Tech, AI, money, and occasional programmer pain.
🚀$0 🟨⬜️⬜️⬜️⬜️ $1k MRR
- Tweets 2,412
- Following 110
- Followers 118
- Likes 5,648
Pretty cool! But how to avoid this?
If you can write CSS, you can write GSS { ● }
I've been dreaming of this for years: a language to build 3D scenes like you write CSS. Selectors, cascade, :hover, keyframes… compiled to one GPU shader.
Meet GSS: GPU Style Sheets.
gss-lang.dev
#shader #glsl #gss
1
30
Apparently it’s cheaper using DeepSeek v4.1 flash via @OpenRouter during the peak hours of @deepseek_ai 🤯
Yes, even with the 5.5% up charge.
22
Made a dashboard for daily quick market briefing. Open to suggestions 🤗
swiftnews.xyz
2
8
100
AI be like: sorry I saw your key without your permission, please rotate the key
Why would you peek without asking me?
2
32
Soft relaunching my daily market briefing website. I’m my first user. Free for all.
Still need some work, here’s the first glance.
28
I lowered my internet bill by $15 a month plus a one time $25 credit. Thanks @ChatGPT
1
45
This game was initially written by Codex. I asked my local qwen3.8 27B model to add the new themes at the beginning of the video. Took it 10 minutes 😂
I see a lot of potentials in it.
1
1
55
I learned something new. Based on this, the memory bandwidth of DGX Spark is even lower. Looks like RTX 3090/4090/5090 or other graphics card with higher memory bandwidth is the way to go for higher tok/s.
My Mac is still useable for running overnight task for my local jobs. I’m not in a hurry to invest in a new machine.
11 tok/s on 8-bit is already near the M2 Max ceiling. 400GB/s over ~27GB per token is ~14 theoretical, so you are at about 75% efficiency, which is normal for decode.
4-bit halves the bytes per pass: ceiling ~28, low 20s measured is the realistic jump.
A 27B fits that machine fine, no Spark needed.
1
2
230
