2024 was a failure year for me.
Still continued pursuing and diving deep into genAI. Had many ideas, made some prototypes but couldn’t launch any products. Pretty terrible.
What I Learned in 2024
GenAI - LLMs
I work with genAI daily through ChatGPT/Claude and cursor lol. Did a few projects at work too, though I found them quite boring.
I learned some new techniques, mainly in LLM optimization. The optimization problem I encountered and thought about the most was how to improve response time with fixed hardware (without solving it by scaling up hardware). Also did a bit with Agents. Generally, I know all the LLM techniques but haven’t deployed them in real-world scale, so it’s whatever.
Still believe in local LLMs, and working hard to create products for local LLMs.
GenAI - Visual things
I learned comfy UI, which is an awesome backend + frontend for image generation.
Did a couple of projects with genAI image early in the year, learned bits and pieces.
New Coding Paradigm
Năm nay thậm chí đã dùng cursor dài hạn, 20$/tháng, mình nghĩ cũng là cái giá tạm tạm hợp lý. Mình tự tin code những thứ mà 2 3 năm trước sẽ rất khó nhằn, ví như có thể code 1 cái browser extension trong 1 ngày. Tuy vậy, vì làm việc nhiều nên mình biết nhiều cái khó khăn bên coding. Có lẽ trong tương lai 5 năm, chưa thể thay thế vị trí programmer được (vì data có vẻ là giới hạn rồi).
This year I even used cursor long-term, $20/month, I think it’s a fairly reasonable price. I’m confident coding stuff that would’ve been super tough 2-3 years ago, like coding a browser extension in a day (without any knowledge of JS and extension). However, because I work a lot, I know limit of AI coding. Perhaps in the next 5 years, it still can’t replace programmers (seems like data is 1 big limitation).
UI things
I learned fasthtml for making backend + frontend for web-based products. It is best in the town for this task.
However, I’m currently focusing more on games/apps. One of the things I’m putting all my effort into is LLM-based games. Initially learned godot (godot 2d is really amazing), but because godot went woke, I’ll use Unity, which is also a better choice (much more support and guide, also I believe that AI know Unity better than godot).
Really want to work with Love2d but now’s not the time.
My current stack/flow
- Claude for chat and research: free claude is enough, I find it smarter than chatgpt.
- Cursor for coding: Cursor is much better than VScode copilot. I tried Cline but it doesn’t work well.
- fasthtml for quick MVP: much better than gradio/fastapi.
- Obsidian for Logs and Notes: still love it.
What I Plan to Learn/Make in 2025
- make thing
- launch thing
Last year I launched nothing. It is a failed year.
Learn to build and launch is my highest priority in 2025, and no second priority. I have a killer idea that I believe it could be both technical possible and marketable, also 99% sure I could make some money from it (1 mil at least, I guess).
So, do it.
