Ai

Previously used products, during code development, were largely satisfactory, but Byte’s SOLO introduced significant differences. Initially, participation in beta testing was through invitation codes, and now it involves submitting an email address for review. After approval, the product can be used. I’ve forgotten when I submitted the application, but today I received a notification from Trae indicating that my application had been approved.

ByteSolo’s Advantages

Following a standard project development process, it incorporates UI design, requirements analysis, functional design, technical solution implementation, and finally code development. In other words, the overall interaction logic is more friendly for beginners. For complex projects, the traditional approach remains more effective – AI as an auxiliary tool to modify and optimize existing code.

Using AI too much, we start thinking of AI for everything – often, learning new developments is more reliable with search engines plus official project documentation.

Hong Kong stocks entered on the dip, encountered a pullback, and then just randomly traded, resulting in basic losses.

Strategy Trading

Not necessarily to actually practice and make money, but rather to improve my own abilities through learning. I don’t believe in these trading indicators; I mostly trust the national fortune (luck) and investing in broad market indices via a buy-and-hold strategy.

Currently, no large model stands out as particularly superior; each company has its own strengths and preferred use cases.

Technical Documentation

For feeding code or asking IT technical questions: ChatGPT and Gemini

Write Code

Gather requirements and request code modifications: Claude

AI has been integrated into daily development workflows, and recent investments have shifted focus from external funds to internal stocks and ETFs.

Open Source Projects

Project Log

Last week, I was bored and tried to obtain GitHub badges, starting to use the Issue module. Previously, when writing code, I always wanted to find a place to record the content of each AI modification, but creating a separate document to record it felt disorganized. Now that we have the Issue module, tagging them to distinguish between bugs, features, enhancements, etc., has made the records clear and efficient. Even if I might not need them in the future, recording them is still a form of accumulation. View Issue List

Cultural Propagation: Its ideological influence, subtle and pervasive. AI Programming: Not performing software design, resulting in a lot of rework.

Cultural Propagation

Initially, the project only supported English, Japanese, and Korean. Thinking it was just AI translation, we wondered if supporting more languages wouldn’t be a good idea. So, French, Russian, and Hindi were added. At this point, no problems were detected; when the program executed translations, formatting issues arose due to historical code problems, requiring re-translation of archived articles.

The initial design of the blog translation project was overly complex – first parsing Markdown format, then using placeholders to protect the content, and finally sending it to a large model for translation. This was entirely unnecessary; large models inherently possess the ability to recognize Markdown syntax and can directly process the original content while maintaining formatting during translation.

Our work shifted from debugging code to debugging the prompting of the model. Model: google/gemma-3-4b Hardware: Nvidia 3060 12GB Indeed, we chose a non-thinking model – thinking models were inefficient when executing translation tasks. We compared the performance of 4b and 12b parameters, and for translation purposes, gemma3’s 4b parameter was sufficient; there was no significant advantage in terms of 12b parameters. 12b parameter speed: 11.32 tok/sec , 4b parameter speed: 75.21 tok/sec.

For many years, I’ve focused on backend development, and recently started to explore AI programming while dipping my toes into some frontend-related content. However, during this period of tinkering, I gradually realized I was falling back into an old habit – being dazzled by shiny new things. I constantly try to use AI to create a frontend interface, but in reality, these attempts haven’t provided much practical benefit for my current work and are actually wasting my energy.

This site is developed using Hugo, but I’ve always used Chinese titles, which results in less friendly generated article links. In simpler terms, when shared, they don’t look as good because the Chinese characters are escaped into formats like %E4%BD%A0%E5%A5%BD within the links. While you can solve this by setting a slug, it’s tedious to do manually every time.

Therefore, I decided to try using Claude4 to develop a translation assistant that automatically converts Chinese titles to English slugs and adds hyperlinks within the articles. This would eliminate the need for manual setup.