All Articles

In the recent speech by DeepSeek's Liang Wenfeng that has been circulating, there is an interesting data point that happens to intersect with my recent thoughts. Therefore, I will write down some predictions about the entire large model industry for future falsification and correction of my thinking

For a long time, Chinese AI models could be roughly divided into three routes. The first route is actively making models smaller. For example, minimax-m3, kimi k2.5, hy3, and other models that actively limit activated parameters to around 20B. Whether 20B activated parameters count as small is itsel

nobody really teaches you research. you get a desk, a problem someone else picked, and a vague instruction to produce something novel. so most people reverse-engineer the job from what they can see, which is papers, threads, and announcements, and what they end up learning is how to look like a rese

Go to DMIT to buy a server, but it's currently unavailable. Buying from Bandwagon Host or other service providers that support CN2 GIA lines is also fine. After payment, save and enter the server page, click to save the SSH private key. Let AI handle all the subsequent work: Here I used the latest R

In the past few days, several friends @mentioned me, saying my open-source tools were actively recommended when they asked AI. Without doing anything, they got indexed. I thought maybe I should spend an hour structuring the content properly, which would be even better. So after finishing, I quickly

On March 31, 2026, security researcher Chaofan Shou discovered that the source map files in the Claude Code package published by Anthropic to npm were not stripped. This means: The complete TypeScript source code of Claude Code, 512,000 lines, 1903 files, was exposed on the public internet just like

A recap of current design skills available in the market: Impeccable Enhanced frontend-design skill, featuring 17 design commands (/polish, /audit, /distill, etc.). https://impeccable.style UI Skills Includes 15 independent skills: baseline-ui (Tailwind design consistency), fixing-accessibility (acc

In 2016, when AI was far from becoming the widespread wave it is today, I had already begun earnestly attempting something: using language models to understand, modify, and generate code. I named the company ai.codes. This name hardly needed explanation: AI codes—treating 'code' as a verb, meaning '

今天这篇文章源于最近半年深度使用 Claude Code、两个账号每月 40 刀氪金换来的一些踩坑经验，希望能给大伙一些输入。 刚开始我也把它当 ChatBot 用，后来很快发现不对劲：上下文越来越乱、工具越来越多但效果越来越差、规则越写越长却越不遵守，折腾了一段时间，研究了 Claude Code 本身之后才意识到，这不是 Prompt 问题，而是这套系统的设计就是这样的。 这篇文章想和大伙聊聊这几个事：Claude Code 底层怎么运作、上下文为什么会乱以及怎么治理、Skills 和 Hooks 应该怎么设计、Subagents 的正确用法、Prompt Caching 的架构影响，以及

By Lan Xi The wave of "Lobster Fever" sparked by OpenClaw is becoming increasingly surreal. If, right after the Spring Festival holiday, having Lobster installed at home was still a joke, still stuck at the stage of using Nano Banana to create images for a laugh, then last weekend's long queue outsi