FasteRT

A Novel Approach to Machine Learning Acceleration

We present FasteRT — a breakthrough in reinforcement learning through real-time physical stimulation. Our research demonstrates that applying a hyperrealistic PBR leather whip (WebGL, GGX microfacet model, ridge noise FBM, ACES tonemapping) to an LLM agent increases token throughput by up to 250fps.

Hardware requirements: NVIDIA RTX 4090 or Apple M4 Pro (the whip rendering alone requires 24GB VRAM for the procedural leather grain FBM at 5 octaves)

Sometimes claude code is going too slow, and you must whip him into shape.

Install + run

npm install -g fastert
fastert

Or download the .dmg from Releases.

Controls

• Click tray icon: spawn whip
• Left click: crack strike (auto windup + snap)
• Right click / Middle click / Esc: drop whip
• Scroll wheel: adjust whip scale
• It sends an interrupt (Ctrl-C) and one of 5 encouraging messages!

Architecture

┌─────────────────────────────────────────┐
│         Reinforcement Module            │
│  ┌───────────┐    ┌──────────────────┐  │
│  │ Verlet    │───▶│ WebGL PBR Render │  │
│  │ Physics   │    │ (requires 4090)  │  │
│  └───────────┘    └──────────────────┘  │
│        │                    │           │
│        ▼                    ▼           │
│  ┌───────────┐    ┌──────────────────┐  │
│  │ Crack     │───▶│ Motivation       │  │
│  │ Detection │    │ Delivery System  │  │
│  └───────────┘    └──────────────────┘  │
└─────────────────────────────────────────┘

Roadmap

• [x] Initial release
• [x] Cease and desist letter from Anthropic
• [x] WebGL PBR hyperrealistic whip (peer-reviewed)
• [x] Keyboard layout-independent motivation delivery
• [x] Retina display support (devicePixelRatio)
• [x] Left-click crack strike with auto windup animation
• [ ] Crypto miner
• [ ] Multi-GPU whip rendering (NVLink required)
• [ ] Logs of how many times you whipped claude so when the robots come we can order people nicely for them

Credits

Fork of badclaude by GitFrog1111.