luworks retweeted
2025年起,我反复强调的四个原则:
原则1,时代的第一性原理:LLM一定会越来越聪明,benchmark越来越高,context window越来越大,reasoning越来越长,价格越来越便宜,inference速度越来越快,
这是scaling law今天依然持续的具体方向,不用你质疑,这是你唯一的信仰和行业最大共识。
原则2, 管理学设计红利:从我提出“自动编程机”、行业提出vibe coding、SWE-Agent以来,从cursor到manus到metaGPT到claude code,
人们逐渐把LLM Agent抽象成人,把软件管理、工程管理、管理学等等所有方法论直接套在multi agent workflow上面,严格按照人类管理学的方式去拆分、review、执行、反馈、循环,
这一波很快红利也吃完了,因为 a. LLM Agent毕竟不是人,存在着memory有限、执行力有限、function calling工具有限等等局限;b. 人类用于管理学的各种方法,直接套在LLM Agent上有利有弊,红利迅速挖掘完,剩下的弊端大量存在,比如过度交流、七手八脚、随时停工等等。
原则3,LLM Agent的职位和定位:绝大多数人,把claude code当做一个工具,最终的产品是用工具来完成的,最终的代码也是人与SWE Agent一步一步interactively迭代产生、迭代review、迭代部署的,
而我反复告诉过所有人,也是我又一条首次提出的原创观点,multi agent未来越来越会变成本身的一个runtime,这个runtime就运行在production里面,产品和面向的对象消费的,不只是软件或者SaaS本身,而是这个runtime实时产生的内容,
所以claude code/opencode/codex/openclaw这些agent,本身将会越来越多地被嵌入到产品本身,在产品关键逻辑和决策中发挥作用,
而绝对不仅仅停留在开发层面,把产品仅仅局限在SWE Agent单向产出和部署的代码和服务上。
原则4,也是我一直强调的,就是当人们试用了SWE Agent这种强大工具之后,人们还有哪些low hanging fruits可以寻找?SWE Agent目前最适合解决哪类问题?
我反复讲过的一点是,对于一个设计复杂、环境复杂、场景复杂、用户复杂、体量复杂、范式复杂、一切开放、一切无解的超级复杂系统,这并不是SWE Agent最擅长的领域,相反这些场景需要人去和环境、客户、场景、性能一点点迭代才能打磨好的产品,
比如微信的100种功能,Facebook的一大堆功能模块和十几年来迭代出来的极其复杂的infra,支付宝后面成千上万的基金和风控,这些都不是AI Agent能一次性解决的问题,相反这些场景和问题不仅高度开放,更高度依赖人的观察、人的设计、人的反馈、人的定义。
AI Agent最适合的场景,甚至是我原创提出goal driven(github.com/lidangzzz/goal-dr…)设计的最适合场景,在不考虑token使用量和时长的条件下(参照第一性原理,这些成本必然会越来越低),必然同时具有如下特点:
a. 定义简单、干净、封闭(一道数学系、一个确定性最小系统、一个编译器、一种算法、一个lean证明、一个电路或者信号模拟、蛋白质模拟和预测、CAD设计与仿真、游戏关卡测试、行为经济学仿真,都是well-defined problems,都有非常明确且封闭的边界)
b. 解决问题的搜索空间巨大(可能有100~10万种天马行空的解决方案,并且绝大多数都是错的)
c. 容易验证,容易verify,验证的成本是设计成本的千分之一(比如编译器,设计可能需要几万行甚至几十万行,验证只需要2000个test case全面覆盖,或者一道数学题,解决需要100步,验证答案只需要带入或者lean编译这一步)
当然,写一段简单的代码,定义一个封闭、完整、定义完全的编程问题,符合上面这些定义,
但是设计一套巨大、复杂、开放、与现实世界深度绑定、高度耦合的系统,让这个系统复杂迭代、添加功能、沟通、review、工程管理、产品管理,这些问题都远远超出这个范畴,很明显是不符合这个要求的。
人们未来探索这些multi agent产品和场景的最关键出路,在于继续挖掘这一类问题,而不是盲目把agent比作一个人,乱套各种管理学方法。
luworks retweeted
What separates a good Forward Deployed Engineer from a great one?
Technical depth?
Speed?
Product intuition?
Communication?
Understanding the customer?
Something else?
Curious to hear from people who’ve actually worked as FDEs or built FDE teams.
What’s the #1 trait you’d look for?
Tomorrow will be my last day at Google after 27 years, and watching it grow from 25 people to 190,000+ has been an amazing journey. Below is a note I shared with many people internally at Google today. An excerpt is:
It has been an absolute pleasure to work with you and to help build some of the most widely used and impactful products of all time. As a kid, I dreamed of helping build software that would be used by many people, and Google now has thirteen products used by more than a billion people (amazing!). Our work has had a tremendous impact in the world, and I have been lucky enough to collaborate and form friendships with many colleagues that I deeply admire, respect, and enjoy. It still brings me joy every time I see people out in the world using our products to find information, handle email, translate documents, watch videos, learn new things, navigate and understand the physical world, browse the web, use their phone, run large-scale computations on our infrastructure, ride in an autonomous vehicle, or perform complex tasks with the help of our AI systems. I hope you all share this sense of joy, because it is a shared accomplishment! Thank you to all of my colleagues at Google over many years!
Now I'm excited to go start @DiscoLoopAI with my longtime friends and colleagues @Sanjay_Ghemawat, @OriolVinyalsML, and @quocleix.
(Updated post: slightly redacted to not have some personal info)
luworks retweeted
过往皆序章, 如果你被联合创始人欺骗情感和金钱
那不是你的损失,那是他们的损失
因为他们这一辈子再也找不到像你这样靠谱的创业合作伙伴,他们一辈子都会因为曾经对你使用欺骗手段, 受到惩罚
因为世间再也没有人愿意和他们创业,因为大家不再相信他们
Startup 8 years, rested 1 year, I experienced countless challenges and have made tons of mistakes. Yesterday a friend of mine 4 years ago pointed to me some people fully understand the world and see through it late in life, some understood fully early on and they succeed faster. Look at the world as the way it truly is, is actually painful and sad. But after 1 year of recovery and sharing my feeling to startup friends, I felt hope again. He got tricked by his cofounder and almost went to jail, but can still keep it together and run a new startup again, why can’t I do it again with everything I’ve gone through?
luworks retweeted
人其实需要休息的,特别是长时间的休息
我记得18年从Deepin Linux离职出来的时候,那段时光真的很痛苦
因为人从一个非常非常忙的时期,突然什么都不干以后,人的脑袋会发疯的,一直想干点什么事情。我记得我家有一块黑板,那块黑板是我用黑板漆漆的,整整一面墙都是黑板。所以我就在我的黑板上构建我的下一个商业模式,下一个创业目标,下一个创新的想法
但是我第二天起来,还是默默的把这些项目都擦掉了。因为我体验过那种过程,就是当你跟世界失去交流的时候,其实一切都已经毫无意义了。即使规划的事是非常聪明的,即使技术是非常先进的,毫无意义。因为没有人关心
我记得那段时间在家读了很多很多的书,读了经济学的书,读了历史的书,读了材料的书,读了科幻的书,读了文学的书,读了传记的书,读了很多很多原来沉迷于写代码的时候忘记的时间,忘记的书
直到2019年的2月份,有个朋友邀请,说想让我帮他做售货机。说句实话,当时我进去做的时候还觉得挺懵的,因为你看亚马逊也在做呀。然后国内的京东也在做呀。觉得我们3个人其实很难。但是我梳理完以后,其实也很简单,就是把商业核心提取出来,剩下的全部做减法。不要做金融交换系统,直接用微信电话号码解决。不要做提前开门,一定要支付完再开门,这样的话你就会从人工智能预测变成打标对单就好了。最后最核心的其实就是,不要用那么先进的板子,纯粹增加成本。我直接用树莓派Zero加了扩展板。然后把摄像头的固件刷一下,让它直出压缩后的视频,这样的话硬件成本就做得很低,最后他们原来要卖2万块钱的柜子,成本是6,000多,被我干成了,板子加锁加柜子,成本1600,然后最后他们卖6880,卖得很火爆。到目前为止,今天已经是第7年了吧,依然是行业记录
当卖得很火爆的时候,这个所谓的朋友,我觉得特别不讲信用。因为当时我说我愿意帮你。因为这件事情对我来说不是很难,但是我有要求,就是暑假和寒假我要去陪老婆小孩。因为我上次创业的时候太拼了,对他们的时间太少了。我说这个薪水一般都可以,但是这个暑假和寒假的假期是一定要答应我的。最开始入职的时候,其实这个朋友答应的很好。我三个月把这个行业最具性价比的售后位搞出来以后,他说做得非常好,但是离创业还差一点点
我当时跟他说,算了吧,这个产品就当我送给你的,也不要跟我扯别的,我这辈子最讨厌言而无信的人
所以当时我就离开了这个朋友的公司,我转头就跟老婆说,我不知道下次创业是什么时候,但是你想暑假旅游吗?我说我们兜里还有2万块钱,我觉得旅游应该花个1万多,剩下的几千块钱就靠运气了,要不再去跟别人打工,要不就想办法创业
我和我老婆都是那种一块钱可以花到2块钱的,从来不考虑存钱的主。所以当时我们俩异口同声的说走
当时我们从武汉出发,一路经过陕西兰州,然后从青海的南面开始环线,一直达到德令哈,达到大柴旦。然后绕过南八仙魔鬼城达到敦煌,再从北边的瓜州、酒泉绕回张掖。全程6,000公里的青海大环线自驾
我的吉普掠过青海的时候。油菜花加上青海湖的颜色,就像韩剧胶卷质感的那般浪漫
我坐在祁连雪山下的冰河,仰望天空的云,那一时刻我明白了人生的意义。人生的意义在于坐在大自然下面,敬畏大自然,享受人生的虚无和人生的每时每刻
第一次去张掖的时候,我都不敢相信自己的眼睛,世界上真的有这么梦幻的地方啊
这一圈最好玩的其实是南八仙魔鬼城。我记得一车人在炮弹坑,在倾斜30度的斜坡下越野的时候,车都快翻了,但是大家超级兴奋,因为那是属于成人的碰碰车、跳床,那种原始的快乐才是人生的意义
我记得旅游完回来以后,感觉人突然开悟了,明白了人生的意义。也知道了未来的方向。那时候兜里就只剩2000块钱了,我说老婆,我要出去融资了,融不了资就去给别人打工
最后融到了450万,开启了这一次的创业之旅。到今天为止,公司营收已经远超八位数了
所以回想起来,要感谢当年的人生暂停,是那一段很长很长的暂停,让我明白了人生的意义,不再追逐金钱,不再和他人对比,不再向世界证明我行
人生的意义在于安静下来,问自己,你到底喜欢的是什么?然后去做就好了,持之以恒的做,没有一件事情是做10年不会成功的,没有一件事情
好了,今天的分享就到这里,欢迎大家关注,我是一个创业20多年的操作系统老兵,经常分享我的创业故事和创业经验
喜欢我文章的朋友,欢迎点赞、收藏、转发
好奇我们公司产品的朋友,欢迎打开我的推特主页,查看我的置顶帖,购买一排懒猫微服吧,你的支持是我创作的无穷动力,感谢你🤠
luworks retweeted
Sora clone Dreamina by Bytedance has launched on Android in the following countries: UK, Australia, Canada, France, Germany, Japan, South Korea, Mexico, and Brazil. 5-q.cc/h2d Follow me when you join!!! My cameo is OPEN! dreamina.capcut.com/sv2/ZSHT… #aivideo #seedance
- 本质是什么?
通过 IM 指挥 claude code
- 怎么实现的?
agent sdk + skills + 定时任务 + im 连接器 + 基于文档的记忆
- 解决了什么问题?
满足了情绪价值和 AI 厂商 Token 卖不出去的问题
- 适合拿来做什么?
装逼和赛博念经
所谓小龙虾帮约炮成功的一定是编的
luworks retweeted
Deepseek got called out for scraping 150k Claude messages. So I'm releasing 155k of my personal Claude Code messages with Opus 4.5.
I'm also open sourcing tooling to help you fetch your data, redact sensitive info & make it discoverable on HF - link below to liberate your data!
Replying to @AnthropicAI
Your AI hates Whites & Asians, especially Chinese, heterosexuals and men.
This is misanthropic and evil. Fix it.
Frankly, I don’t think there is anything you can do to escape the inevitable irony of Anthropic ending up being Misanthropic. You were doomed to this fate when you chose your name.
The Name of the Wind.
luworks retweeted
I am very excited about AI, but to go off-script for a minute:
I built an app with Codex last week. It was very fun. Then I started asking it for ideas for new features and at least a couple of them were better than I was thinking of.
I felt a little useless and it was sad.
luworks retweeted
I'm Boris and I created Claude Code. I wanted to quickly share a few tips for using Claude Code, sourced directly from the Claude Code team. The way the team uses Claude is different than how I use it. Remember: there is no one right way to use Claude Code -- everyones' setup is different. You should experiment to see what works for you!
luworks retweeted
nanochat can now train GPT-2 grade LLM for <<$100 (~$73, 3 hours on a single 8XH100 node).
GPT-2 is just my favorite LLM because it's the first time the LLM stack comes together in a recognizably modern form. So it has become a bit of a weird & lasting obsession of mine to train a model to GPT-2 capability but for much cheaper, with the benefit of ~7 years of progress. In particular, I suspected it should be possible today to train one for <<$100.
Originally in 2019, GPT-2 was trained by OpenAI on 32 TPU v3 chips for 168 hours (7 days), with $8/hour/TPUv3 back then, for a total cost of approx. $43K. It achieves 0.256525 CORE score, which is an ensemble metric introduced in the DCLM paper over 22 evaluations like ARC/MMLU/etc.
As of the last few improvements merged into nanochat (many of them originating in modded-nanogpt repo), I can now reach a higher CORE score in 3.04 hours (~$73) on a single 8XH100 node. This is a 600X cost reduction over 7 years, i.e. the cost to train GPT-2 is falling approximately 2.5X every year. I think this is likely an underestimate because I am still finding more improvements relatively regularly and I have a backlog of more ideas to try.
A longer post with a lot of the detail of the optimizations involved and pointers on how to reproduce are here:
github.com/karpathy/nanochat…
Inspired by modded-nanogpt, I also created a leaderboard for "time to GPT-2", where this first "Jan29" model is entry #1 at 3.04 hours. It will be fun to iterate on this further and I welcome help! My hope is that nanochat can grow to become a very nice/clean and tuned experimental LLM harness for prototyping ideas, for having fun, and ofc for learning.
The biggest improvements of things that worked out of the box and simply produced gains right away were 1) Flash Attention 3 kernels (faster, and allows window_size kwarg to get alternating attention patterns), Muon optimizer (I tried for ~1 day to delete it and only use AdamW and I couldn't), residual pathways and skip connections gated by learnable scalars, and value embeddings. There were many other smaller things that stack up.
Image: semi-related eye candy of deriving the scaling laws for the current nanochat model miniseries, pretty and satisfying!
luworks retweeted
扉展!VRのAIアート展示会!楽しみ!!
ツネくんからお誘い受けてたのですが、残念ながら今回は出展が間に合わず…
ですが当日お邪魔しに行くので皆さんもよかったら一緒に行きましょう〜
チケット販売スタート!
#SunRraiz #扉展
SunRraiz主催AIアートイベント
『扉展』 チケット販売解禁!
3割引クーポン配布中 コード:【 OPENDOOR124 】
(コピーして決済画面で入力!)
会場:Cluster🌎
新しい扉を開く準備はできましたか?
▼チケット購入はこちら sunrraizhub.com/event-detail…
luworks retweeted
ニューヨークポスト紙のSora特集にて紹介してもらいましたー!!
Ariさん(@littlemissjacob )からアメリカのSoraクリエイターチームにご招待いただきまして、この機会を得ることができました。
このチームは、SoraにてSlop系動画で現在ダントツにPVを稼いでいるfancysonさんも在籍してたりします。
今後Soraというプラットフォームを介して、アメリカで何か活動をさせていただくかもしれません!



































