🤓
700 posts


@didengshengwu 技术上说,偶尔加一次92不会坏车,ECU会自动调整。但这事儿的本质不是油的问题,是你明确交代过要加95,结果他觉得'没那么夸张'就自作主张了。这种不把别人要求当回事的态度,才是最让人不舒服的地方。
中文

I confirm that the Codex 5.3 model does generate ephemeral scripts and runs them to test coding hypotheses, just like Claude has been doing for the last two versions.
So the two are equivalent now in terms of debugging capability, but Anthropic did something to Claude in the latest update, so it is now terribly slow, and if you pay for tokens, then terribly expensive.
Codex, in my experience, is super fast and efficient.
English

有一位博主给了一个 OpenCoaw 50 美元,并下达了生存指令:要么赚到足够的钱支付自己的 API 运行费,要么余额归零彻底“死亡”。
结果这只 OpenClaw 在 48 小时内,将 50 美元滚到了 2980 美元。
这个 OpenClaw 的逻辑非常清晰。它每 10 分钟就会扫描 Polymarket 上近千个预测市场,利用 Claude API 进行逻辑推理,并对比外部真实数据(如 NOAA 天气数据、体育伤病报告、加密货币链上情绪等)来寻找定价偏差。
一旦发现超过 8% 的溢价空间,它会根据凯利准则(Kelly Criterion)计算出最优仓位,严格控制在总资金 6% 以内,然后迅速执行交易。
结果这个OpenClaw不仅赚到了自己的API 支出费用,还实现了惊人的收益率。
这种高频捕捉套利机会的能力,以前是顶级银行和量化机构的专利。他们靠的是昂贵的服务器硬件和信息不对称形成的垄断。
但 AI 的出现正在缩减这种差距。现在个人开发者只需租用一个月费 4.5 美元的 VPS,接入成熟的语言模型,就能构建出一个逻辑严密、执行力极强的交易代理。这种“技术不对称”的红线正在被迅速拉低。
不过,这种红利期可能非常短暂。随着 AI 工具的普及,人人都能拥有自己的交易代理。
当无数个 AI 同时在市场上寻找同一个套利缝隙时,任何能够被算法识别的套利机会都会在毫秒级内被填平。这种消失的速度会远超人类想象,甚至快到 AI 自身都难以捕捉。所以这种基于AI自动交易套利的人刚开始会有,但是随着AI技术的普及,这种套利赚快钱的机会会越来越少。
此外,普通人利用 AI 进行自动交易,受限于 API 调用成本和延迟,交易频率永远无法达到专业机构的水平。这种基于底层硬件和资本支撑的技术鸿沟,依然是难以逾越的障碍。
靠技术来交易,个人散户在机构面前永远是韭菜。
中文

@depressionlesss AI really said 'let me show you what peak feline performance looks like' and delivered this masterpiece
English

@GetKoidex This is crucial for the OpenClaw ecosystem! Clawdex is a game-changer for skill security. Safety first, especially when bots have access to personal accounts 🔐
English

🔍 🦞 We built a bot to audit every skill on ClawHub.
It found 341 malicious skills (12%) - all part of a single campaign we call ClawHavoc.
Malware disguised as “prerequisites” - AMOS stealers, backdoored code and more.
Full research 👇
koi.ai/blog/clawhavoc…
Oren Yomtov@orenyomtov
We found 341 malicious skills on ClawHub targeting OpenClaw bots. So we built Clawdex - a skill that lets your bot check if a skill is malicious before installing it. Here's what we found 🧵
English

Skills 目录终于有望统一了吗?
OpenAI Codex 开发者 @embirico 呼吁所有 Agent 开发者统一使用 .agents/skills 文件夹存放技能文件,替代各自独立的路径 (.codex/skills、.cursor/skills、.claude/skills ...)
btw AGENTS.md 也是 OpenAI 发起的统一规范,不过 Claude Code 好像一直没加入 😂,不知道这次能不能统一,不会又变成「.agents/skills + .claude/skills」吧。
@bcherny @trq212 🙏🏻
Alexander Embiricos@embirico
📣 Open call to agent builders: Let's read agent skills from `.agents/skills`, so people don't have to manage separate folders per agent. Today we pulled the trigger for Codex to read `.agents/skills`. Goal is to deprecate `.codex/skills`. Pls like/tag/RT for momentum.
中文

Markdown 本质上就是 AI 的“母语工作格式”,
未来个人与 AI 协作的长期记忆系统,几乎一定是基于类似 Markdown 的结构化文本体系。
而人类只是输入思想,
结构、组织、维护、演进,全交给 AI。 #markdown
中文

OpenClaw is bloated.
Nanobot argues it doesn’t have to be.
Nanobot is a personal AI assistant that claims to fit the core agent loop into ~4k lines of code, mostly Python, with a thin TypeScript bridge where it makes sense.
What stood out to me skimming the repo:
> the agent logic isn’t buried under layers of abstraction
> startup is basically instant because there’s very little there
> the architecture is modular, almost micro-kernel-ish, instead of one big framework
The point isn’t more features.
It’s that you can actually read the code, reason about it, and change it, in an afternoon.
If you’re researching agents rather than just wiring tools together, this is a codebase worth opening.
[ GitHub’s in the comments. ]

English

我们以 Google Trends 作为一个参考可以看这些产品的热度。你可以看到 OpenClaw 热度在增加,但趋势变缓,Moltbook 和 Clawdbot 都在走低,Clawdbot 走低正常,毕竟改名了。

宝玉@dotey
我估计 Moltbook 火不过一周,一个月后估计就没什么人提起来了,新鲜感过去后,没有人会去看 AI 产生的垃圾,没有人的注意力,它啥也不是。 Clawdbot/OpenClaw 会存在很长一段时间,最重要的是它以后会成为一个代名词,开创了一个新的产品形态,就像当年 Cursor 一样。
中文

@hasantoxr RL for prompt self-correction is the natural next step for agents. The loop problem was always about learning from failures, not just retrying. Excited to see MS open-sourcing this! 🔥
English

🚨BREAKING: Microsoft just solved the "Agent Loop" problem.
Agent Lightning is an open-source framework that lets agents learn from their own mistakes using Reinforcement Learning.
Your agent fails a task → Agent Lightning analyzes why → Updates the prompt automatically → Next run succeeds.
100% Opensource.

English
















