RL-trained agent trains other models with RL, loses $1.3k

A developer trained an RL agent that uses reinforcement learning to train other models, costing $1,300 in compute. The agent explores hyperparameters and training strategies autonomously.