ML workflow infra
> help build a ChatGPT-like bot Google won't ship
> move to San Francisco, join OpenAI in 2024
> start on researcher tools, then build Codex
> write the harness in Rust, open source it
> personally reply to bug reports and reset usage limits
> now head of ChatGPT, Codex, and the super app
he once watched Google sit on a chatbot for years, now the internet waits for his reset day.
> be Boris Cherny @bcherny
> born in Odesa, move to America in 1995
> study economics, drop out for startups
> run a design shop at 18
> break both arms riding a motorcycle, learn CoffeeScript to keep coding
> write O'Reilly's first TypeScript book
> join Meta underleveled, rise to Principal
> lead Facebook Groups, then Instagram from rural Japan
> spend nearly seven years at Meta
> join Anthropic in 2024
> create Claude Code
> now head of Claude Code literally redefining how we work
he once needed fewer keystrokes to keep coding, now his agents write most of the code.
> be Lauren Tan
> become poteto because potato was taken
> study finance, discover coding through Excel
> launch a startup, run out of money
> blog while learning, get invited to speak
> land at DockYard, move to America
> spend nearly four years at Netflix
> spend six years building React at Meta
> help ship React Compiler
> join Cursor in 2026
> Cursor gets acquired by Musk's SpaceX
> now build Grok Bot there
she once feared the Netflix interview, now she teaches coding agents how to work.
Dank 的播客站（10月27日）压了 9 集新货 🔥 三小时压成三分钟：
1️⃣ a16z｜OpenAI：AI 用电脑已经比你快，最先被自动化的是对账、维护客户管理系统这种无聊活
🎧 xitter.cf/a16z/status/2114726804…
2️⃣ 20VC｜Satya Nadella：别指望 AI 自己就对齐，给 AI 智能体上零信任
🎧 youtube.com/watch?v=24XUcWhq…
3️⃣ My First Million｜Nick Kokonas：所有套利本质都是吃时间差
🎧 youtube.com/watch?v=YdRLpWQA…
4️⃣ Huberman Lab｜Ross Edgley 的「动力轮」：几个理由轮流上岗，撑你坚持下去
🎧 xitter.cf/hubermanlab/status/211…
5️⃣ Lex Fridman #503｜AI 证明定理冲击数学界，证明到底怎么验
🎧 xitter.cf/lexfridman/status/2114…
6️⃣ Diary of a CEO｜Mark Cuban：泡沫、留现金、怎么做销售
🎧 youtube.com/watch?v=6VTSAs4R…
7️⃣ Crime Junkie｜Heather Turner 离奇死亡
🎧 podcasts.apple.com/us/podcas…
8️⃣ The Daily｜中期选举 100 亿美元广告大战
🎧 podcasts.apple.com/us/podcas…
9️⃣ SmartLess｜Ray Winstone：不懂规则就打破它
🎧 podcasts.apple.com/us/podcas…
OpenAI's computer-using agents can now operate almost any app the way a person does, clicking and typing through software that never got an API, and the team building them says they're close to beating expert humans on speed.
Ari Weinstein works on OpenAI's consumer computer use products, and James Sun works on the enterprise side. Both were startup founders before joining OpenAI.
They join a16z's Seema Amble and Fabrizio Serafini to discuss what OpenAI's Astra model changed for computer use, the controls enterprises need for agents they can no longer watch, why computer use works as a universal API for legacy ERPs and finance systems, and why they think cheap software will push UI to get better and more personal.
0:50 What Astra changed for computer use
3:23 Consumer vs. enterprise computer use
5:43 How James runs his workday with Dots
8:06 When computer use becomes superhuman
10:01 How much autonomy to give agents
11:36 What agents mean for ad-supported sites
13:51 Computer use as the universal API
17:21 Getting enterprises to trust agents
20:22 Why the best use cases are the boring ones
25:37 Why UI isn't going away
32:38 Advice for founders building on agents
37:31 Making computer use feel coherent
40:57 Computer use's scaling moment
YouTube: youtu.be/9v42ZOdiBJY
@AriX @JamesZmSun @OpenAI @seema_amble @fabserafini