I felt I needed to get better at expressing myself, so I decided to pick up the habit of writing weekly notes again...
Also, a lot of the time I read an article once and then it just slips away — nothing settles. Maybe writing it down can help the thinking actually land.
Last weekend I went to AI Maker Summit. It started because I saw Bojie Li post on fedi that he was going to give a talk. I opened the summit site, saw the ticket was over 300 yuan, hesitated, then bit the bullet and signed up. When I got there, there was no talk by Bojie. The next day I saw on fedi that he seemed to have gone to the US. On Friday he published a blog post about his AI observations in Silicon Valley; I found it really interesting and read it two or three times.
https://01.me/2025/12/silicon-valley-ai-insights-2025/
The part I found most useful personally was the best practices for AI coding: learn to split tasks, keep AI-written code within about 500 lines per task. You can also use different models to review the code.
Back to AI Maker Summit: after sitting through the whole thing, I found Daojie's talk, a talk on post-training, and an investor's talk the most interesting. My favorite line from the investor was, "We've looked at so, so many projects..." — it gave a sense of being grounded.
During Daojie's talk he mentioned an agent memory system that would be presented next door. He said if large models fully solved memory, projects like that wouldn't exist at all.
That was a pretty big takeaway, and together with Bojie's blog: indie developers or small companies need to know their niche, make sure the product they ship won't be swallowed as models keep getting stronger, and also make sure it won't just get built by a big company. So I think even engineers and programmers need to read AI papers and companies' tech reports to keep up with where large models are.
A day or two after it ended, I got pulled into the AI Maker Summit group chat. The next day a founder posted a Product Hunt launch.
Because I also believe reading itself can't be replaced by AI, I went to take a look on Product Hunt. It was about reading a book with famous people. I clicked into the app and tried it — it was actually really fun!
Seeing those lines from Jobs and Munger next to it, I thought of Poor Charlie's Almanack and the Steve Jobs biography I'd read before. I realized this was also a review of books I'd already read, and it made it easier to connect the dots. So after grabbing a 50% discount on Discord, I subscribed to the Annual Plan without hesitation — buy more, save more, right.
Then I kept reading Tony Dinh's My Indie Book on Readever. Toward the end I was getting a bit tired, impatient to finish. I might finish it tonight or tomorrow.
Just now I saw another interesting example.
Jason Young spent a huge amount of effort and money converting OpenRouter's chat format into a Claude Code-compatible mode, and then OpenRouter rolled out a compatible API.
Tony Dinh made a similar choice in the book.
The first three days of this week I was preparing for the thesis proposal defense. I thought I was well prepared, but the reviewers still said it felt like a product manager pitching a product. A lab-mate who defended after me said one of the professors might have given me a score in the seventies. I don't know if I'll still need to defend at the school level — results come out next week...
I've decided to wait until the 31st to write this year's year-end review, because every day now might still bring a surprise. You have to wait until the last day for the year to actually be over.