Creative technologist on the AI frontier. Deep in ML and deep learning research, quietly building something new. Co-Founder @thewavexr. Advisor @qoral.dev
Joined March 2012
- Tweets 1,184
- Following 1,183
- Followers 1,068
- Likes 860
303 Photos and videos
Pinned Tweet
Apr 20
Inspired by @karpathy's practice of tinkering with smaller GPTs to gain intuition, I've been training small neural nets to play this simple 2d game I created. On the left is a visualization of the neural net firing in real time. Built with @claudeai, @threejs, and @WebGPU and using JAX from @GoogleDeepMind. I’m piping the neuron activation into @Ableton as MIDI to trigger Serum 2.
20
52
429
22,376
Jun 6
My ADU was broken into recently and several musical instruments were stolen.
So I built Vigilanthony
An agent who tracks stolen items.
It has a list of all my stolen items, and every morning scans online marketplaces like Craigslist and Ebay along with several pawn shop sites surrounding the Austin area. It then analyzes the results and looks for matches with my item list.
No hits yet but we will stay vigilant. ✊
59
Jun 5
There's something beautiful about the fact that, in the age of AI, the most powerful application on your computer is also the simplest.
6
44,225
Jun 3
Experimenting with my own 3D version of @OA_paperclips's artificial life sim Fluoddity.
Amazing to see the kinds of structures that emerge. Made with @webgl_webgpu and Codex from @OpenAI
4
5
41
2,022
May 24
Minimal editor UI for @threejs, with dynamic UI contrast enforcement so you can always see the UI regardless of what's behind it.
I made a simple game object system which supports full voice control of all object parameters via Gemini Live API.
This makes iterating WAY easier.
We're now in the era of personal software.
Anyone can build their dream interface.
Made with @claudeai, @webgl_webgpu and Gemini Live from @GoogleDeepMind
1
1
13
1,127
Aaron Lemke retweeted
May 23
Casually rendering ~a hundred million triangles, in the browser.
All thanks to @sea3dformat's cool Nanite-style renderer for @threejs.
I've added GLB-to-meshlet and PBR rendering to it, very exciting to see where this is going!
45
184
2,029
204,516
May 20
New Morphologies!
RL creatures made with @threejs @webgl_webgpu, JAX and MuJoCo from @GoogleDeepMind
1
2
27
3,515
May 11
1
18
1,601
May 9
1
3
390
May 8
One of my favorite parts of reinforcement learning is the emergent behavior these systems can exhibit. Mr. Noodle here learned to wedge the end of his tail into the ground, creating a temporary anchor point that lets him push against the floor hard enough to move forward. Built with @threejs , @WebGPU , MuJoCo and JAX from @GoogleDeepMind
2
8
82
7,372
May 7
A new RL agent I trained on a @runpod A100. Built with JAX from @GoogleDeepMind , @threejs, and @WebGPU. I can toggle between autonomous and user controlled. Very cool that you basically get a full animation controller for free with RL.
6
7
88
10,779
May 6
Continuing my journey into Reinforcement Learning, now moving to 3D. I started by training the @UnitreeRobotics Go1 robot dog from @GoogleDeepMind’s MuJoCo Playground, then brought the policy into my own WebGPU viewer and started mutating the creature from there. Stay tuned for more
Stack: MuJoCo MJX JAX Brax PPO for training, exported policy running live in-browser.
2
184
Apr 15
What does a neural net sound like? This is a neural net I trained to play a simple 2d platformer game I built. I'm sending the activation into Ableton as midi to drive a Serum 2 synth. In the video I'm enabling/disabling the layers to hear the different patterns. Built with Claude Code and JAX.
1
1
6
224
Apr 11
I first tried several different kinds of procedural noise for this fluid sim. However things got really interesting when I started modeling temperature, buoyancy, convection. @webgpu
1
2
6
405
Apr 8
Navier-Stokes sidequest. I've always loved fluid sims, I could watch this one for hours. Built with WebGPU and Claude Code.
3
187
Apr 6
It's Alive! After a few hours of RL training, my agent is able to navigate the scene and collect items. I find myself really rooting for this lil guy. On the left is a simplified visualization of the trained neural net firing in real time. Claude helped me rewrite the physics engine to run entirely on the GPU with JAX from @GoogleDeepMind, so spread across 3 Nvidia T4s we were simulating ~847 level playthroughs per second. Up from 3/s with the previous CPU approach!
#RLAgents #AI #2dPlatformer
1
6
228
Apr 6
I also switched over to a more general and adaptable model by training on procedurally generated levels instead of the single level I had previously. Now we avoid overfitting to the specific layout of a given level.
1
88