paddy (@0xPD33) header
paddy (@0xPD33) profile picture

schizo tech wizard. 🧙‍♂️ Christ is King ✝️ NixOS ❄️ | AI/ML | Game Dev | DeFi

heaven on earth
Joined May 2023
Photo by paddy (@0xPD33)Photo by paddy (@0xPD33)Photo by paddy (@0xPD33)Photo by paddy (@0xPD33)Photo by paddy (@0xPD33)Photo by paddy (@0xPD33)Photo by paddy (@0xPD33)Photo by paddy (@0xPD33)Photo by paddy (@0xPD33)Photo by paddy (@0xPD33)
Pinned Tweet
I haven't written a single line of code in 1 year. 2 of the projects I was working on are still alive in a way. I decided to continue building out the FPS game and rewrite the Godot Open World in a custom C + Vulkan engine. There are no limits to what you can do in game development at this point, the latest closed source and open source models are good enough for everything you can imagine.
I haven't written a single line of code in months. And yet, I was able to start building 4 large-scale games. 7 years in game dev. Here's what's actually different now. The projects, all side projects, all built with AI over the past few months: → Open world RPG prototype (GDScript/Godot) → Space sim in Godot-Rust (Rimworld × Starsector) → FPS in C + SDL3 → Life-sim engine in C + SDL3, think Sims 2/3 but ultra-scalable with planned LLM integration Here is what the leading models handled without issues: - Custom render pipelines on top of SDL - Shaders for everything I could describe or give references for - Procedural world and model generation (everything you see in the screenshots is procedural, no external assets) - Complex, interlocked gameplay systems Years of programming work condensed into weeks. The engine doesn't matter anymore. Godot and GDScript, Rust and Bevy, C and SDL/Raylib, Javascript and ThreeJS. Pick whatever fits. The models handle it. What actually matters is domain knowledge. 7 YOE meant I knew what to ask for, how to structure things, which patterns to reach for. That's the real multiplier. There are things these models still can't do well. I'll break down each project with the specific challenges and what I'm planning in future posts. But one thing is clear: the floor has shifted. If you can clearly describe your dream game, you can almost certainly start building it today. Fire up Claude or Codex and give it a shot. Any questions? Feel free to ask.
7
2,091
paddy retweeted
Asked one of our developers how he uses his AI agents and he just sent me this:
177
1,324
260
16,166
1,736,814
5
18
584
7,873
you just need html to build a website btw motherfuckingwebsite.com
62
119
11
2,246
313,266
Damn, another one
[ ZOOMER ] GITBOOK, WHICH IS HEAVILY USED FOR CRYPTO DOCUMENTATION, DISCLOSES AI AGENT SECURITY BREACH - URGES USERS TO ROTATE SECRETS: BLOG
2
206
135
3,754
125
50,867
985,801
Ayo this seems pretty good
The newest #MaixStation Dock board design is here! Outperforming #DGX Spark's AI power in just 10x10cm, Check out the new option for local AI deployment: sipeed.com/maixstation/
2
204
Even faster answers are coming at no accuracy loss. I've resorted to patching llama.cpp with specific improvements, once they are proven solid I'll open PR's upstream as well.
The next release, Commonplace 0.4.0, gets Gemma's first word 31% sooner on a Pixel 10, fully offline: 1.86 s → 1.28 s (median, 34 questions). The trick: the CPU and the Tensor NPU no longer take turns. 🧵
1
1
11
373
🚨 VITALIK: “We really need to treat data as toxic waste by default.” At Devcon Mumbai, he calls for open-source tech and zero-knowledge proofs to protect privacy and reduce reliance on trust.
154
145
66
1,082
164,027
paddy retweeted
83
636
22
3,666
125,719
oh what a mess this is no way I'm really hard stuck on Gemma
1
3
255
@googlegemma would be nice if I could get some insight on how to properly run more LLMs on the TPU on newer Pixels. So far Gemma E2B apparently is the best thing I can run for my offline research app @askcommonplace Even an upgrade to E4B would go a long way.
72
nice I love herdr, started using it recently. I was one of those psychos who just rawdogged tmux with multiple panes and different shells on all of my niri workspaces. Now I'm a bit more organized without running any of the labs official desktop apps.
*ehm ehm* BIG ANNOUNCEMENTS DAY 📣 two new things cooking in herdr labs 🐏🧪 a brand new herdr-core, and a desktop app built on it. open source. first, the app: software that fits you. every part of it is a file your agent can change. you just ask.
3
221
reminder while Opus 5.5 is still the best model out there, most tasks really don't need it get some compute if you can, open source will only be up from here
you probably have no idea but we are beyond the point of needing cloud AI after recent Qwen, Deepseek and GLM releases. 95% of tasks you are doing, work or personal, can be done by open source models. get compute and be free now. i have 2 DGX sparks on the way.
3
270
34
657
21
9,261
95,246
wow!
I just built a *functioning* React app by describing what I wanted to GPT-3. I'm still in awe.
3
415
your years of practice don't disappear with AI. they become the filter that makes everything it produces actually good
2
4
32
676
Open models on your own hardware
Claude gonna start snitching on Tuesday. Anthropic just emailed customers. Starting Nov 3, text carries an invisible watermark. You can't see it. It survives copy and paste, and it can survive some editing. Only regulators, journalists, fact-checkers and researchers can check for it right now. It doesn't say who you are. It says Claude touched the words. Even a quick proofread counts. If your brand voice runs on Claude, write like someone might check. Because now someone can.
5
6
93
3,294
JUST IN: Anthropic bans Claude Code from recreating existing video games amid a viral trend of developers using AI to reverse-engineer popular games.
48
1,260
14
20,772
315,574
Interested in how Commonplace works and what makes it stand out? There is a lot of documentation on the backend on GitHub (just follow the README): github.com/0xPD33/commonplac…
3
1
14
755