Seeing Ma Boyong share his diary-writing method was very interesting! And it's quite similar to Karpathy's recording method, which I shared before. This method can be well integrated with AI, and I'd like to try it out in 2026. Let me first introduce how they both recorded: To reduce friction, he only recorded facts (basically a diary entry), without recording opinions, reflections, embellishments, or worrying about word choice. Karpathy's diary also used an append-only mode; he would just throw whatever came to mind at the top of the document, without tags, categorization, or hierarchy, using the diary as a clipboard. By only recording facts, the space for "performing for others" or "self-indulgence" is eliminated, significantly reducing the friction of the input process. Furthermore, it can also be used for voice input; after writing something down, simply turn on voice input and say a sentence—very convenient. Furthermore, they share a common characteristic: they use only one file for recording. Andrej Karpathy uses a single document in his notes app. Ma Boyong uses a single Word document. This eliminates the need for organization; they can simply use Command + F to search for keywords when needed. Moreover, this kind of diary-like record is relatively short, perhaps only tens of thousands of words a year, making it easy to retrieve relevant information. Ma Boyong primarily uses this method to address his mid-life memory decline. Andrej Karpathy views it as a tool for freeing up the brain's RAM: through note-taking, the brain can store this information, allowing it to focus on other tasks requiring concentration. I found that their recording method is also very compatible with AI: This kind of data is excellent native AI corpus. Because it possesses high-density semantics, including: names, book titles, events, commands, and code—all easily extracted structured information. Furthermore, this method is very context-friendly. Because of its small size, perhaps only tens of thousands of words a year, it perfectly fits the context of current large language models. Furthermore, because it's plain text, AI can query and organize it very efficiently and conveniently. This low-friction, high-information-density, plain text recording method is inadvertently very suitable for modern AI. Through this method, you can quickly build your own AI memory, similar to ChatGPT. And plain text prevents you from being tied to any particular product.

Karpathy's note-taking method is here: You can find Ma Boyong's on his WeChat official account of the same name.

Haha, this meme!

Vibe has created an app: