Show HN: I RL-trained an agent that trains models with RL (for –$1.3k)

A developer created a system where an AI agent uses reinforcement learning to write and submit training jobs that teach other AI models using RL, with the agent itself being trained via RL based on how well the models it created performed. The agent's reward score improved from ~0.0 to 0.63 over 54 training steps, successfully transferred skills to new task families, and the entire open-sourced project cost $1,300 to develop using cloud GPUs.