Real Users, Real Excitement! 👾✨
While others talked about future adoption, Gamegear stole the show in Denver. Watch users light up playing games and testing agents. This isn't a whitepaper - it's live tech making crypto fun for real people right now.
Results over roadmaps. 🎮
This is the first thing I've ever made that's too fucking rad and cool to tweet about openly.
UniRig was shit so I built a better one.
Runs with one gemini-3.1-flash call for $0.0025.
Rigging costs 5 credits through Meshy which is ~$.10 on their basic $20 plan.😘
I hacked the engine out of comfyUI's image-to-3d to utilize llama.cpp and now it can run quantized models and this workflow is down to 4g of VRAM.
You're paying $400 a month in meshy credits for something you can run on a GPU from 2018.
I'm shipping a fully integrated emulator for abandonware that plays on a 2D screen inside these 3D prompted digital worlds, the entire project after this merge will be ~250,000 lines of javascript. A game within a game.
Long Live Gamegear.🎮⚙️
we have hung out and got a full demo of quddit. actually mad insane a single person built all of this.
prompt to game, multiplayer, custom editor, water, l system trees, impostors...
games that work. very cool shit.
part of the fun of building this game engine is just randomly stumbling into visual vibes that trip me back to gaming on my n64 in 2001. LLMs are capable of such amazing art if you teach them properly.
"White Hillside Buildings Overlooking The Sea".
Cost: $0.34, Time: 2.5m
I've been teaching LLMs how to write threeJS + create their own texture assets for almost a year.
2 minutes 40 seconds. Enjoy. ❤️
zeroshot: "𝘱𝘭𝘢𝘯𝘦𝘵 𝘟, 𝘮𝘺𝘴𝘵𝘦𝘳𝘪𝘰𝘶𝘴 𝘢𝘳𝘤𝘵𝘪𝘤 𝘱𝘭𝘢𝘯𝘦𝘵 - 𝘴𝘵𝘦𝘢𝘮 𝘷𝘦𝘯𝘵𝘴 𝘢𝘯𝘥 𝘸𝘦𝘪𝘳𝘥 𝘪𝘤𝘦 𝘴𝘵𝘳𝘶𝘤𝘵𝘶𝘳𝘦𝘴"
Guys I got the threeJS outputs of my model to be fucking unreal. This prompt is 4 words.
"3D voxel, snow cabin."
Behind the scenes, it takes the user intent and generates an text-to-image prompt on the fly for a 2D reference image that Gemini then uses to code the 3D art.
I'm fucking cracked?
Kimi K2 given principal component decompositions (PCA) of a basket of 2000 anonymized stock tickers, tasked with managing a portfolio for risk adjusted returns.
Just traditional data science, a little prompt magic - and 4$ of API credits.
Guys I'm freaking out. I've reran the numbers 3x times. I've checked for data leaks. I was so paranoid I anonymized the stock tickers in my dataset.
After a decade of programming and trading - I wrote my first HIGHLY profitable quant algorithm.🙏❤️
Details this weekend.⬇️