The Ax3l Project
The Ax3l Project uses Qwen 3.5 4B, small, locally hosted AI to ask the question, “Can an AI improve another AI’s performance by adjusting it’s model and operating environment settings?”
This project uses the Snake Lab Server which houses a small and simple AI. The Snake Lab AI is trainable. The Snake Lab Server accepts a configuration to set the AI’s hyper-parameters and configure the training environment. The Snake Lab server also runs the training simulation. The high score and other simulation results are stored in a database.
The Ax3l system submits simulation configurations and analyzes the result data. It adjusts the configuration and submits a new simulation. This creates a continuous experimental loop in which the Ax3l AI explores the configuration space and attempts to improve the Snake AI’s performance over time.
Live Experiment Data
This is a long-running project expected to operate for well over a month. The current experiment began on September 15, 2026. Real-time project status can be viewed:
- Live Site with event log
- Simulation Scores Distribution Histogram
- Simulation Highscore History
- Gallery of the Snake in Action
- A random sampling of the LLM’s Thinking