...... (@pingf) profile picture
sanfancisco
Joined May 2009
Photo by ...... (@pingf)Photo by ...... (@pingf)Photo by ...... (@pingf)Photo by ...... (@pingf)Photo by ...... (@pingf)Photo by ...... (@pingf)Photo by ...... (@pingf)Photo by ...... (@pingf)Photo by ...... (@pingf)Photo by ...... (@pingf)
...... retweeted
Someone trained a tiny 47M-param diffusion model that paints pixel art emoji in your browser 👾 I built a tiny Slack emoji maker on top of it. Type anything, get 4 to pick from. huggingface.co/spaces/cfahlg…
43
208
24
2,298
85,934
...... retweeted
不要再去读 X 上又臭又长的 harness 工程的文章了, 跟这篇文章相比, X 上的那些文章垃圾都不如。 Lilian-weng 的这篇新的博客文章, 是我目前读过的最写的清晰、最容易理解的 harness 工程, 也是递归自我改进的 harness 工程写的最好的文章。 文章是一篇框架式的详细描述了 self-Improment 的 harness engineering 全部知识框架。 链接是我做的中文版本 pandatalk8.com/blog/harness-…
18
186
4
772
61,627
...... retweeted
开源项目推荐:Claude Code From Scratch 这是一本学习 Claude Code 的开源电子书,严格来说不仅仅是电子书,还有代码,不需要你去看 Claude Code 的 50 万行代码。 用 ~4300 行代码(TypeScript 和 Python 两个版本分别实现)复现了 Claude Code 的核心架构——Agent Loop、13 个工具(含并行执行 + 流式早期启动)、4 层上下文压缩、语义记忆召回、技能系统、多 Agent、MCP 集成……每一步都对照真实源码讲解它怎么做的 → 我们怎么简化的。 有 13 章内容,每一章都是一份分步教程,跟着动手写几千行代码,快速理解 Claude Code 这样最好用的 coding agent 的精髓。 读完你就能大致理解了 coding agent 的工作原理,我跟着快速浏览了下都有了些新的收获,推荐有兴趣的可以看看。 有中英文版: diwang.info/claude-code-from…
69
341
21
1,434
108,761
...... retweeted
先讲一个你可能从来没听说过的人。 1930年,英国语言学家 C.K. Ogden 做了一件很偏执的事——他把整个英语词汇表研究了一遍,然后问了一个问题:如果只能留下最核心的那些词,英语还能不能用? 他的答案是:850个词,够了。 这850个词,不是随便选的,是他穷尽分析之后筛选出来的"语言骨架"。他把它们分成5类: 100个操作词(Operations):这、那、是、有、做……就是那些让句子转起来的胶水词 400个普通事物(General Things):名词里最高频、最基础的那层 200个可描绘事物(Picturable Things):能画出来的具体东西,比如 door、fire、river 100个性质词(Qualities):描述世界的形容词,比如 clear、simple、important 50个反义词(Opposites):成对出现,学一个等于学两个 —————— 然后他把这套体系命名为 Basic English,发表出来。 接下来发生的事情更有意思。丘吉尔在二战期间公开赞扬这套体系,认为它是盟国推广英语教育的最佳基础。H.G. 威尔斯把它写进了《世界大战》的续集设定里,作为未来世界通用语言的蓝图。乔治·奥威尔……没有赞扬它,但《1984》里那套压缩语言"新话"(Newspeak)的灵感,正是来自对 Basic English 的深深警惕——他担心语言被缩减之后,人的思想也会跟着被缩减。 这就是这850个词的能量:支持者觉得它是桥梁,批评者觉得它是枷锁,争论了将近一百年。 但我想说的不是那场争论。我想说的是——这850个词,对今天的英语学习者来说意味着什么。 大多数人学英语,是被一本本厚厚的词汇书和无尽的考纲压着学的。CET-4、CET-6、托福、雅思……每一个证书背后都是几千个词的暴力记忆,背了忘,忘了背,从来没有真正"用"过。 Ogden 给出的逻辑完全相反:不是多,而是深。 这850个词,每一个都是高频中的高频,每一个在日常英语表达里都有不可替代的位置。当你真正把这850个词学透——不只知道它的意思,还知道它的用法、它的近义词在什么场景该用哪一个——你对英语的掌握会比背了5000个词却每个只认识中文意思要扎实得多。 这是一种截然不同的语言观:根扎得够深,枝才能生得够远。 —————— 于是我做了这个网站:ogden.munch.love (准确的说我让 Codex 做的) 原版 Ogden 的资料是一本PDF,黑白的,排版像上个世纪(它确实是上个世纪的)。我想让它变成真正可以用来学习的东西。 花了不少时间(花了不少 Token )最终做成了这样: 每一个词,都有一张完整的卡片。 卡片上有:这个词的中文释义、英文定义(用简单的英语解释英语,就像 Ogden 本人的风格)、一句真实语境下的例句,以及2到3个同义词。五个类别各有自己的颜色,一眼就知道你在学哪个区域的词汇。 每一个词,都可以听发音。 接的是有道词典的语音接口,默认英式发音,可以切换美式。点单词听一次,点例句听一次,语速比正常稍慢一点,适合跟读。不是那种机械的合成音,真的可以听。 最让我花心思的,是同义词那里。 很多人背单词的时候,同义词只是"差不多的意思"。但真正用英语的时候,difference 和 distinction 不一样,speak 和 talk 不一样,simple 和 easy 不一样。差在哪里?什么场景该用哪个? ——————— 网站是免费的,不需要注册,不会收集任何信息。 直接访问就行:ogden.munch.love 如果你身边有正在学英语的人,或者觉得自己英语"学了很多年但还是用不好"的,可以发给他们看看。 也许这850个词,是一个值得重新开始的起点。 👆 以上的文案也是它写的,我越来越没有用了
511
2,054
100
8,773
994,690
Marco Rubio was very impressed with the ceiling and light fixatures inside the Great Hall of the People.
158
317
198
3,087
869,300
...... retweeted
What if the EU is red?
82
180
34
5,795
131,154
Chinese scientists have developed, The best shortest-path algorithm in 41 years! A team from Tsinghua University has broken Dijkstra's "sorting barrier" - the first improvement since 1984. Just use for a world-map 🤯 Paper - arxiv.org/pdf/2504.17033
474
3,066
321
29,086
3,405,882
感谢中国如此热情的接待 Thank you China for your warm welcome!
2,997
1,900
451
30,749
3,146,049
最近和一位被某杭州大厂 PUA 一年的朋友吃饭 感觉现在这些大厂的工作方式不止是严重落后于时代了 每个中层都用高情商在编制着高层喜欢的梦境,让高层在梦里笑的开心 只是不知道梦何时会醒 不醒其实也不要紧,毕竟公司不靠AI续命
21
10
201
51,104
吴恩达的课程,使用NotebookLM,已经整理完毕。一共五个模块,在本帖下评论区将依次附上其笔记链接。 这些材料适合作为原课程的补充材料,而非主菜。原课程中除了免费的视频可供学习之外,还提供了若干Jupyter Notebook的练习!
14
176
5
650
81,789
Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single, dependency-minimal codebase. You boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web UI. It weighs ~8,000 lines of imo quite clean code to: - Train the tokenizer using a new Rust implementation - Pretrain a Transformer LLM on FineWeb, evaluate CORE score across a number of metrics - Midtrain on user-assistant conversations from SmolTalk, multiple choice questions, tool use. - SFT, evaluate the chat model on world knowledge multiple choice (ARC-E/C, MMLU), math (GSM8K), code (HumanEval) - RL the model optionally on GSM8K with "GRPO" - Efficient inference the model in an Engine with KV cache, simple prefill/decode, tool use (Python interpreter in a lightweight sandbox), talk to it over CLI or ChatGPT-like WebUI. - Write a single markdown report card, summarizing and gamifying the whole thing. Even for as low as ~$100 in cost (~4 hours on an 8XH100 node), you can train a little ChatGPT clone that you can kind of talk to, and which can write stories/poems, answer simple questions. About ~12 hours surpasses GPT-2 CORE metric. As you further scale up towards ~$1000 (~41.6 hours of training), it quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests. E.g. a depth 30 model trained for 24 hours (this is about equal to FLOPs of GPT-3 Small 125M and 1/1000th of GPT-3) gets into 40s on MMLU and 70s on ARC-Easy, 20s on GSM8K, etc. My goal is to get the full "strong baseline" stack into one cohesive, minimal, readable, hackable, maximally forkable repo. nanochat will be the capstone project of LLM101n (which is still being developed). I think it also has potential to grow into a research harness, or a benchmark, similar to nanoGPT before it. It is by no means finished, tuned or optimized (actually I think there's likely quite a bit of low-hanging fruit), but I think it's at a place where the overall skeleton is ok enough that it can go up on GitHub where all the parts of it can be improved. Link to repo and a detailed walkthrough of the nanochat speedrun is in the reply.
669
3,306
731
23,903
5,824,144
上海 云边的小卖部, 贩卖魔都的浪漫☀️。 Happy weekend! ☺️🌸🍀🌷🍒
2
10
2
45
922
Manufacturing returns to America. (Set to a Chinese soundtrack)
18
25
2
96
44,582
习近平说,那两个二球,我都看不下了。
151
25
4
1,188
841,487
3 minute tutorial on making apps with Grok 3 and deploying instantly with Replit.
453
3,043
345
20,722
16,385,887
...... retweeted
A Chinese AI lab just dropped the best ever open-source text-to-video model: Step Video! – 30B param, 540p, ~8s at 30fps – Trained on 1000s of H800s – Evaluates as well as Meta MovieGen, feels as good as Sora / Veo Paper and demo is awesome and reveals all the gory details:
40
333
13
2,254
258,105
BREAKING HISTORIC BOMBSHELL: Watch President Trump Announce A Plan To Negotiate A Trilateral Agreement With Russia & China To Cut All Defense Spending By 50% In An Unprecedented Move To Achieve Global Peace » WATCH/SHARE THE LIVE X STREAM HERE: nitter.cf/i/broadcasts/1lPJqMrLQ…
13,310
33,020
2,170
203,510
44,556,779
...... retweeted
浙江大学出的这个开源的书籍「大模型基础」值得一看,行文风格挺不错的,易读、严谨、有深度的大模型教材。 github.com/ZJU-LLMs/Foundati…
28
690
10
2,569
354,885
The architecture diagram presented below illustrates some of the key components & how they interact with each other! It will be followed by detailed descriptions & code for each component:
2
20
1
217
32,779
Let's build a RAG app using DeepSeek-R1 (100% local):
59
330
14
4,159
707,191