I know some readers have concerns about today’s AI, or large language models. A lot of websites used to be written by people, word by word, and now they’re full of AI-generated articles with nothing much in them, the so-called “AI slop.” When you see a post tagged “AI-Assisted,” your first thought might be “more AI junk,” and you might not want to read any further. So I want to explain here how this blog uses AI, and how I think about it myself.
Why I Don’t Just Type
Actually, I’m not bad at typing. I wrote code for many years before there was any AI. I can touch-type with no problem, and I’m pretty fast at writing code. But after all these years of writing code, I still don’t like typing out articles, because typing blocks my thinking. When I write an article, I check a sentence two or three times, then rework the structure and rephrase it, and my thinking stops flowing. Basically, there’s a bit of laziness in this too, but I think it’s reasonable.
I’ve actually had this idea for a long time. In 2016, in Thoughts on the Form of the Internet of Things (Part 2) (in Chinese), I wrote this in parentheses, almost in passing: “Typing is so tiring. It would be great if the things in my head could turn straight into text. Holding them in is really uncomfortable.” Now this is more or less possible.
How My Posts Are Written Now
This is how I write posts now: I say my ideas out loud, the speech is transcribed, and then AI organizes it into an article in my own style. I have only one requirement for the AI: it doesn’t change my thinking or my meaning, and the result should read as if I’m talking to the reader directly. The views and examples in the articles are all mine, and the AI doesn’t make up experiences, data or quotes for me. I read and revise the outline and the draft before anything is published.
For example, GpgFrontend’s New Version Model and Release Mechanism was written this way: I talked for almost an hour, and that was turned into the article. This page was written the same way.
So behind these articles is my own thinking, not an AI. I really put effort into them. I don’t let AI generate a whole article while I know nothing about what’s in it, and I don’t hand everything over to AI either. That’s not how I want to use AI.
Other Places Where AI Is Used
Besides organizing my dictation, this blog uses AI in two other places.
One is the English translations. The site is mainly in Chinese, and most of the English versions are translated with the help of AI. The translations have to stay faithful to the original, and they should sound the way I talk when I speak English.
The other is a few older posts that were originally generated or polished by AI, so their style wasn’t mine. In November 2026, I had them rewritten in my own style. The rewriting was also done by AI, so they’re tagged “AI-Assisted” too (these posts are in Chinese):
- The Relativity and Conditionality of Knowledge
- Knowing and Doing Are Not One
- On a Psychological “Anti-Qualitative Change”
- On Focusing on a Goal and Trying to Reach It
- On What the Essence of Things Is
- My Experience Using AWS Services
- Weekly Technical Report for the Second Week of April 2023
- A Partial Solar Eclipse Without Baader Film: Observing with a Makeshift “Projector”
About the “AI-Assisted” Tag
Every post where AI took part in drafting, organizing or rewriting gets the “AI-Assisted” tag. In Chinese it’s “AI辅助.” You can find all the posts with this tag on the tag page. The tag is a note that tells readers how a post was written. It doesn’t mean the post is AI slop. So if you see this tag, maybe give the post a read first, and see whether the ideas in it are my own.