Natural language rewards replace hand-coded scoring in RL agent training

Original: Karpathy's prediction about RL is coming true now!

Source: x.com ↗