Robin Xie

The Introvert Innovator

[Ray Dalio @RayDalio]

https://t.co/krgBEJX0m8

[Pickle Cat @0xPickleCati]

https://t.co/2ghAA5aLZB

[Guohao Li 🐫 @guohao_li]

Anthropic Claude Cowork just killed our startup product 😅 So we did the most rational thing: open-sourced it.

Meet Eigent 👉 https://t.co/R82WRFoh41

[Guohao Li 🐫 @guohao_li]

Frontier labs spend millions purchasing RL environments for training terminal agents. But we decided to open source it.

Introducing SETA: Scaling Environments for Terminal Agents, the largest open source training RL environments for terminal agents.

We released:
- 400 termianl agent training environments, more to come
- SOTA agent harness on terminal-bench with CAMEL terminal toolkit
- The RL training pipeline and trained SETA-RL-Qwen3-8B model weights

[BNB🔶蛙蛙 @jiamibtc]

贵人最怕的一种回馈方式就是请吃饭，上等的帮助，最怕下等的回应。

滴水之恩，当涌泉相报，可最怕的是，你以为的涌泉，可能只是滴水。

多数人不是不想报恩，是太着急把恩情处理掉。于是请一顿饭、送点土特产，忙着表示一下。

但越是这样表示，越显得浅薄。请吃饭，有时候吃不出深意，很有可能只会吃出尴尬。

我们每个人这一生，大概都会遇到一两个贵人。他们不一定天天联系，不一定话多热情，但他们做了一件事，在你犹豫的时候点了你一下，在你无力的时候托了你一下，或者，只是在人前替你说了一句话，便改变了你在整个系统里的坐标。这样的人，是稀缺的。

而他们之所以帮你，不是看你嘴甜，更不是图你饭局周到，是他们在你身上，看到了值得押注的一点什么，那一点潜力、锋芒，或隐忍的干净。可当你转头说：“哥，哪天请你吃个饭！”你以为你在感恩，从另一个角度，是告诉他：你帮的这事，差不多值顿饭，咱俩扯平了。

你不是在还人情，而是用最短平快的方式，把一次本可以长线共赢的升维协作，变成了一次人情往来，甚至有点像“赶紧把债清一清”的小心思。古人讲：君子之交淡如水。

水是流动的，是不设防的，是相处时留有余地、不急着结算的。而你一请吃饭，就把余地变成了边界，贵人一看，明白了，你不是那种能跑得远的人，你是等着被奖赏的那种人。

请吃饭，是懒人最体面的敷衍。因为请吃饭最容易，不用用脑，不必改变，花钱买个流程感就行了。真正让贵人动心的回馈，从来都不是一顿饭，而是你变了。

你变得不再慌张了，能在风口浪尖稳住自己的频率；你变得靠谱了，能把交代的事办干净，甚至多走一步；你变得有力量了，能自己开路，不再次次都等人搭梯子。这些变化，才是一个人对贵人最沉默也最动人的回应。

但很多人没做到，也不打算真做，于是拿出“请吃饭”这张遮羞布。说好听是回馈，说直白了，是我还没什么成绩，就先走个流程吧。这就像，你答应了老师要写一篇长文，最后交了一张贺卡；你承诺过父母会努力，最后送了他们一个保温杯。形式没错，但分量不对。

最讽刺的，是那种带着情绪的“感恩”，菜刚上桌，人还没坐稳，就开始絮絮叨叨“我是真心感谢你，那天你那句话我一直记得，我这人吧，嘴笨，想来想去也只能请你吃个饭”。说的每一句，听上去都挺真诚，可在贵人眼里，是在说：做不到，所以说点好听的。这请的不是饭，是自己成长的替身啊。

有个概念，叫“象征性交代”，人在面对难以完成的期待时，会用一些表面化的动作来“模拟回应”。它能短期缓解内疚，但也彻底封死了改变的可能。而真正的感谢，不是嘴说出来的，是人活出来的，你活成了他愿意继续帮的人，这才叫高质量回馈。

请吃饭，会打乱别人的时间和节奏。贵人有一个共性，时间密度高、注意力集中、能量闭环完整。他们一天的日程往往像一块块切割好的方糖，紧凑、有序，该输出就输出，该收敛就收敛，很少有“闲着瞎聊”的时间窗口。甚至连走神，都是有规划的。

你突如其来的一句：“哪天请你吃个饭吧。”听上去客气、礼貌、有心，但对方内心多半像是听到一个系统弹窗：你有一个无用但无法拒绝的待办事项。你以为你在维护关系，他看到的，是一个人把社交逻辑当通用钥匙，硬塞进他的高效系统里，打乱了节奏，也浪费了注意力。

贵人怕的不是吃饭，是饭局之后的那段低效能耗。三个小时挂着人情笑脸，听一些客气话、绕圈话、重复话，然后花半天时间恢复专注力，把情绪状态重启一遍，回头一看，当天唯一没“推进结果”的，就是这顿饭。

这顿饭吃掉的，是他本该用于推进项目、梳理战略、陪孩子看一场电影的时间块。这不是你没礼貌，而是你不识系统。

真正的高段位人际交往，从来不是“靠近他”，而是不打扰他，让他觉得你是节奏对齐的人，而不是能量消耗者。

你只在他看得见结果的时候出现，在他需要的时候恰好在场，他自然会想：“这人，不错，跟着舒服。”知音者不言，近人者不扰。越是关键的人际关系，越不靠靠近，而靠对频。

人和人的关系，不是请一次饭就能巩固的，更不是送一次礼就能绑定的。真正走得久的，是价值与节奏对齐，是互不打扰里的彼此信任。

贵人不是怕你欠人情，怕你太快划清了“我已经还过了”的边界。一还，就结束了。一请，就打扰了。一说“我欠你一个人情”，其实是说，我不打算再继续兑现你对我的期待了。

所以，送礼这件事是非常有艺术的，但我说的礼不是那么功利的送。正确的打开方式是，你认为对你好或者重要的人，每当有应季的特产下来给人寄点。夏天寄广东荔枝、江苏桃子，秋天寄阳澄湖大闸蟹，春天寄点春茶，或者你当地的特产最好。

总之就是应季的地方特产，你错开了时间送，一年送个两三次。这玩意你找个供应商，朋友圈好多都是自家卖的，东西也不贵，一圈发下来也用不了五百一千的，自己少喝两顿酒就够了。

[L2WTrades @L2WTrades]

got a DM from a guy who said he made $1.2M trading last year

looked at his profile. 847 followers. no course. no discord. posts maybe twice a week.

thought it was cap

asked him to prove it

he sent broker statements. verified. real.

so I asked him one question: "what's your strategy?"

his answer changed how I think about trading forever.

"I trade one pattern. literally one. for 9 years."

"just one?"

"just one. bullish engulfing at daily support. that's it."

"you made $1.2M from bullish engulfings?"

"I made $1.2M from being the most patient motherfucker on planet earth."

he explained:

when he started, he was like everyone else. 47 indicators. 12 strategies. telegram signals. news trading. scalping. swing. everything.

blew $340k in 4 years.

his wife told him he was done. trading was over. get a real job.

"that's when I got serious."

he spent 6 months backtesting ONE setup. bullish engulfing at daily support. over 3 years of data. (i don't even know how the fuck support/resistance work lmao)

found it hit 61% of the time with average 2.8R when traded correctly.

"that was enough. I didn't need anything else."

now he wakes up. checks daily charts for 15 minutes. if the setup is there, he takes it. if not, he plays golf.

"some weeks I take zero trades."

"doesn't that drive you crazy?"

"it drove me crazy for the first 6 months. now I don't even think about it. no setup = no trade. the market owes me nothing."

I asked about his typical month:

- trades: 8-12
- winners: 5-7
- losers: 3-5
- average month: $70-120k

"you make six figures taking 10 trades a month?"

"bro most months I make more than I made the entire year when I was taking 10 trades a day."

this broke something in my brain.

we've been told MORE effort = MORE results

but trading is inverted

LESS effort = MORE results

because effort in trading usually means:
- more trades (more mistakes)
- more screen time (more emotional decisions)
- more analysis (more paralysis)
- more indicators (more confusion)

he told me his screen time went from 8 hours per day to 45 minutes per week.

his income went from -$80k/year to +$1.2M/year.

inverse correlation

I asked for his advice to new traders:

"pick one setup. test it. prove it works. then do ONLY that setup for 3 years. you'll be rich and everyone will think you're a genius. you're not. you're just patient."

before I ended the call:

"one more thing. everyone who tells you trading is hard is right. but not because the market is hard. because DOING NOTHING is hard. that's the actual skill. doing nothing while everyone else is busy losing money."

he hung up.

probably went to go do nothing.

[Beginnersblog @beginnersblog1]

I spent months studying AI side hustles.

Most fail within weeks.

But 12 are quietly making people $3K-$10K/month working just 10 hours a week.

Here's what's actually working (save this):

↓

First, the brutal truth:

The people making money with AI aren't building the next ChatGPT.

They're doing boring stuff like:
→ Automating appointment reminders for dentists
→ Summarizing contracts for solo lawyers
→ Qualifying leads for overwhelmed realtors

Boring problems = Real money.

Here are the 12 that work:

𝟭. 𝗖𝗼𝗻𝘁𝗿𝗮𝗰𝘁 𝗔𝗻𝗮𝗹𝘆𝘀𝗶𝘀 𝗳𝗼𝗿 𝗦𝗺𝗮𝗹𝗹 𝗟𝗮𝘄 𝗙𝗶𝗿𝗺𝘀
Only 20% of small law firms use AI.
Use Claude to summarize contracts + flag risks. Add your human review layer.
Pricing: $50-150/doc or $500-2K/month retainer
No law degree needed.

𝟮. 𝗡𝗼-𝗦𝗵𝗼𝘄 𝗥𝗲𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗳𝗼𝗿 𝗠𝗲𝗱𝗶𝗰𝗮𝗹 𝗣𝗿𝗮𝗰𝘁𝗶𝗰𝗲𝘀
Dental offices lose 15-23% of appointments to no-shows.
Set up AI voice reminders with or Vapi.
Pricing: $200-500/month per practice
Pitch: "Pay only if no-shows decrease."

𝟯. 𝗥𝗲𝗮𝗹 𝗘𝘀𝘁𝗮𝘁𝗲 𝗟𝗲𝗮𝗱 𝗤𝘂𝗮𝗹𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻
Leads contacted in 5 mins convert way better.
Solo agents can't respond at 9 PM on Friday.
Build AI chatbots that qualify + schedule showings.
Pricing: $300-1,500 setup + $200-500/month

𝟰. 𝗜𝗻𝘀𝘂𝗿𝗮𝗻𝗰𝗲 𝗣𝗼𝗹𝗶𝗰𝘆 𝗖𝗼𝗺𝗽𝗮𝗿𝗶𝘀𝗼𝗻
Only 17% of independent agents trust AI.
But 41% plan to adopt it within 6 months.
Be the person who does it for them.
Pricing: $200-500/month retainer

𝟱. 𝗧𝗲𝗻𝗮𝗻𝘁 𝗖𝗼𝗺𝗺𝘂𝗻𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗳𝗼𝗿 𝗟𝗮𝗻𝗱𝗹𝗼𝗿𝗱𝘀
Small landlords (10-100 units) waste hours on repetitive messages.
AI handles maintenance requests + common questions.
Pricing: $5-15/unit/month
20 landlords × $250/month = $5,000/month

𝟲. 𝗔𝗜 𝗥𝗲𝗰𝗲𝗽𝘁𝗶𝗼𝗻𝗶𝘀𝘁 𝗳𝗼𝗿 𝗟𝗼𝗰𝗮𝗹 𝗕𝘂𝘀𝗶𝗻𝗲𝘀𝘀𝗲𝘀
Salons and restaurants lose customers to missed calls.
AI answers 24/7, takes bookings, routes urgent calls.
Pricing: $150-400/month per location
Walk in. Show a 2-min demo. Close.

𝟳. 𝗕𝗼𝗼𝗸𝗸𝗲𝗲𝗽𝗶𝗻𝗴 𝗧𝗿𝗮𝗻𝘀𝗮𝗰𝘁𝗶𝗼𝗻 𝗖𝗮𝘁𝗲𝗴𝗼𝗿𝗶𝘇𝗮𝘁𝗶𝗼𝗻
Accountants fear AI errors in compliance work.
Be the human verification layer.
Partner with a local bookkeeper for referrals.
Pricing: $300-800/month per firm

𝟴. 𝗣𝗿𝗼𝗱𝘂𝗰𝘁 𝗗𝗲𝘀𝗰𝗿𝗶𝗽𝘁𝗶𝗼𝗻𝘀 𝗳𝗼𝗿 𝗘-𝗰𝗼𝗺𝗺𝗲𝗿𝗰𝗲
Sellers need hundreds of descriptions.
AI + SEO + brand voice = what generic tools miss.
Start on Fiverr. Build reviews. Raise prices.
Pricing: $500-1,500 for store rewrites

𝟵. 𝗩𝗶𝗱𝗲𝗼 𝗘𝗱𝗶𝘁𝗶𝗻𝗴 𝗳𝗼𝗿 𝗖𝗼𝗮𝗰𝗵𝗲𝘀
Coaches film hours of content but can't edit.
Descript + Opus Clip = 70% less editing time.
Turn 1 long video into 10 shorts.
Pricing: $500-1,500/month retainers

𝟭𝟬. 𝗥𝗲𝗰𝗿𝘂𝗶𝘁𝗶𝗻𝗴 𝗔𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝗰𝗲 𝗳𝗼𝗿 𝗦𝗺𝗮𝗹𝗹 𝗛𝗥 𝗧𝗲𝗮𝗺𝘀
Only 14% of companies use AI in recruiting.
But 81% plan to invest.
Screen resumes. Write job posts. Draft outreach.
Pricing: $500-1,500/month retainer

𝟭𝟭. 𝗛𝘂𝗺𝗮𝗻𝗶𝘇𝗶𝗻𝗴 𝗔𝗜 𝗖𝗼𝗻𝘁𝗲𝗻𝘁
Companies rushed into AI content.
Now it sounds robotic
They need humans to fix it.
Pricing: $1,000-3,000/month retainers
You're not competing with AI. You're fixing its output.

𝟭𝟮. 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗼𝗻 𝗣𝗮𝗰𝗸𝗮𝗴𝗲𝘀 𝗳𝗼𝗿 𝗦𝗽𝗲𝗰𝗶𝗳𝗶𝗰 𝗜𝗻𝗱𝘂𝘀𝘁𝗿𝗶𝗲𝘀
Don't sell "automation."
Sell "The 5 automations every recruiting agency needs.
Specific beats generic. Always.
Pricing: $500-2,000 setup + $100-300/month

𝗧𝗵𝗲 𝗽𝗮𝘁𝘁𝗲𝗿𝗻 𝗯𝗲𝗵𝗶𝗻𝗱 𝗮𝗹𝗹 𝟭𝟮:
→ Pick a boring industry
→ Find their repetitive pain
→ Add AI + your human layer
→ Charge monthly

That's it.

𝗛𝗼𝘄 𝘁𝗼 𝗴𝗲𝘁 𝘆𝗼𝘂𝗿 𝗳𝗶𝗿𝘀𝘁 𝗰𝗹𝗶𝗲𝗻𝘁 𝗶𝗻 𝟲 𝘄𝗲𝗲𝗸𝘀:
Week 1: Pick ONE opportunity. Create positioning.
Week 2-3: Build 3 sample deliverables.
Week 4: Reach out to 20 prospects. Offer 2-3 free pilots.
Week 5-6: Deliver. Document results. Get testimonials.

𝗥𝗲𝗮𝗹𝗶𝘀𝘁𝗶𝗰 𝘁𝗶𝗺𝗲𝗹𝗶𝗻𝗲
• First client: 2-6 weeks
• $1,000/month: 1-2 months
• $3,000-5,000/month: 3-6 months
• $10,000+/month: 6-12 months

The bottleneck is marketing, not skills.

𝟯 𝗺𝗶𝘀𝘁𝗮𝗸𝗲𝘀 𝘁𝗵𝗮𝘁 𝗸𝗶𝗹𝗹 𝗺𝗼𝘀𝘁 𝗔𝗜 𝘀𝗶𝗱𝗲 𝗵𝘂𝘀𝘁𝗹𝗲𝘀:
Going too broad (niche until it hurts)
Expecting passive income (AI helps you work faster - it doesn't work for you)
Building on one tool (your value is problem-solving, not tool access)

The opportunity isn't access to better AI.

Everyone has ChatGPT.

- Pick a boring problem.
- Solve it with AI.
- Add human oversight.
- Charge monthly.

That's the formula.

PS:) Like, REPOST, and Bookmark!
.
.
.

[周期教授 @Zhouqi2013]

我后来发现，真正能赚到钱的人，思路都很简单：

不追概念，不讲宏大叙事，只盯三件事——谁在反复花钱、钱为什么花、凭什么给你。

第一层：钱跟着“生活刚需”走
钱最稳定的地方，永远在重复发生的生活场景里。吃、住、行、健康、陪伴、效率，这些东西一年要付无数次账。
比如宠物、老人、单身人群，消费频率高、情绪黏性强，只要服务靠谱，复购自然来。见过做老年手机教学的，一个月单客才几十块，但客户能跟一年，后面设备、服务全从他这走。

第二层：价格由“省事程度”决定
很多钱不是赚技术，是赚“你帮我少想一步”。
忙的人不想研究、不想比较、不想踩坑，只要你把流程捋顺，价格就能翻。陪诊、代办、代运营，本质都一样：把复杂变简单。
有个朋友做企业资料代整理，单子不大，但客户黏，一年下来比很多风口项目稳。

第三层：溢价来自“感觉和身份”
大量消费，看起来没用，但情绪价值极高。
定制、圈层、仪式感、被理解，只要打中，掏钱非常果断。小众社群、情绪内容、身份标签，粉丝不多，但转化率很吓人，因为认同感本身就在付费。

第四层：专业度会自己抬价
一件事，只要你在某个细节上做到“别人懒得学、学不会、没耐心”，价格自然往上走。
修图、剪辑、方案、分析，单点突破比全面开花值钱得多。客户不是看你会多少，是看“换你行不行”。

第五层：信息差和位置差一直存在
同样的东西，有人知道门路，有人不知道；有人站中间，有人站两头。
渠道、资源、撮合，本质是位置问题。很多人不生产产品，但站在链路中间，反而更安全。

第六层：注意力会慢慢变现
只要你在一个领域持续出现，哪怕不完美，信任也会累积。
很多成交不是因为某一条内容，而是“我看你很久了”。长期输出，本身就是筛选用户。

第七层：能跑长线的项目，一开始都慢
订阅、会员、长期服务，前期很无聊，后期很稳。
多数人不是没机会，是熬不过最开始那段“钱不多但在增长”的阶段。

钱不在风口上，在生活里。你只要在一个重复出现的场景里，站稳一个能被替代成本高的位置，剩下的就是时间问题。

[Long Chen @LongChenNotes]

我本来以为Agent就是人的原型，通过学习各种skill(技能)打磨workflow(工作流)来提升自己。

我通过和一个高人聊过之后发现大错特错❌

这里面有三个概念，我是搞反了。

第一、Workflow（工作流）
你每天回邮件、做表格、撕逼、写周报。
在AI眼里，这叫Inference（推理）。
这就是纯耗材。
资本买你的时间，就是为了买你的推理。
他们最希望把你的Temperature设为 0。
不要创新，不要随机，只要像头驴一样稳定输出。
很多大厂P7、P8，哪怕年薪百万
本质上也就是个高频低延迟的API接口。
过去一年你跑了多少个Token？
跑得越多，显卡磨损越快，离报废越近。

第二、Skills（技能）
你会Python，会剪辑，会写复杂的prompt。
这叫Function Calling（工具调用）。
以前我们信奉技多不压身。
现在？
这是贬值最快的资产。
你苦练了三年的画图技巧， Nano Banana一出来就给你秒了。
你背了一万个单词，也就是个低配版的谷歌翻译。
沉迷堆砌Skills，就是在跟摩尔定律赛跑。
注定输得底裤都不剩。

第三、LLM（大模型）
你的世界观，你的底层逻辑，你的决策框架。
这才是你的Base Model。
也是你唯一的护城河。
但悲剧就在这：99%的人，出厂设置即巅峰。
往后几十年，都在用同一个版本的Windows 95跑新地图。
你在公司卷了五年
Workflow跑得冒烟了
Skills背包塞满了
但你的LLM版本还是GPT-2。
遇到点复杂问题，直接蓝屏。

这个世界的真相很残酷：
系统设计出来，就是为了锁死你的算力。
它给你发工资，用Workflow填满你的时间
用房贷车贷，锁住你的风险偏好
就是为了不让你有空闲去Fine-tuning（微调）你的大脑。
毕竟，觉醒的 Agent 不好管。

听一句劝：
2026年，别再傻乎乎地卷工作量了。
那叫无效熵增。

停下来。
减少Inference，增加Training。
去读难读的书，去见厉害的人，去复盘最痛苦的亏损。
哪怕每天发呆一小时
也比在群里抢两块钱红包有价值。
那是你在更新权重。

你是想做一辈子随时可被替换的 API？
还是想做一个拥有Opus 4.5 级别认知的操盘手？

算力有限。
别把你的CPU，烧在毫无意义的死循环里。

哦，对了和我聊的高人叫“脑总”

[Curiosity @CuriosityonX]

Right now, more than 24 billion kilometers from Earth, a tiny spacecraft is drifting through the darkness—Voyager 1, the most distant object we've ever sent into space. 🌌

[Robin Xie @nanobin1984]

Who knows Snail 🐌 Intimacy is so tender and amazing!!! Snail 🐌 💕 🐌
#snail

[Robin Xie @nanobin1984]

Happy Birthday, Ethereum! 🎂

Ethereum turns ten,
from miners’ fire to stakers’ zen.

No kings, just sovereign code,
a chain where value freely flowed.

We walked through hacks, we built again,
in this new world, we remain.

@ethereum @VitalikButerin @ethereumJoseph @ethereumfndn

[Elon Musk @elonmusk]

Wow, a prime number with 420 digits!

[Ｍａｒｃｕｓ @NeoSe7ens]

@PaulleyTicks

[Robin Xie @nanobin1984]

@grok Now, with “deep search” it indeed returns the correct answer…

[Robin Xie @nanobin1984]

@grok please fact check this

[Robin Xie @nanobin1984]

If no “deep search” then @grok will say: this picture is likely to be fake news…

[Nansen 🧭 @nansen_ai]

𝗬𝗼𝘂 𝗱𝗼𝗻’𝘁 𝗻𝗲𝗲𝗱 𝘁𝗼 𝗮𝘀𝗸 “𝘄𝗵𝗼’𝘀 𝗯𝘂𝘆𝗶𝗻𝗴?” 𝗶𝗳 𝘆𝗼𝘂 𝗯𝘂𝗶𝗹𝗱 𝘆𝗼𝘂𝗿 𝗼𝘄𝗻 𝗳𝗶𝗹𝘁𝗲𝗿𝘀.

Smart Segments on Nansen = a workflow for slicing through wallets like an onchain sushi chef.

Here’s how to set up a custom segment to track exactly the kind of wallets you care about 🧵