Medicare Will Pay $50k for Your Heart Attack. It Should Pay $100 to Prevent It
Summary
The article documents post-training a Qwen3-4B model as a pilot to fly a Tesla Roadster in a wind simulator, showing that RL-trained small models can outperform larger ones under real-time constraints. It details a two-stage warm-start plus RL training, comparisons across wind conditions, and the open-source SkyPilot workflow that orchestrates training, rollouts, and evaluation. The piece highlights cost efficiency (~$13) and emphasizes the value of imitation plus RL in achieving robust, real-time control.