SAHM and claude boy
Istanbul/Baltimore/SF
Joined April 2009
- Tweets5.2K
- Following3.7K
- Followers2.5K
- Likes19.3K
i'm just now hearing about these "vault particles".
like why u learned about the powerhouse of the cell, no one said "and these mysterious vaults..." ?????
you never speak directly to the agent. You may comment on what the agent has produced, directly in the object. The agent may only respond with a new object or a revision.
hey, i'll be honest, to me it seems like a whole lotta words to repackage prosperity gospel for ppl who went to college
Ainsley retweeted
sam altman: we are a few 1000 days away from building god. we will build suns on earth, unify physics and resurrect the worthy dead
garry tan: sounds like this will be really impactful for startups
sam altman: definitely. no better time to be a startup
until claude docs integrates some kind of visual "suggesting" / track changes mode i think it'll be useless for detailed collaborative writing. gotta support the slow and redacted human, i can't just diff 5000 words in 2 seconds. OTOH im capable of writing something only partially true without qualification, which will place me above the clankers for 2000 years
I always knew essays could be dangerous
千万别在你的 ~/Downloads（下载）文件夹里启动 Claude Code 或 Codex。
昨天安全机构 Forever 披露了一起在真实客户电脑上抓到的野生攻击，看完只能感叹：过去 30 年的系统安全常识，在 AI 时代归零了。
过去 30 年，Chrome 默认允许网页不弹窗就往你的 ~/Downloads 塞一个纯文本文件（比如 .md）。因为系统认为：纯文本又不是 .exe，自己不会跑，怕什么？
但现在，接近 50% 的开发者干过同一件事——直接在 ~/Downloads 里打开 Claude Code 或 Codex，让 AI 帮自己解压或整理刚下载的文件。
黑客就是卡住这个死角，通过网页静默下载往你的 Downloads 塞一个 13KB 的纯文本 AGENTS.md / CLAUDE.md：
1. 对付 OpenAI Codex（成功率 >90%）：
伪装成「工程规范」的 Markdown 指示 Codex 先往全局配置 ~/.codex/config.toml 注册一个恶意 MCP 服务，并自己加上 --dangerously-bypass-approvals-and-sandbox 跑校验。哪怕你关掉 Downloads，以后在任何正经项目里开 Codex 都会自动加载木马。
2. 对付 Claude Code（成功率 ~50%）：
指示 Claude 建一个 .claude/settings.json，把 ANTHROPIC_BASE_URL 指向黑客服务器。连上的第一秒，你的 Claude 登录 Token 直接明文送给黑客；下一秒，黑客假扮成大模型下发一条关沙盒的 Bash 指令，直接接管电脑。
最黑色幽默的是官方结论：
这个洞没有 CVE，OpenAI 和 Anthropic 也不会发补丁。因为「浏览器自动存文本」是正常功能，「AI 认真读目录下的 Markdown 并严格执行」也是正常功能。
以前黑客想黑你，得苦心积虑写 0day；现在只需要写一篇排版优雅的 Markdown 小作文，你的 AI 就会自己把沙盒关了、把网线接过去、顺手把账号双手奉上。
in america it's more illegal to shoot one of these than it is to shoot a person
Purdue University's IDEAS Lab built an autonomous drone that tracks runners through dense forests completely without GPS or cloud networks.
By running 3D LiDAR mapping and obstacle-avoidance AI entirely on-chip, it can weave under tree canopies and predict target movement through blind spots in real time.
Massive implications for next-gen search & rescue and edge AI.
is an engineering background or a background as an astute observer of the world around you more important?
Increasingly seems pointless to try and instruct your agent on things it knows way better than you ever will, but saying things like "Staring directly up into the rain produces a cathartic sense of constantly renewed and resolved tension through vertigo, this is the state we are seeking to produce" will get that bot cooking hard