测试完了，不好用，没我自己手搓的好用。因为我自己手搓的软件只需要有很简单的功能，满足我的需求就可以了。它这个则得兼顾所有人的需求，找出其中的最大公约数。但是通用软件的时代已经过去了，未来是量身定制软件的时代。

A creator’s inbox: two years of DMs, from requests for help to unsolicited career advice.

AI can outperform us and still miss what we value. I compared AI face ratings with Chinese forum scores, then looked at what my cat can teach us about taste.

If AI can replace everything we do, what makes a human life meaningful? My answer starts with taking my daughter to lunch.

If AI prosperity were shared broadly, what would we compete for beyond a paycheck? I imagine more focus on popularity, activities, and self-expression—a kind of second youth. An optimistic thought experiment, with distribution doing the heavy lifting.

AI techniques change faster than any course can keep up. My approach: start using AI on real tasks, then focus on what I want and whether the result is good. Rule-based work is changing; taste and judgment still need human attention.

I made a free MLBB draft helper. Compare lane win rates and check matchups in your browser. Try it: guigulaoshi.github.io/counte… #MLBB

AI makes building easier, not attention less scarce. A game set on your old campus can matter to a particular group. That’s why I focus on niche needs. Gameplay excerpt: Genshin Impact PAX East 2020 demo, via Gematsu.

My AI skills evolve from actual projects. A video workflow should preserve my style; a game workflow should reflect that game’s goals. Even for the same person, one reusable prompt cannot capture every project’s priorities.

I don’t look for a perfect prompt. I speak my ideas aloud, inspect what AI produces, and iterate. The harder part is knowing what I want—and recognizing when the result is right. That’s where taste and judgment matter.

Code readability reflects human limits. Does unfamiliar AI-generated code always mean worse code? I still review code in important projects, but experiment more freely in small personal ones. Here’s how I think about the trade-off.

Why does AI struggle with taste? It learns existing patterns. Great art is often an outlier. Think of a polished Hollywood sequel that hits every beat, yet leaves you cold. AI can generate surprises. Taste is knowing which ones are worth keeping.

I built a webcam-controlled game. You jump, the bunny jumps. Free to play + open source. Code: github.com/guigulaoshi/bunny… Play: guigulaoshi.itch.io/bunny-ru…

I just released a skill for AI coding agents that builds your own racing game. Name a place, like your hometown or your school, and your agent turns its real roads into a game. Made with it: World Tour, 13 cities. Skill: github.com/guigulaoshi/silic… Play: guigulaoshi.github.io/silico…

Made a browser game that allows you to race the real Bay Area: guigulaoshi.itch.io/silicon-… I will post a video explain my workflow and open source it later.

I switched to codex because it has image generation that comes along with the subscription. For Claude I have to pay extra.

An AI skill for auto-cut MVs: github.com/guigulaoshi/music…