Show HN: I RL-trained an agent that trains models with RL (for –$1.3k)
Summary
A GitHub project showcases an RL-based meta-training loop where an agent writes and runs RL training jobs for other models, using Runpod GPUs and Thinking Machines' Tinker. It includes reward design, task families, results, and cost accounting; the codebase is open source with pretrained weights and a getting-started guide. It highlights automated AutoML style workflows and cost transparency.